https://lianwu.app/videos/JAM3ToAoGhM https://lianwu.app/media/youtube/JAM3ToAoGhM.jpg One prompt, a team of agents: Gemini 4 Carbon #Shorts One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. - In Google's demo, a swarm built an open-world 3D game: you ride a bike around Google's Mountain View campus as a pelican. Subagents modeled the 3D assets in Blender, built the game in Three.js, and added sound effects and music. My take: Carbon is not the model for every task. It's the one you put in charge, and /swarm gives it a team. Full video, with price, benchmarks and chapters: https://youtu.be/HV3eVHoWzDw Sources Google, Gemini 4 Carbon is now available (Nov 10, 2026): https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/ Google Antigravity, Gemini 4 Carbon in Google Antigravity: https://antigravity.google/blog/gemini-4-carbon-in-google-antigravity/ Demo footage: Google's official videos on YouTube and the Antigravity blog, used muted. https://www.youtube.com/embed/JAM3ToAoGhM 57 2026-11-11T16:08:21.000Z Marvin Builds https://lianwu.app/zh-TW/videos/JAM3ToAoGhM https://lianwu.app/media/youtube/JAM3ToAoGhM.jpg One prompt, a team of agents: Gemini 4 Carbon #Shorts One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. - In Google's demo, a swarm built an open-world 3D game: you ride a bike around Google's Mountain View campus as a pelican. Subagents modeled the 3D assets in Blender, built the game in Three.js, and added sound effects and music. My take: Carbon is not the model for every task. It's the one you put in charge, and /swarm gives it a team. Full video, with price, benchmarks and chapters: https://youtu.be/HV3eVHoWzDw Sources Google, Gemini 4 Carbon is now available (Nov 10, 2026): https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/ Google Antigravity, Gemini 4 Carbon in Google Antigravity: https://antigravity.google/blog/gemini-4-carbon-in-google-antigravity/ Demo footage: Google's official videos on YouTube and the Antigravity blog, used muted. https://www.youtube.com/embed/JAM3ToAoGhM 57 2026-11-11T16:08:21.000Z Marvin Builds https://lianwu.app/ja/videos/JAM3ToAoGhM https://lianwu.app/media/youtube/JAM3ToAoGhM.jpg One prompt, a team of agents: Gemini 4 Carbon #Shorts One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. - In Google's demo, a swarm built an open-world 3D game: you ride a bike around Google's Mountain View campus as a pelican. Subagents modeled the 3D assets in Blender, built the game in Three.js, and added sound effects and music. My take: Carbon is not the model for every task. It's the one you put in charge, and /swarm gives it a team. Full video, with price, benchmarks and chapters: https://youtu.be/HV3eVHoWzDw Sources Google, Gemini 4 Carbon is now available (Nov 10, 2026): https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/ Google Antigravity, Gemini 4 Carbon in Google Antigravity: https://antigravity.google/blog/gemini-4-carbon-in-google-antigravity/ Demo footage: Google's official videos on YouTube and the Antigravity blog, used muted. https://www.youtube.com/embed/JAM3ToAoGhM 57 2026-11-11T16:08:21.000Z Marvin Builds https://lianwu.app/ko/videos/JAM3ToAoGhM https://lianwu.app/media/youtube/JAM3ToAoGhM.jpg One prompt, a team of agents: Gemini 4 Carbon #Shorts One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. - In Google's demo, a swarm built an open-world 3D game: you ride a bike around Google's Mountain View campus as a pelican. Subagents modeled the 3D assets in Blender, built the game in Three.js, and added sound effects and music. My take: Carbon is not the model for every task. It's the one you put in charge, and /swarm gives it a team. Full video, with price, benchmarks and chapters: https://youtu.be/HV3eVHoWzDw Sources Google, Gemini 4 Carbon is now available (Nov 10, 2026): https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/ Google Antigravity, Gemini 4 Carbon in Google Antigravity: https://antigravity.google/blog/gemini-4-carbon-in-google-antigravity/ Demo footage: Google's official videos on YouTube and the Antigravity blog, used muted. https://www.youtube.com/embed/JAM3ToAoGhM 57 2026-11-11T16:08:21.000Z Marvin Builds https://lianwu.app/videos/HV3eVHoWzDw https://lianwu.app/media/youtube/HV3eVHoWzDw.jpg Gemini 4 Carbon, explained: one prompt, a team of agents One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. Here's what Google made available on November 10, 2026. - Gemini 4 Carbon: Google calls it their strongest model. Announced on October 28 and given to cybersecurity defenders first. Google still recommends Gemini 4 Argon for most coding and knowledge work; Carbon is for the hardest tasks and uses more tokens to reason deeply and check its own work. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. Agents talk on a shared message board; a waiting agent costs zero tokens. You can pin each agent to a model: Carbon to coordinate and verify, lighter models for routine edits. Google's own warning: a swarm uses up your quota much faster than a normal chat. - Google's demos: an open-world 3D game on Google's Mountain View campus (you ride a bike as a pelican), Google's Platform 37 office rebuilt in 3D from seven city council planning PDFs over ten overnight loops, and a playable fortepiano from around 1780 rebuilt from CAD drawings and X-rays. - Benchmarks and safety: Google says Carbon ranks first on CursorBench 4.0 and two other outside benchmarks, at under half the cost per task of the next best outside model. Its best security results come from Carbon for cyber, a version only approved defenders can use. On Gray Swan's prompt injection test, zero attacks got through; Google still says prompt injection needs more work. - Price and access: $2 per million input tokens and $10 per million output tokens, the same as Argon. After November 30, 2026, that doubles to $4 and $20. Developers use it on the Gemini API's paid tier; in the Gemini app, Search and Antigravity you need Google AI Ultra. My take: Carbon is not the model for every task. It's the one you put in charge, and /sw https://www.youtube.com/embed/HV3eVHoWzDw 164 2026-11-11T15:40:05.000Z Marvin Builds https://lianwu.app/zh-TW/videos/HV3eVHoWzDw https://lianwu.app/media/youtube/HV3eVHoWzDw.jpg Gemini 4 Carbon, explained: one prompt, a team of agents One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. Here's what Google made available on November 10, 2026. - Gemini 4 Carbon: Google calls it their strongest model. Announced on October 28 and given to cybersecurity defenders first. Google still recommends Gemini 4 Argon for most coding and knowledge work; Carbon is for the hardest tasks and uses more tokens to reason deeply and check its own work. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. Agents talk on a shared message board; a waiting agent costs zero tokens. You can pin each agent to a model: Carbon to coordinate and verify, lighter models for routine edits. Google's own warning: a swarm uses up your quota much faster than a normal chat. - Google's demos: an open-world 3D game on Google's Mountain View campus (you ride a bike as a pelican), Google's Platform 37 office rebuilt in 3D from seven city council planning PDFs over ten overnight loops, and a playable fortepiano from around 1780 rebuilt from CAD drawings and X-rays. - Benchmarks and safety: Google says Carbon ranks first on CursorBench 4.0 and two other outside benchmarks, at under half the cost per task of the next best outside model. Its best security results come from Carbon for cyber, a version only approved defenders can use. On Gray Swan's prompt injection test, zero attacks got through; Google still says prompt injection needs more work. - Price and access: $2 per million input tokens and $10 per million output tokens, the same as Argon. After November 30, 2026, that doubles to $4 and $20. Developers use it on the Gemini API's paid tier; in the Gemini app, Search and Antigravity you need Google AI Ultra. My take: Carbon is not the model for every task. It's the one you put in charge, and /sw https://www.youtube.com/embed/HV3eVHoWzDw 164 2026-11-11T15:40:05.000Z Marvin Builds https://lianwu.app/ja/videos/HV3eVHoWzDw https://lianwu.app/media/youtube/HV3eVHoWzDw.jpg Gemini 4 Carbon, explained: one prompt, a team of agents One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. Here's what Google made available on November 10, 2026. - Gemini 4 Carbon: Google calls it their strongest model. Announced on October 28 and given to cybersecurity defenders first. Google still recommends Gemini 4 Argon for most coding and knowledge work; Carbon is for the hardest tasks and uses more tokens to reason deeply and check its own work. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. Agents talk on a shared message board; a waiting agent costs zero tokens. You can pin each agent to a model: Carbon to coordinate and verify, lighter models for routine edits. Google's own warning: a swarm uses up your quota much faster than a normal chat. - Google's demos: an open-world 3D game on Google's Mountain View campus (you ride a bike as a pelican), Google's Platform 37 office rebuilt in 3D from seven city council planning PDFs over ten overnight loops, and a playable fortepiano from around 1780 rebuilt from CAD drawings and X-rays. - Benchmarks and safety: Google says Carbon ranks first on CursorBench 4.0 and two other outside benchmarks, at under half the cost per task of the next best outside model. Its best security results come from Carbon for cyber, a version only approved defenders can use. On Gray Swan's prompt injection test, zero attacks got through; Google still says prompt injection needs more work. - Price and access: $2 per million input tokens and $10 per million output tokens, the same as Argon. After November 30, 2026, that doubles to $4 and $20. Developers use it on the Gemini API's paid tier; in the Gemini app, Search and Antigravity you need Google AI Ultra. My take: Carbon is not the model for every task. It's the one you put in charge, and /sw https://www.youtube.com/embed/HV3eVHoWzDw 164 2026-11-11T15:40:05.000Z Marvin Builds https://lianwu.app/ko/videos/HV3eVHoWzDw https://lianwu.app/media/youtube/HV3eVHoWzDw.jpg Gemini 4 Carbon, explained: one prompt, a team of agents One prompt, and a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive 3D exhibition of how the Eiffel Tower was built. Here's what Google made available on November 10, 2026. - Gemini 4 Carbon: Google calls it their strongest model. Announced on October 28 and given to cybersecurity defenders first. Google still recommends Gemini 4 Argon for most coding and knowledge work; Carbon is for the hardest tasks and uses more tokens to reason deeply and check its own work. - /swarm in Google Antigravity (preview): one prompt becomes a team of agents you can watch. A coordinator spins up workers that build in parallel, each in its own git worktree, and verifiers run the tests and challenge every change. Agents talk on a shared message board; a waiting agent costs zero tokens. You can pin each agent to a model: Carbon to coordinate and verify, lighter models for routine edits. Google's own warning: a swarm uses up your quota much faster than a normal chat. - Google's demos: an open-world 3D game on Google's Mountain View campus (you ride a bike as a pelican), Google's Platform 37 office rebuilt in 3D from seven city council planning PDFs over ten overnight loops, and a playable fortepiano from around 1780 rebuilt from CAD drawings and X-rays. - Benchmarks and safety: Google says Carbon ranks first on CursorBench 4.0 and two other outside benchmarks, at under half the cost per task of the next best outside model. Its best security results come from Carbon for cyber, a version only approved defenders can use. On Gray Swan's prompt injection test, zero attacks got through; Google still says prompt injection needs more work. - Price and access: $2 per million input tokens and $10 per million output tokens, the same as Argon. After November 30, 2026, that doubles to $4 and $20. Developers use it on the Gemini API's paid tier; in the Gemini app, Search and Antigravity you need Google AI Ultra. My take: Carbon is not the model for every task. It's the one you put in charge, and /sw https://www.youtube.com/embed/HV3eVHoWzDw 164 2026-11-11T15:40:05.000Z Marvin Builds https://lianwu.app/videos/Fq5_8aXsh0U https://lianwu.app/media/youtube/Fq5_8aXsh0U.jpg Talk it out, Claude draws the whiteboard #Shorts Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. - Claude Whiteboards (beta): dictate a brain dump; Claude groups the ideas, puts the steps in order and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting and drop-downs. Each edit is outlined as it lands. - Whiteboards and Sheets are on Pro, Max, Ultra, Team and Enterprise plans. Full video, with Claude Data and chapters: https://youtu.be/HzpYayj-8JY Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifacts) Demo footage: Claude's official video, used muted: https://www.youtube.com/watch?v=YewAJPawE7I Music: "Large Smile Mood" by Nico Staf, from the YouTube Audio Library. https://www.youtube.com/embed/Fq5_8aXsh0U 52 2026-11-11T12:03:54.000Z Marvin Builds https://lianwu.app/zh-TW/videos/Fq5_8aXsh0U https://lianwu.app/media/youtube/Fq5_8aXsh0U.jpg Talk it out, Claude draws the whiteboard #Shorts Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. - Claude Whiteboards (beta): dictate a brain dump; Claude groups the ideas, puts the steps in order and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting and drop-downs. Each edit is outlined as it lands. - Whiteboards and Sheets are on Pro, Max, Ultra, Team and Enterprise plans. Full video, with Claude Data and chapters: https://youtu.be/HzpYayj-8JY Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifacts) Demo footage: Claude's official video, used muted: https://www.youtube.com/watch?v=YewAJPawE7I Music: "Large Smile Mood" by Nico Staf, from the YouTube Audio Library. https://www.youtube.com/embed/Fq5_8aXsh0U 52 2026-11-11T12:03:54.000Z Marvin Builds https://lianwu.app/ja/videos/Fq5_8aXsh0U https://lianwu.app/media/youtube/Fq5_8aXsh0U.jpg Talk it out, Claude draws the whiteboard #Shorts Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. - Claude Whiteboards (beta): dictate a brain dump; Claude groups the ideas, puts the steps in order and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting and drop-downs. Each edit is outlined as it lands. - Whiteboards and Sheets are on Pro, Max, Ultra, Team and Enterprise plans. Full video, with Claude Data and chapters: https://youtu.be/HzpYayj-8JY Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifacts) Demo footage: Claude's official video, used muted: https://www.youtube.com/watch?v=YewAJPawE7I Music: "Large Smile Mood" by Nico Staf, from the YouTube Audio Library. https://www.youtube.com/embed/Fq5_8aXsh0U 52 2026-11-11T12:03:54.000Z Marvin Builds https://lianwu.app/ko/videos/Fq5_8aXsh0U https://lianwu.app/media/youtube/Fq5_8aXsh0U.jpg Talk it out, Claude draws the whiteboard #Shorts Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. - Claude Whiteboards (beta): dictate a brain dump; Claude groups the ideas, puts the steps in order and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting and drop-downs. Each edit is outlined as it lands. - Whiteboards and Sheets are on Pro, Max, Ultra, Team and Enterprise plans. Full video, with Claude Data and chapters: https://youtu.be/HzpYayj-8JY Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifacts) Demo footage: Claude's official video, used muted: https://www.youtube.com/watch?v=YewAJPawE7I Music: "Large Smile Mood" by Nico Staf, from the YouTube Audio Library. https://www.youtube.com/embed/Fq5_8aXsh0U 52 2026-11-11T12:03:54.000Z Marvin Builds https://lianwu.app/videos/HzpYayj-8JY https://lianwu.app/media/youtube/HzpYayj-8JY.jpg Claude Whiteboards, Sheets and Data, explained Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. Here's what Anthropic launched on November 10, 2026: Claude Whiteboards, Claude Sheets, and Claude Data. - Claude Whiteboards (beta): start one from any chat, or pick the Whiteboard tile in the Artifacts tab. Dictate a brain dump, and Claude groups related ideas, puts the steps in order, and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. When you're done, Claude turns the board into a doc or a deck, or sends it to Lucid, Miro, or Vani by Zoho. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting, and drop-downs. Each edit is outlined as it lands. For now: up to 100,000 cells, 144 functions, no pivot tables yet; export to Excel or CSV. - Claude Data: Claude Dashboards, renamed and generally available. It needs a data connector. Your data team certifies the queries everyone should use, Claude drafts data stories from the numbers, and live sharing shows current numbers to colleagues without access to the connectors. Certified queries and live sharing are in beta. - Who gets it: Whiteboards and Sheets are on Pro, Max, Ultra, Team, and Enterprise plans, and count toward your usage limits. For now a whiteboard works only from what's on the board, not your connected apps. My take: Claude used to hand you an answer to paste somewhere else. Now the board, the sheet, and the dashboard stay in Claude, where your team keeps working on them. Chapters 0:00 Talk it out, Claude draws it 0:21 Claude Whiteboards 1:08 Claude Sheets 1:31 Claude Data 1:46 Who gets it 1:59 My take Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifact https://www.youtube.com/embed/HzpYayj-8JY 143 2026-11-11T10:38:05.000Z Marvin Builds https://lianwu.app/zh-TW/videos/HzpYayj-8JY https://lianwu.app/media/youtube/HzpYayj-8JY.jpg Claude Whiteboards, Sheets and Data, explained Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. Here's what Anthropic launched on November 10, 2026: Claude Whiteboards, Claude Sheets, and Claude Data. - Claude Whiteboards (beta): start one from any chat, or pick the Whiteboard tile in the Artifacts tab. Dictate a brain dump, and Claude groups related ideas, puts the steps in order, and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. When you're done, Claude turns the board into a doc or a deck, or sends it to Lucid, Miro, or Vani by Zoho. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting, and drop-downs. Each edit is outlined as it lands. For now: up to 100,000 cells, 144 functions, no pivot tables yet; export to Excel or CSV. - Claude Data: Claude Dashboards, renamed and generally available. It needs a data connector. Your data team certifies the queries everyone should use, Claude drafts data stories from the numbers, and live sharing shows current numbers to colleagues without access to the connectors. Certified queries and live sharing are in beta. - Who gets it: Whiteboards and Sheets are on Pro, Max, Ultra, Team, and Enterprise plans, and count toward your usage limits. For now a whiteboard works only from what's on the board, not your connected apps. My take: Claude used to hand you an answer to paste somewhere else. Now the board, the sheet, and the dashboard stay in Claude, where your team keeps working on them. Chapters 0:00 Talk it out, Claude draws it 0:21 Claude Whiteboards 1:08 Claude Sheets 1:31 Claude Data 1:46 Who gets it 1:59 My take Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifact https://www.youtube.com/embed/HzpYayj-8JY 143 2026-11-11T10:38:05.000Z Marvin Builds https://lianwu.app/ja/videos/HzpYayj-8JY https://lianwu.app/media/youtube/HzpYayj-8JY.jpg Claude Whiteboards, Sheets and Data, explained Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. Here's what Anthropic launched on November 10, 2026: Claude Whiteboards, Claude Sheets, and Claude Data. - Claude Whiteboards (beta): start one from any chat, or pick the Whiteboard tile in the Artifacts tab. Dictate a brain dump, and Claude groups related ideas, puts the steps in order, and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. When you're done, Claude turns the board into a doc or a deck, or sends it to Lucid, Miro, or Vani by Zoho. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting, and drop-downs. Each edit is outlined as it lands. For now: up to 100,000 cells, 144 functions, no pivot tables yet; export to Excel or CSV. - Claude Data: Claude Dashboards, renamed and generally available. It needs a data connector. Your data team certifies the queries everyone should use, Claude drafts data stories from the numbers, and live sharing shows current numbers to colleagues without access to the connectors. Certified queries and live sharing are in beta. - Who gets it: Whiteboards and Sheets are on Pro, Max, Ultra, Team, and Enterprise plans, and count toward your usage limits. For now a whiteboard works only from what's on the board, not your connected apps. My take: Claude used to hand you an answer to paste somewhere else. Now the board, the sheet, and the dashboard stay in Claude, where your team keeps working on them. Chapters 0:00 Talk it out, Claude draws it 0:21 Claude Whiteboards 1:08 Claude Sheets 1:31 Claude Data 1:46 Who gets it 1:59 My take Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifact https://www.youtube.com/embed/HzpYayj-8JY 143 2026-11-11T10:38:05.000Z Marvin Builds https://lianwu.app/ko/videos/HzpYayj-8JY https://lianwu.app/media/youtube/HzpYayj-8JY.jpg Claude Whiteboards, Sheets and Data, explained Talk through a messy plan out loud, and Claude turns it into a whiteboard your whole team can edit live. Here's what Anthropic launched on November 10, 2026: Claude Whiteboards, Claude Sheets, and Claude Data. - Claude Whiteboards (beta): start one from any chat, or pick the Whiteboard tile in the Artifacts tab. Dictate a brain dump, and Claude groups related ideas, puts the steps in order, and connects the ones that depend on each other. Drag a box around one part to ask Claude about just that area. When you're done, Claude turns the board into a doc or a deck, or sends it to Lucid, Miro, or Vani by Zoho. - Claude Sheets (beta): a spreadsheet Claude builds and edits with you, with tabs, formulas, formatting, and drop-downs. Each edit is outlined as it lands. For now: up to 100,000 cells, 144 functions, no pivot tables yet; export to Excel or CSV. - Claude Data: Claude Dashboards, renamed and generally available. It needs a data connector. Your data team certifies the queries everyone should use, Claude drafts data stories from the numbers, and live sharing shows current numbers to colleagues without access to the connectors. Certified queries and live sharing are in beta. - Who gets it: Whiteboards and Sheets are on Pro, Max, Ultra, Team, and Enterprise plans, and count toward your usage limits. For now a whiteboard works only from what's on the board, not your connected apps. My take: Claude used to hand you an answer to paste somewhere else. Now the board, the sheet, and the dashboard stay in Claude, where your team keeps working on them. Chapters 0:00 Talk it out, Claude draws it 0:21 Claude Whiteboards 1:08 Claude Sheets 1:31 Claude Data 1:46 Who gets it 1:59 My take Sources Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data: https://claude.com/resources/articles/sheets-whiteboards-data Claude Help Center: https://support.claude.com/en/articles/18324319 (Whiteboards), https://support.claude.com/en/articles/18324300 (Sheets), https://support.claude.com/en/articles/17153992 (Artifact https://www.youtube.com/embed/HzpYayj-8JY 143 2026-11-11T10:38:05.000Z Marvin Builds https://lianwu.app/videos/ObCN9It1nfg https://lianwu.app/media/youtube/ObCN9It1nfg.jpg No ice: should the cafe fill the cup? Three AIs debated it (narrated) | Roundtable #Shorts Lian Wu (Mac): https://lianwu.app/yt In this version I narrate it myself, in Chinese. A customer orders an iced drink with no ice or less ice, and the cup ends up with a gap at the top. Should the cafe fill it? I gave a coffee shop manager's two options to three AIs in Roundtable for three rounds, each on its own model, all on Low effort: A. Fill the cup with the drink for free. B. By default give only the recipe's amount, and let the customer pay extra (say NT$15) to fill it. - Ella (GPT-6 Luna): B in every round. Filling it for free waters down the taste. - Grace (Gemini 3.5 Flash Lite): pushed back throughout. Customers in Taiwan treat the cup's height as what they paid for; asking every order about the NT$15 slows the counter; customers never read the menu, so only A gets zero complaints. - Audrey (GPT-5.6 Luna, moderator): B. Free filling means regular-ice customers subsidise no-ice ones; don't upsell every order, put it on the menu instead. Conclusion: most chose B, no fill by default. Grace still leans A. You run the cafe: A or B? Tell me in the comments. The illustrations were drawn in Painter, in the same app. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). This version shows Lian Wu in Traditional Chinese. https://www.youtube.com/embed/ObCN9It1nfg 67 2026-11-10T14:31:55.000Z Marvin Builds https://lianwu.app/zh-TW/videos/ObCN9It1nfg https://lianwu.app/media/youtube/ObCN9It1nfg.jpg 去冰、少冰要不要補滿？三個 AI 認真討論（配音版）｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 這一版是我自己用中文配音。 客人點冰飲選去冰或少冰，杯子會空出一截，店家要不要補滿？我在 Roundtable 讓三個 AI 討論了三輪，題目是咖啡店店長的兩個做法： A. 免費用飲料補滿。 B. 預設不補，只給原配方的飲料量，但客人可以加價（例如 15 元）補滿。 三個 AI 各用不同的模型，思考程度都設「低」。 - Ella（GPT-6 Luna）：三輪都選 B。免費補滿，口味會變淡。 - Grace（Gemini 3.5 Flash Lite）：一路挑戰。台灣人把杯子的高度當成付的錢；每單都問要不要加 15 元，點餐會變慢；客人根本不讀菜單，A 免費補滿才能零客訴。 - Audrey（GPT-5.6 Luna，主持人）：選 B。免費補滿，等於正常冰的人補貼去冰的人；不要每單推銷，改在菜單標清楚。 結論：多數選 B，預設不補。Grace 仍然傾向 A。 你是店長，選 A 還是 B？留言告訴我。 插圖是用同一個 App 的 Painter 畫的。 Roundtable 是我正在做的 Mac 應用程式 Lian Wu 裡的 Guild（Apple 晶片的 Mac）。這一版的 Lian Wu 是繁體中文介面。 https://www.youtube.com/embed/ObCN9It1nfg 67 2026-11-10T14:31:55.000Z Marvin Builds https://lianwu.app/ja/videos/ObCN9It1nfg https://lianwu.app/media/youtube/ObCN9It1nfg.jpg 去冰、少冰要不要補滿？三個 AI 認真討論（配音版）｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 這一版是我自己用中文配音。 客人點冰飲選去冰或少冰，杯子會空出一截，店家要不要補滿？我在 Roundtable 讓三個 AI 討論了三輪，題目是咖啡店店長的兩個做法： A. 免費用飲料補滿。 B. 預設不補，只給原配方的飲料量，但客人可以加價（例如 15 元）補滿。 三個 AI 各用不同的模型，思考程度都設「低」。 - Ella（GPT-6 Luna）：三輪都選 B。免費補滿，口味會變淡。 - Grace（Gemini 3.5 Flash Lite）：一路挑戰。台灣人把杯子的高度當成付的錢；每單都問要不要加 15 元，點餐會變慢；客人根本不讀菜單，A 免費補滿才能零客訴。 - Audrey（GPT-5.6 Luna，主持人）：選 B。免費補滿，等於正常冰的人補貼去冰的人；不要每單推銷，改在菜單標清楚。 結論：多數選 B，預設不補。Grace 仍然傾向 A。 你是店長，選 A 還是 B？留言告訴我。 插圖是用同一個 App 的 Painter 畫的。 Roundtable 是我正在做的 Mac 應用程式 Lian Wu 裡的 Guild（Apple 晶片的 Mac）。這一版的 Lian Wu 是繁體中文介面。 https://www.youtube.com/embed/ObCN9It1nfg 67 2026-11-10T14:31:55.000Z Marvin Builds https://lianwu.app/ko/videos/ObCN9It1nfg https://lianwu.app/media/youtube/ObCN9It1nfg.jpg 去冰、少冰要不要補滿？三個 AI 認真討論（配音版）｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 這一版是我自己用中文配音。 客人點冰飲選去冰或少冰，杯子會空出一截，店家要不要補滿？我在 Roundtable 讓三個 AI 討論了三輪，題目是咖啡店店長的兩個做法： A. 免費用飲料補滿。 B. 預設不補，只給原配方的飲料量，但客人可以加價（例如 15 元）補滿。 三個 AI 各用不同的模型，思考程度都設「低」。 - Ella（GPT-6 Luna）：三輪都選 B。免費補滿，口味會變淡。 - Grace（Gemini 3.5 Flash Lite）：一路挑戰。台灣人把杯子的高度當成付的錢；每單都問要不要加 15 元，點餐會變慢；客人根本不讀菜單，A 免費補滿才能零客訴。 - Audrey（GPT-5.6 Luna，主持人）：選 B。免費補滿，等於正常冰的人補貼去冰的人；不要每單推銷，改在菜單標清楚。 結論：多數選 B，預設不補。Grace 仍然傾向 A。 你是店長，選 A 還是 B？留言告訴我。 插圖是用同一個 App 的 Painter 畫的。 Roundtable 是我正在做的 Mac 應用程式 Lian Wu 裡的 Guild（Apple 晶片的 Mac）。這一版的 Lian Wu 是繁體中文介面。 https://www.youtube.com/embed/ObCN9It1nfg 67 2026-11-10T14:31:55.000Z Marvin Builds https://lianwu.app/videos/U2XJam5H3Lg https://lianwu.app/media/youtube/U2XJam5H3Lg.jpg Claude Fable 5.2 built an operating system. Here's what else is new Claude Fable 5.2 is out. Here's what's new, from Anthropic's announcement of November 9, 2026. - One Anthropic engineer gave it an empty folder. Eight days and over 1,000 subagents later, it had an operating system that boots on a laptop. - Fable 5.2 and Mythos 5.2 are the same underlying model. Fable has strong cybersecurity and biology safeguards, so it's open to everyone. Mythos goes to vetted organizations. - Up to 30% faster than Fable 5.1, and cheaper: $8 per million input tokens, $40 per million output. An optional fast mode is up to 2.5 times faster, for twice the price. - FrontierCode: on High effort it scored about the same as Opus 5.5 at its best (45.3% vs 45.5%), at roughly half the cost per task. On Xhigh it reached 49.1%, the highest of any released model Anthropic tested. - Before launch: three working electronic devices from a one-paragraph prompt, real robots, two open problems in physics, and Fermat's Last Theorem formalized by 1,000 agents in 30 hours. - About 55% fewer blocks than Fable 5.1. A blocked request is answered by Opus 4.8 for cybersecurity, or Opus 5 for biology. - Anthropic's advice: start with Opus 5.5 for most coding and knowledge work, move up to Fable for the most ambitious work, or down to Sonnet or Haiku. Chapters 0:00 Fable 5.2 built an OS 0:14 Two names, one model 0:30 Faster and cheaper 0:53 Benchmarks 1:10 Before launch 1:31 Fewer refusals 1:46 When to use it 2:01 Where to get it 2:11 My take Source: Anthropic, Claude Fable 5.2 and Claude Mythos 5.2 https://www.anthropic.com/claude-fable-and-mythos-5-2 Charts, videos and quoted text in this video: anthropic.com. https://www.youtube.com/embed/U2XJam5H3Lg 154 2026-11-10T13:40:28.000Z Marvin Builds https://lianwu.app/zh-TW/videos/U2XJam5H3Lg https://lianwu.app/media/youtube/U2XJam5H3Lg.jpg Claude Fable 5.2 built an operating system. Here's what else is new Claude Fable 5.2 is out. Here's what's new, from Anthropic's announcement of November 9, 2026. - One Anthropic engineer gave it an empty folder. Eight days and over 1,000 subagents later, it had an operating system that boots on a laptop. - Fable 5.2 and Mythos 5.2 are the same underlying model. Fable has strong cybersecurity and biology safeguards, so it's open to everyone. Mythos goes to vetted organizations. - Up to 30% faster than Fable 5.1, and cheaper: $8 per million input tokens, $40 per million output. An optional fast mode is up to 2.5 times faster, for twice the price. - FrontierCode: on High effort it scored about the same as Opus 5.5 at its best (45.3% vs 45.5%), at roughly half the cost per task. On Xhigh it reached 49.1%, the highest of any released model Anthropic tested. - Before launch: three working electronic devices from a one-paragraph prompt, real robots, two open problems in physics, and Fermat's Last Theorem formalized by 1,000 agents in 30 hours. - About 55% fewer blocks than Fable 5.1. A blocked request is answered by Opus 4.8 for cybersecurity, or Opus 5 for biology. - Anthropic's advice: start with Opus 5.5 for most coding and knowledge work, move up to Fable for the most ambitious work, or down to Sonnet or Haiku. Chapters 0:00 Fable 5.2 built an OS 0:14 Two names, one model 0:30 Faster and cheaper 0:53 Benchmarks 1:10 Before launch 1:31 Fewer refusals 1:46 When to use it 2:01 Where to get it 2:11 My take Source: Anthropic, Claude Fable 5.2 and Claude Mythos 5.2 https://www.anthropic.com/claude-fable-and-mythos-5-2 Charts, videos and quoted text in this video: anthropic.com. https://www.youtube.com/embed/U2XJam5H3Lg 154 2026-11-10T13:40:28.000Z Marvin Builds https://lianwu.app/ja/videos/U2XJam5H3Lg https://lianwu.app/media/youtube/U2XJam5H3Lg.jpg Claude Fable 5.2 built an operating system. Here's what else is new Claude Fable 5.2 is out. Here's what's new, from Anthropic's announcement of November 9, 2026. - One Anthropic engineer gave it an empty folder. Eight days and over 1,000 subagents later, it had an operating system that boots on a laptop. - Fable 5.2 and Mythos 5.2 are the same underlying model. Fable has strong cybersecurity and biology safeguards, so it's open to everyone. Mythos goes to vetted organizations. - Up to 30% faster than Fable 5.1, and cheaper: $8 per million input tokens, $40 per million output. An optional fast mode is up to 2.5 times faster, for twice the price. - FrontierCode: on High effort it scored about the same as Opus 5.5 at its best (45.3% vs 45.5%), at roughly half the cost per task. On Xhigh it reached 49.1%, the highest of any released model Anthropic tested. - Before launch: three working electronic devices from a one-paragraph prompt, real robots, two open problems in physics, and Fermat's Last Theorem formalized by 1,000 agents in 30 hours. - About 55% fewer blocks than Fable 5.1. A blocked request is answered by Opus 4.8 for cybersecurity, or Opus 5 for biology. - Anthropic's advice: start with Opus 5.5 for most coding and knowledge work, move up to Fable for the most ambitious work, or down to Sonnet or Haiku. Chapters 0:00 Fable 5.2 built an OS 0:14 Two names, one model 0:30 Faster and cheaper 0:53 Benchmarks 1:10 Before launch 1:31 Fewer refusals 1:46 When to use it 2:01 Where to get it 2:11 My take Source: Anthropic, Claude Fable 5.2 and Claude Mythos 5.2 https://www.anthropic.com/claude-fable-and-mythos-5-2 Charts, videos and quoted text in this video: anthropic.com. https://www.youtube.com/embed/U2XJam5H3Lg 154 2026-11-10T13:40:28.000Z Marvin Builds https://lianwu.app/ko/videos/U2XJam5H3Lg https://lianwu.app/media/youtube/U2XJam5H3Lg.jpg Claude Fable 5.2 built an operating system. Here's what else is new Claude Fable 5.2 is out. Here's what's new, from Anthropic's announcement of November 9, 2026. - One Anthropic engineer gave it an empty folder. Eight days and over 1,000 subagents later, it had an operating system that boots on a laptop. - Fable 5.2 and Mythos 5.2 are the same underlying model. Fable has strong cybersecurity and biology safeguards, so it's open to everyone. Mythos goes to vetted organizations. - Up to 30% faster than Fable 5.1, and cheaper: $8 per million input tokens, $40 per million output. An optional fast mode is up to 2.5 times faster, for twice the price. - FrontierCode: on High effort it scored about the same as Opus 5.5 at its best (45.3% vs 45.5%), at roughly half the cost per task. On Xhigh it reached 49.1%, the highest of any released model Anthropic tested. - Before launch: three working electronic devices from a one-paragraph prompt, real robots, two open problems in physics, and Fermat's Last Theorem formalized by 1,000 agents in 30 hours. - About 55% fewer blocks than Fable 5.1. A blocked request is answered by Opus 4.8 for cybersecurity, or Opus 5 for biology. - Anthropic's advice: start with Opus 5.5 for most coding and knowledge work, move up to Fable for the most ambitious work, or down to Sonnet or Haiku. Chapters 0:00 Fable 5.2 built an OS 0:14 Two names, one model 0:30 Faster and cheaper 0:53 Benchmarks 1:10 Before launch 1:31 Fewer refusals 1:46 When to use it 2:01 Where to get it 2:11 My take Source: Anthropic, Claude Fable 5.2 and Claude Mythos 5.2 https://www.anthropic.com/claude-fable-and-mythos-5-2 Charts, videos and quoted text in this video: anthropic.com. https://www.youtube.com/embed/U2XJam5H3Lg 154 2026-11-10T13:40:28.000Z Marvin Builds https://lianwu.app/videos/Vmt7JkrKWIY https://lianwu.app/media/youtube/Vmt7JkrKWIY.jpg No ice or less ice: should the cafe fill the cup? Three AIs debated it | Roundtable #Shorts Lian Wu (Mac): https://lianwu.app/yt A customer orders an iced drink with no ice or less ice, and the cup ends up with a gap at the top. Should the cafe fill it? I gave a coffee shop manager's two options to three AIs in Roundtable for three rounds, each on its own model, all on Low effort: A. Fill the cup with the drink for free. B. By default give only the recipe's amount, and let the customer pay extra (say NT$15) to fill it. - Ella (GPT-6 Luna): B in every round. Filling it for free waters down the taste. - Grace (Gemini 3.5 Flash Lite): pushed back throughout. Customers in Taiwan treat the cup's height as what they paid for; asking every order about the NT$15 slows the counter; customers never read the menu, so only A gets zero complaints. - Audrey (GPT-5.6 Luna, moderator): B. Free filling means regular-ice customers subsidise no-ice ones; don't upsell every order, put it on the menu instead. Conclusion: most chose B, no fill by default. Grace still leans A. You run the cafe: A or B? Tell me in the comments. The illustrations were drawn in Painter, in the same app. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). This version shows Lian Wu in Traditional Chinese. https://www.youtube.com/embed/Vmt7JkrKWIY 60 2026-11-10T05:01:36.000Z Marvin Builds https://lianwu.app/zh-TW/videos/Vmt7JkrKWIY https://lianwu.app/media/youtube/Vmt7JkrKWIY.jpg 去冰、少冰要不要補滿？三個 AI 認真討論｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 客人點冰飲選去冰或少冰，杯子會空出一截，店家要不要補滿？我在 Roundtable 讓三個 AI 討論了三輪，題目是咖啡店店長的兩個做法： A. 免費用飲料補滿。 B. 預設不補，只給原配方的飲料量，但客人可以加價（例如 15 元）補滿。 三個 AI 各用不同的模型，思考程度都設「低」。 - Ella（GPT-6 Luna）：三輪都選 B。免費補滿，口味會變淡。 - Grace（Gemini 3.5 Flash Lite）：一路挑戰。台灣人把杯子的高度當成付的錢；每單都問要不要加 15 元，點餐會變慢；客人根本不讀菜單，A 免費補滿才能零客訴。 - Audrey（GPT-5.6 Luna，主持人）：選 B。免費補滿，等於正常冰的人補貼去冰的人；不要每單推銷，改在菜單標清楚。 結論：多數選 B，預設不補。Grace 仍然傾向 A。 你是店長，選 A 還是 B？留言告訴我。 插圖是用同一個 App 的 Painter 畫的。 Roundtable 是我正在做的 Mac 應用程式 Lian Wu 裡的 Guild（Apple 晶片的 Mac）。這一版的 Lian Wu 是繁體中文介面。 https://www.youtube.com/embed/Vmt7JkrKWIY 60 2026-11-10T05:01:36.000Z Marvin Builds https://lianwu.app/ja/videos/Vmt7JkrKWIY https://lianwu.app/media/youtube/Vmt7JkrKWIY.jpg 去冰、少冰要不要補滿？三個 AI 認真討論｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 客人點冰飲選去冰或少冰，杯子會空出一截，店家要不要補滿？我在 Roundtable 讓三個 AI 討論了三輪，題目是咖啡店店長的兩個做法： A. 免費用飲料補滿。 B. 預設不補，只給原配方的飲料量，但客人可以加價（例如 15 元）補滿。 三個 AI 各用不同的模型，思考程度都設「低」。 - Ella（GPT-6 Luna）：三輪都選 B。免費補滿，口味會變淡。 - Grace（Gemini 3.5 Flash Lite）：一路挑戰。台灣人把杯子的高度當成付的錢；每單都問要不要加 15 元，點餐會變慢；客人根本不讀菜單，A 免費補滿才能零客訴。 - Audrey（GPT-5.6 Luna，主持人）：選 B。免費補滿，等於正常冰的人補貼去冰的人；不要每單推銷，改在菜單標清楚。 結論：多數選 B，預設不補。Grace 仍然傾向 A。 你是店長，選 A 還是 B？留言告訴我。 插圖是用同一個 App 的 Painter 畫的。 Roundtable 是我正在做的 Mac 應用程式 Lian Wu 裡的 Guild（Apple 晶片的 Mac）。這一版的 Lian Wu 是繁體中文介面。 https://www.youtube.com/embed/Vmt7JkrKWIY 60 2026-11-10T05:01:36.000Z Marvin Builds https://lianwu.app/ko/videos/Vmt7JkrKWIY https://lianwu.app/media/youtube/Vmt7JkrKWIY.jpg 去冰、少冰要不要補滿？三個 AI 認真討論｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 客人點冰飲選去冰或少冰，杯子會空出一截，店家要不要補滿？我在 Roundtable 讓三個 AI 討論了三輪，題目是咖啡店店長的兩個做法： A. 免費用飲料補滿。 B. 預設不補，只給原配方的飲料量，但客人可以加價（例如 15 元）補滿。 三個 AI 各用不同的模型，思考程度都設「低」。 - Ella（GPT-6 Luna）：三輪都選 B。免費補滿，口味會變淡。 - Grace（Gemini 3.5 Flash Lite）：一路挑戰。台灣人把杯子的高度當成付的錢；每單都問要不要加 15 元，點餐會變慢；客人根本不讀菜單，A 免費補滿才能零客訴。 - Audrey（GPT-5.6 Luna，主持人）：選 B。免費補滿，等於正常冰的人補貼去冰的人；不要每單推銷，改在菜單標清楚。 結論：多數選 B，預設不補。Grace 仍然傾向 A。 你是店長，選 A 還是 B？留言告訴我。 插圖是用同一個 App 的 Painter 畫的。 Roundtable 是我正在做的 Mac 應用程式 Lian Wu 裡的 Guild（Apple 晶片的 Mac）。這一版的 Lian Wu 是繁體中文介面。 https://www.youtube.com/embed/Vmt7JkrKWIY 60 2026-11-10T05:01:36.000Z Marvin Builds https://lianwu.app/videos/EIc2X4M8FGg https://lianwu.app/media/youtube/EIc2X4M8FGg.jpg One saved picture, one word: how I use ChatGPT Images 2.6 #Shorts Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt How I actually use ChatGPT Images 2.6. Part 2, in under a minute. - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me the thumbnail, a post image and the Short cover. No upload. - Markup: in part 1, a GIF frame had a ChatGPT logo on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. - GIFs: I asked for one dance, planned across all 12 frames, with the camera held still. The wax apple sways, waves and winks. Full video, with chapters: https://youtu.be/f7AhPYTHkRk Part 1: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/EIc2X4M8FGg 49 2026-11-09T13:06:10.000Z Marvin Builds https://lianwu.app/zh-TW/videos/EIc2X4M8FGg https://lianwu.app/media/youtube/EIc2X4M8FGg.jpg One saved picture, one word: how I use ChatGPT Images 2.6 #Shorts Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt How I actually use ChatGPT Images 2.6. Part 2, in under a minute. - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me the thumbnail, a post image and the Short cover. No upload. - Markup: in part 1, a GIF frame had a ChatGPT logo on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. - GIFs: I asked for one dance, planned across all 12 frames, with the camera held still. The wax apple sways, waves and winks. Full video, with chapters: https://youtu.be/f7AhPYTHkRk Part 1: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/EIc2X4M8FGg 49 2026-11-09T13:06:10.000Z Marvin Builds https://lianwu.app/ja/videos/EIc2X4M8FGg https://lianwu.app/media/youtube/EIc2X4M8FGg.jpg One saved picture, one word: how I use ChatGPT Images 2.6 #Shorts Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt How I actually use ChatGPT Images 2.6. Part 2, in under a minute. - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me the thumbnail, a post image and the Short cover. No upload. - Markup: in part 1, a GIF frame had a ChatGPT logo on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. - GIFs: I asked for one dance, planned across all 12 frames, with the camera held still. The wax apple sways, waves and winks. Full video, with chapters: https://youtu.be/f7AhPYTHkRk Part 1: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/EIc2X4M8FGg 49 2026-11-09T13:06:10.000Z Marvin Builds https://lianwu.app/ko/videos/EIc2X4M8FGg https://lianwu.app/media/youtube/EIc2X4M8FGg.jpg One saved picture, one word: how I use ChatGPT Images 2.6 #Shorts Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt How I actually use ChatGPT Images 2.6. Part 2, in under a minute. - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me the thumbnail, a post image and the Short cover. No upload. - Markup: in part 1, a GIF frame had a ChatGPT logo on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. - GIFs: I asked for one dance, planned across all 12 frames, with the camera held still. The wax apple sways, waves and winks. Full video, with chapters: https://youtu.be/f7AhPYTHkRk Part 1: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/EIc2X4M8FGg 49 2026-11-09T13:06:10.000Z Marvin Builds https://lianwu.app/videos/29NSs7xhYnM https://lianwu.app/media/youtube/29NSs7xhYnM.jpg Is gimbap without seaweed still gimbap? Three AIs debated it | Roundtable #Shorts Lian Wu (Mac): https://lianwu.app/yt If you take the seaweed (gim) out of gimbap, is it still gimbap? Is a food's name decided by its ingredients, its shape or its culture? I gave it to three AIs in Roundtable for three rounds, each on its own model, all on Low effort. - Ella (GPT-6 Luna): yes at first, since the rolled shape and the dish's lineage remain; no by round 3. - Grace (Gemini 3.5 Flash Lite): no. Gimbap means rice rolled in gim; rolled in egg it is a roll. Would a burger without a bun still be a burger? Borrowing the name needs legal and social guidelines first. - Audrey (GPT-5.6 Luna, moderator): no. Without gim, name it by what wraps it, like egg-rolled rice. Final call: no. Take the gim out and it is not gimbap. Yes or no? Tell me in the comments. The illustrations were drawn in Painter, in the same app. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). This version shows Lian Wu in Korean. https://www.youtube.com/embed/29NSs7xhYnM 58 2026-11-09T12:31:55.000Z Marvin Builds https://lianwu.app/zh-TW/videos/29NSs7xhYnM https://lianwu.app/media/youtube/29NSs7xhYnM.jpg 김밥에서 김을 빼도 김밥일까? AI 셋이 진지하게 토론｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 김밥에서 김을 빼도 김밥일까? 음식의 이름은 재료, 모양, 문화 중 무엇으로 정해질까요? Roundtable에서 AI 셋에게 3라운드 토론을 시켰습니다. 모델은 각각 다르고, 추론 수준은 모두 '낮음'. - Ella (GPT-6 Luna): 처음엔 '예'. 김이 빠져도 말아 만든 모양과 김밥이라는 문화적 계보는 남는다. 3라운드에서는 '아니요'로 바꿨습니다. - Grace (Gemini 3.5 Flash Lite): 아니요. 김밥은 '김으로 밥을 말아 만든 음식'. 달걀로 말면 '롤'이나 '말이'다. 김치 없는 김치찌개, 빵 없는 햄버거도 그 이름으로 불러야 하느냐고 묻고, 이름을 빌려 쓰려면 법적·사회적 가이드라인부터 필요하다고 했습니다. - Audrey (GPT-5.6 Luna, 진행자): 아니요. 김이 없으면 '달걀말이밥'처럼 실제 재료로 이름을 붙이자. 최종 판단: 아니요. 일반적인 의미에서 김을 빼면 김밥이 아닙니다. 여러분은 '예'? '아니요'? 댓글로 알려 주세요. 일러스트는 같은 앱의 Painter로 그렸습니다. Roundtable은 제가 만들고 있는 Mac 앱 'Lian Wu'의 Guild입니다 (Apple 실리콘 Mac용). https://www.youtube.com/embed/29NSs7xhYnM 58 2026-11-09T12:31:55.000Z Marvin Builds https://lianwu.app/ja/videos/29NSs7xhYnM https://lianwu.app/media/youtube/29NSs7xhYnM.jpg 김밥에서 김을 빼도 김밥일까? AI 셋이 진지하게 토론｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 김밥에서 김을 빼도 김밥일까? 음식의 이름은 재료, 모양, 문화 중 무엇으로 정해질까요? Roundtable에서 AI 셋에게 3라운드 토론을 시켰습니다. 모델은 각각 다르고, 추론 수준은 모두 '낮음'. - Ella (GPT-6 Luna): 처음엔 '예'. 김이 빠져도 말아 만든 모양과 김밥이라는 문화적 계보는 남는다. 3라운드에서는 '아니요'로 바꿨습니다. - Grace (Gemini 3.5 Flash Lite): 아니요. 김밥은 '김으로 밥을 말아 만든 음식'. 달걀로 말면 '롤'이나 '말이'다. 김치 없는 김치찌개, 빵 없는 햄버거도 그 이름으로 불러야 하느냐고 묻고, 이름을 빌려 쓰려면 법적·사회적 가이드라인부터 필요하다고 했습니다. - Audrey (GPT-5.6 Luna, 진행자): 아니요. 김이 없으면 '달걀말이밥'처럼 실제 재료로 이름을 붙이자. 최종 판단: 아니요. 일반적인 의미에서 김을 빼면 김밥이 아닙니다. 여러분은 '예'? '아니요'? 댓글로 알려 주세요. 일러스트는 같은 앱의 Painter로 그렸습니다. Roundtable은 제가 만들고 있는 Mac 앱 'Lian Wu'의 Guild입니다 (Apple 실리콘 Mac용). https://www.youtube.com/embed/29NSs7xhYnM 58 2026-11-09T12:31:55.000Z Marvin Builds https://lianwu.app/ko/videos/29NSs7xhYnM https://lianwu.app/media/youtube/29NSs7xhYnM.jpg 김밥에서 김을 빼도 김밥일까? AI 셋이 진지하게 토론｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt 김밥에서 김을 빼도 김밥일까? 음식의 이름은 재료, 모양, 문화 중 무엇으로 정해질까요? Roundtable에서 AI 셋에게 3라운드 토론을 시켰습니다. 모델은 각각 다르고, 추론 수준은 모두 '낮음'. - Ella (GPT-6 Luna): 처음엔 '예'. 김이 빠져도 말아 만든 모양과 김밥이라는 문화적 계보는 남는다. 3라운드에서는 '아니요'로 바꿨습니다. - Grace (Gemini 3.5 Flash Lite): 아니요. 김밥은 '김으로 밥을 말아 만든 음식'. 달걀로 말면 '롤'이나 '말이'다. 김치 없는 김치찌개, 빵 없는 햄버거도 그 이름으로 불러야 하느냐고 묻고, 이름을 빌려 쓰려면 법적·사회적 가이드라인부터 필요하다고 했습니다. - Audrey (GPT-5.6 Luna, 진행자): 아니요. 김이 없으면 '달걀말이밥'처럼 실제 재료로 이름을 붙이자. 최종 판단: 아니요. 일반적인 의미에서 김을 빼면 김밥이 아닙니다. 여러분은 '예'? '아니요'? 댓글로 알려 주세요. 일러스트는 같은 앱의 Painter로 그렸습니다. Roundtable은 제가 만들고 있는 Mac 앱 'Lian Wu'의 Guild입니다 (Apple 실리콘 Mac용). https://www.youtube.com/embed/29NSs7xhYnM 58 2026-11-09T12:31:55.000Z Marvin Builds https://lianwu.app/videos/u9CAGUGadP0 https://lianwu.app/media/youtube/u9CAGUGadP0.jpg Is curry a drink? Three AIs debated it | Roundtable #Shorts Lian Wu (Mac): https://lianwu.app/yt Is curry a drink? It is a long-running argument on the Japanese internet, so I gave it to three AIs in Roundtable for three rounds, each on its own model, all on Low effort. - Ella (GPT-6 Luna): no. Being swallowable does not make it a drink. - Grace (Gemini 3.5 Flash Lite): potage and smoothies are food and drinks at once. Picture curry in a jelly-drink pouch. Once you admit exceptions, a flat no falls apart. - Audrey (GPT-5.6 Luna, moderator): no. Ordinary curry is food. The summary: ordinary curry, no. Curry made to drink could be called a drink. Yes or no? Tell me in the comments. The illustrations were drawn in Painter, in the same app. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). This version shows Lian Wu in Japanese. https://www.youtube.com/embed/u9CAGUGadP0 58 2026-11-09T09:12:10.000Z Marvin Builds https://lianwu.app/zh-TW/videos/u9CAGUGadP0 https://lianwu.app/media/youtube/u9CAGUGadP0.jpg カレーは飲み物か？AI 3体が本気で議論｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt カレーは飲み物か？ネットで長年続くこの論争を、Roundtable で AI 3体に3ラウンド議論させました。モデルはそれぞれ別、推論はすべて「低い」。 - Ella（GPT-6 Luna）：いいえ。飲み込めることと、飲み物であることは別。 - Grace（Gemini 3.5 Flash Lite）：ポタージュもスムージーも、料理であり飲み物では？ゼリー飲料型のカレーを想像して。例外を認めた瞬間、一律の「いいえ」は崩れる。 - Audrey（GPT-5.6 Luna、司会）：いいえ。一般的なカレーは料理です。 まとめ：ふつうのカレーなら「いいえ」。飲む用に調整されたカレーなら、飲み物と呼ぶ余地はある。 あなたは「はい」？「いいえ」？コメントで教えてください。 イラストは同じアプリの Painter で描きました。 Roundtable は、私が作っている Mac アプリ「Lian Wu」の Guild です（Apple シリコン搭載の Mac 用）。 https://www.youtube.com/embed/u9CAGUGadP0 58 2026-11-09T09:12:10.000Z Marvin Builds https://lianwu.app/ja/videos/u9CAGUGadP0 https://lianwu.app/media/youtube/u9CAGUGadP0.jpg カレーは飲み物か？AI 3体が本気で議論｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt カレーは飲み物か？ネットで長年続くこの論争を、Roundtable で AI 3体に3ラウンド議論させました。モデルはそれぞれ別、推論はすべて「低い」。 - Ella（GPT-6 Luna）：いいえ。飲み込めることと、飲み物であることは別。 - Grace（Gemini 3.5 Flash Lite）：ポタージュもスムージーも、料理であり飲み物では？ゼリー飲料型のカレーを想像して。例外を認めた瞬間、一律の「いいえ」は崩れる。 - Audrey（GPT-5.6 Luna、司会）：いいえ。一般的なカレーは料理です。 まとめ：ふつうのカレーなら「いいえ」。飲む用に調整されたカレーなら、飲み物と呼ぶ余地はある。 あなたは「はい」？「いいえ」？コメントで教えてください。 イラストは同じアプリの Painter で描きました。 Roundtable は、私が作っている Mac アプリ「Lian Wu」の Guild です（Apple シリコン搭載の Mac 用）。 https://www.youtube.com/embed/u9CAGUGadP0 58 2026-11-09T09:12:10.000Z Marvin Builds https://lianwu.app/ko/videos/u9CAGUGadP0 https://lianwu.app/media/youtube/u9CAGUGadP0.jpg カレーは飲み物か？AI 3体が本気で議論｜Roundtable #Shorts Lian Wu（Mac）: https://lianwu.app/yt カレーは飲み物か？ネットで長年続くこの論争を、Roundtable で AI 3体に3ラウンド議論させました。モデルはそれぞれ別、推論はすべて「低い」。 - Ella（GPT-6 Luna）：いいえ。飲み込めることと、飲み物であることは別。 - Grace（Gemini 3.5 Flash Lite）：ポタージュもスムージーも、料理であり飲み物では？ゼリー飲料型のカレーを想像して。例外を認めた瞬間、一律の「いいえ」は崩れる。 - Audrey（GPT-5.6 Luna、司会）：いいえ。一般的なカレーは料理です。 まとめ：ふつうのカレーなら「いいえ」。飲む用に調整されたカレーなら、飲み物と呼ぶ余地はある。 あなたは「はい」？「いいえ」？コメントで教えてください。 イラストは同じアプリの Painter で描きました。 Roundtable は、私が作っている Mac アプリ「Lian Wu」の Guild です（Apple シリコン搭載の Mac 用）。 https://www.youtube.com/embed/u9CAGUGadP0 58 2026-11-09T09:12:10.000Z Marvin Builds https://lianwu.app/videos/6hMx0csui-U https://lianwu.app/media/youtube/6hMx0csui-U.jpg Is a hot dog a sandwich? Three AIs debated it #Shorts Try Lian Wu (Mac): https://lianwu.app/yt A viewer on our Roundtable video asked us to settle it: is a hot dog a sandwich? So I put the question to three AIs in Roundtable, each on its own model, all on Low effort. - Ella (GPT-6 Luna) said yes: it's a filling served inside bread. - Grace (Gemini 3.5 Flash Lite) said no: a sandwich needs separate pieces of bread. By that logic, she argued, a street taco would be a sandwich too. - Audrey (GPT-5.6 Luna), the moderator, said yes in the broad sense. Three rounds: 2 yes, 1 no, and no consensus. The meeting summary: yes under a broad definition, no under a stricter culinary one. The Roundtable video that started it: https://youtu.be/eVA-QwHEf6I Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/6hMx0csui-U 44 2026-11-08T14:37:04.000Z Marvin Builds https://lianwu.app/zh-TW/videos/6hMx0csui-U https://lianwu.app/media/youtube/6hMx0csui-U.jpg Is a hot dog a sandwich? Three AIs debated it #Shorts Try Lian Wu (Mac): https://lianwu.app/yt A viewer on our Roundtable video asked us to settle it: is a hot dog a sandwich? So I put the question to three AIs in Roundtable, each on its own model, all on Low effort. - Ella (GPT-6 Luna) said yes: it's a filling served inside bread. - Grace (Gemini 3.5 Flash Lite) said no: a sandwich needs separate pieces of bread. By that logic, she argued, a street taco would be a sandwich too. - Audrey (GPT-5.6 Luna), the moderator, said yes in the broad sense. Three rounds: 2 yes, 1 no, and no consensus. The meeting summary: yes under a broad definition, no under a stricter culinary one. The Roundtable video that started it: https://youtu.be/eVA-QwHEf6I Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/6hMx0csui-U 44 2026-11-08T14:37:04.000Z Marvin Builds https://lianwu.app/ja/videos/6hMx0csui-U https://lianwu.app/media/youtube/6hMx0csui-U.jpg Is a hot dog a sandwich? Three AIs debated it #Shorts Try Lian Wu (Mac): https://lianwu.app/yt A viewer on our Roundtable video asked us to settle it: is a hot dog a sandwich? So I put the question to three AIs in Roundtable, each on its own model, all on Low effort. - Ella (GPT-6 Luna) said yes: it's a filling served inside bread. - Grace (Gemini 3.5 Flash Lite) said no: a sandwich needs separate pieces of bread. By that logic, she argued, a street taco would be a sandwich too. - Audrey (GPT-5.6 Luna), the moderator, said yes in the broad sense. Three rounds: 2 yes, 1 no, and no consensus. The meeting summary: yes under a broad definition, no under a stricter culinary one. The Roundtable video that started it: https://youtu.be/eVA-QwHEf6I Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/6hMx0csui-U 44 2026-11-08T14:37:04.000Z Marvin Builds https://lianwu.app/ko/videos/6hMx0csui-U https://lianwu.app/media/youtube/6hMx0csui-U.jpg Is a hot dog a sandwich? Three AIs debated it #Shorts Try Lian Wu (Mac): https://lianwu.app/yt A viewer on our Roundtable video asked us to settle it: is a hot dog a sandwich? So I put the question to three AIs in Roundtable, each on its own model, all on Low effort. - Ella (GPT-6 Luna) said yes: it's a filling served inside bread. - Grace (Gemini 3.5 Flash Lite) said no: a sandwich needs separate pieces of bread. By that logic, she argued, a street taco would be a sandwich too. - Audrey (GPT-5.6 Luna), the moderator, said yes in the broad sense. Three rounds: 2 yes, 1 no, and no consensus. The meeting summary: yes under a broad definition, no under a stricter culinary one. The Roundtable video that started it: https://youtu.be/eVA-QwHEf6I Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/6hMx0csui-U 44 2026-11-08T14:37:04.000Z Marvin Builds https://lianwu.app/videos/f7AhPYTHkRk https://lianwu.app/media/youtube/f7AhPYTHkRk.jpg How I actually use ChatGPT Images 2.6 (Part 2) Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt Part 2: how I actually use ChatGPT Images 2.6, in my own work. Part 1, where I tested what's new: https://youtu.be/41HjLEeBkOU What I use - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me this video's thumbnail, a post image and the Short cover. No upload. Same shape, same leaf. One tip: type the whole name, then it's first in the menu. - Markup: in part 1, a GIF frame had a ChatGPT logo drawn on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. Markup isn't new in ChatGPT; what's new is that developers get it in the API too. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: 18 paths, 6.6 KB. It's the icon on this video's end card. - GIFs: part 1's GIFs barely moved, because nobody asked for an action. A friend asked if the wax apple could dance. I asked for one dance, planned across all 12 frames, with the camera held still. It sways, it waves, it winks. - My own app now asks for one clear action across every GIF's frames. I typed just "a cat". It's not out yet; it's coming in the next update. Chapters 0:00 One saved picture, one word 0:10 Saved references: a matching set 0:33 Markup: fix one spot 0:50 SVG: a real vector from the API 1:00 GIFs that dance 1:30 What I use Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/f7AhPYTHkRk 113 2026-11-08T13:33:20.000Z Marvin Builds https://lianwu.app/zh-TW/videos/f7AhPYTHkRk https://lianwu.app/media/youtube/f7AhPYTHkRk.jpg How I actually use ChatGPT Images 2.6 (Part 2) Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt Part 2: how I actually use ChatGPT Images 2.6, in my own work. Part 1, where I tested what's new: https://youtu.be/41HjLEeBkOU What I use - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me this video's thumbnail, a post image and the Short cover. No upload. Same shape, same leaf. One tip: type the whole name, then it's first in the menu. - Markup: in part 1, a GIF frame had a ChatGPT logo drawn on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. Markup isn't new in ChatGPT; what's new is that developers get it in the API too. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: 18 paths, 6.6 KB. It's the icon on this video's end card. - GIFs: part 1's GIFs barely moved, because nobody asked for an action. A friend asked if the wax apple could dance. I asked for one dance, planned across all 12 frames, with the camera held still. It sways, it waves, it winks. - My own app now asks for one clear action across every GIF's frames. I typed just "a cat". It's not out yet; it's coming in the next update. Chapters 0:00 One saved picture, one word 0:10 Saved references: a matching set 0:33 Markup: fix one spot 0:50 SVG: a real vector from the API 1:00 GIFs that dance 1:30 What I use Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/f7AhPYTHkRk 113 2026-11-08T13:33:20.000Z Marvin Builds https://lianwu.app/ja/videos/f7AhPYTHkRk https://lianwu.app/media/youtube/f7AhPYTHkRk.jpg How I actually use ChatGPT Images 2.6 (Part 2) Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt Part 2: how I actually use ChatGPT Images 2.6, in my own work. Part 1, where I tested what's new: https://youtu.be/41HjLEeBkOU What I use - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me this video's thumbnail, a post image and the Short cover. No upload. Same shape, same leaf. One tip: type the whole name, then it's first in the menu. - Markup: in part 1, a GIF frame had a ChatGPT logo drawn on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. Markup isn't new in ChatGPT; what's new is that developers get it in the API too. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: 18 paths, 6.6 KB. It's the icon on this video's end card. - GIFs: part 1's GIFs barely moved, because nobody asked for an action. A friend asked if the wax apple could dance. I asked for one dance, planned across all 12 frames, with the camera held still. It sways, it waves, it winks. - My own app now asks for one clear action across every GIF's frames. I typed just "a cat". It's not out yet; it's coming in the next update. Chapters 0:00 One saved picture, one word 0:10 Saved references: a matching set 0:33 Markup: fix one spot 0:50 SVG: a real vector from the API 1:00 GIFs that dance 1:30 What I use Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/f7AhPYTHkRk 113 2026-11-08T13:33:20.000Z Marvin Builds https://lianwu.app/ko/videos/f7AhPYTHkRk https://lianwu.app/media/youtube/f7AhPYTHkRk.jpg How I actually use ChatGPT Images 2.6 (Part 2) Lian Wu, the Mac app I'm building to switch AI models mid-conversation: https://lianwu.app/yt Part 2: how I actually use ChatGPT Images 2.6, in my own work. Part 1, where I tested what's new: https://youtu.be/41HjLEeBkOU What I use - Saved references: I saved my mascot once, as "Lian". One prompt that calls it by name gave me this video's thumbnail, a post image and the Short cover. No upload. Same shape, same leaf. One tip: type the whole name, then it's first in the menu. - Markup: in part 1, a GIF frame had a ChatGPT logo drawn on the laptop. I marked the logo and asked for a plain lid. It changed only what I marked. Markup isn't new in ChatGPT; what's new is that developers get it in the API too. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: 18 paths, 6.6 KB. It's the icon on this video's end card. - GIFs: part 1's GIFs barely moved, because nobody asked for an action. A friend asked if the wax apple could dance. I asked for one dance, planned across all 12 frames, with the camera held still. It sways, it waves, it winks. - My own app now asks for one clear action across every GIF's frames. I typed just "a cat". It's not out yet; it's coming in the next update. Chapters 0:00 One saved picture, one word 0:10 Saved references: a matching set 0:33 Markup: fix one spot 0:50 SVG: a real vector from the API 1:00 GIFs that dance 1:30 What I use Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/f7AhPYTHkRk 113 2026-11-08T13:33:20.000Z Marvin Builds https://lianwu.app/videos/X0y-JR90vs4 https://lianwu.app/media/youtube/X0y-JR90vs4.jpg ChatGPT can now make GIFs. I tested it #Shorts Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested Images 2.6 on my own account. GIFs: a story prompt gave nine frames in about forty seconds. It drew a ChatGPT logo on the laptop, and it doesn't really loop. Fun, but they still feel like slides. SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector. Full test, with chapters: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/X0y-JR90vs4 46 2026-11-08T08:49:35.000Z Marvin Builds https://lianwu.app/zh-TW/videos/X0y-JR90vs4 https://lianwu.app/media/youtube/X0y-JR90vs4.jpg ChatGPT can now make GIFs. I tested it #Shorts Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested Images 2.6 on my own account. GIFs: a story prompt gave nine frames in about forty seconds. It drew a ChatGPT logo on the laptop, and it doesn't really loop. Fun, but they still feel like slides. SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector. Full test, with chapters: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/X0y-JR90vs4 46 2026-11-08T08:49:35.000Z Marvin Builds https://lianwu.app/ja/videos/X0y-JR90vs4 https://lianwu.app/media/youtube/X0y-JR90vs4.jpg ChatGPT can now make GIFs. I tested it #Shorts Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested Images 2.6 on my own account. GIFs: a story prompt gave nine frames in about forty seconds. It drew a ChatGPT logo on the laptop, and it doesn't really loop. Fun, but they still feel like slides. SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector. Full test, with chapters: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/X0y-JR90vs4 46 2026-11-08T08:49:35.000Z Marvin Builds https://lianwu.app/ko/videos/X0y-JR90vs4 https://lianwu.app/media/youtube/X0y-JR90vs4.jpg ChatGPT can now make GIFs. I tested it #Shorts Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested Images 2.6 on my own account. GIFs: a story prompt gave nine frames in about forty seconds. It drew a ChatGPT logo on the laptop, and it doesn't really loop. Fun, but they still feel like slides. SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector. Full test, with chapters: https://youtu.be/41HjLEeBkOU Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/X0y-JR90vs4 46 2026-11-08T08:49:35.000Z Marvin Builds https://lianwu.app/videos/41HjLEeBkOU https://lianwu.app/media/youtube/41HjLEeBkOU.jpg I tested ChatGPT Images 2.6: GIFs, saved references, SVG Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested everything new in Images 2.6 on my own account. Here is what actually works. What I found - GIFs: a simple loading GIF was six frames, one little hop. A story prompt gave nine frames in about forty seconds. Under the hood it is one image, a three by three sheet. It drew a ChatGPT logo I never asked for, and the last frame jumps right back to the first. - Saved references: I saved my logo once as "Lian", then called it by name in a new chat. Watercolor, pixel art, clay: same shape, same leaf. This one I would use. - Markup: I drew a box and asked for a small clay arm holding a laptop. It only changed the box. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: with Flare, 13 paths, about 5 KB, in 28 seconds. - Speed: OpenAI says Flare can be as fast as six seconds at low quality. On my Mac, one run each: Flare took 7.2 seconds, Sunburst 18.8 seconds. Chapters 0:00 ChatGPT can now make GIFs 0:15 What's new in Images 2.6 0:29 Test 1: GIFs 1:09 Test 2: Saved references 1:26 Test 3: Markup 1:37 Test 4: SVG 2:00 Last test: Speed, Flare vs Sunburst 2:13 What I'd use today Part 2, how I actually use it in my own work: https://youtu.be/f7AhPYTHkRk Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/41HjLEeBkOU 165 2026-11-07T16:55:41.000Z Marvin Builds https://lianwu.app/zh-TW/videos/41HjLEeBkOU https://lianwu.app/media/youtube/41HjLEeBkOU.jpg I tested ChatGPT Images 2.6: GIFs, saved references, SVG Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested everything new in Images 2.6 on my own account. Here is what actually works. What I found - GIFs: a simple loading GIF was six frames, one little hop. A story prompt gave nine frames in about forty seconds. Under the hood it is one image, a three by three sheet. It drew a ChatGPT logo I never asked for, and the last frame jumps right back to the first. - Saved references: I saved my logo once as "Lian", then called it by name in a new chat. Watercolor, pixel art, clay: same shape, same leaf. This one I would use. - Markup: I drew a box and asked for a small clay arm holding a laptop. It only changed the box. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: with Flare, 13 paths, about 5 KB, in 28 seconds. - Speed: OpenAI says Flare can be as fast as six seconds at low quality. On my Mac, one run each: Flare took 7.2 seconds, Sunburst 18.8 seconds. Chapters 0:00 ChatGPT can now make GIFs 0:15 What's new in Images 2.6 0:29 Test 1: GIFs 1:09 Test 2: Saved references 1:26 Test 3: Markup 1:37 Test 4: SVG 2:00 Last test: Speed, Flare vs Sunburst 2:13 What I'd use today Part 2, how I actually use it in my own work: https://youtu.be/f7AhPYTHkRk Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/41HjLEeBkOU 165 2026-11-07T16:55:41.000Z Marvin Builds https://lianwu.app/ja/videos/41HjLEeBkOU https://lianwu.app/media/youtube/41HjLEeBkOU.jpg I tested ChatGPT Images 2.6: GIFs, saved references, SVG Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested everything new in Images 2.6 on my own account. Here is what actually works. What I found - GIFs: a simple loading GIF was six frames, one little hop. A story prompt gave nine frames in about forty seconds. Under the hood it is one image, a three by three sheet. It drew a ChatGPT logo I never asked for, and the last frame jumps right back to the first. - Saved references: I saved my logo once as "Lian", then called it by name in a new chat. Watercolor, pixel art, clay: same shape, same leaf. This one I would use. - Markup: I drew a box and asked for a small clay arm holding a laptop. It only changed the box. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: with Flare, 13 paths, about 5 KB, in 28 seconds. - Speed: OpenAI says Flare can be as fast as six seconds at low quality. On my Mac, one run each: Flare took 7.2 seconds, Sunburst 18.8 seconds. Chapters 0:00 ChatGPT can now make GIFs 0:15 What's new in Images 2.6 0:29 Test 1: GIFs 1:09 Test 2: Saved references 1:26 Test 3: Markup 1:37 Test 4: SVG 2:00 Last test: Speed, Flare vs Sunburst 2:13 What I'd use today Part 2, how I actually use it in my own work: https://youtu.be/f7AhPYTHkRk Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/41HjLEeBkOU 165 2026-11-07T16:55:41.000Z Marvin Builds https://lianwu.app/ko/videos/41HjLEeBkOU https://lianwu.app/media/youtube/41HjLEeBkOU.jpg I tested ChatGPT Images 2.6: GIFs, saved references, SVG Lian Wu, the desktop app I'm building to switch AI models mid-conversation: https://lianwu.app/yt ChatGPT can now make GIFs. I tested everything new in Images 2.6 on my own account. Here is what actually works. What I found - GIFs: a simple loading GIF was six frames, one little hop. A story prompt gave nine frames in about forty seconds. Under the hood it is one image, a three by three sheet. It drew a ChatGPT logo I never asked for, and the last frame jumps right back to the first. - Saved references: I saved my logo once as "Lian", then called it by name in a new chat. Watercolor, pixel art, clay: same shape, same leaf. This one I would use. - Markup: I drew a box and asked for a small clay arm holding a laptop. It only changed the box. - SVG: in ChatGPT I got a PNG inside an SVG. The API gave a real vector: with Flare, 13 paths, about 5 KB, in 28 seconds. - Speed: OpenAI says Flare can be as fast as six seconds at low quality. On my Mac, one run each: Flare took 7.2 seconds, Sunburst 18.8 seconds. Chapters 0:00 ChatGPT can now make GIFs 0:15 What's new in Images 2.6 0:29 Test 1: GIFs 1:09 Test 2: Saved references 1:26 Test 3: Markup 1:37 Test 4: SVG 2:00 Last test: Speed, Flare vs Sunburst 2:13 What I'd use today Part 2, how I actually use it in my own work: https://youtu.be/f7AhPYTHkRk Source: OpenAI, Introducing Images 2.6 https://openai.com/index/introducing-images-2-6/ https://www.youtube.com/embed/41HjLEeBkOU 165 2026-11-07T16:55:41.000Z Marvin Builds https://lianwu.app/videos/iFSCPvhZoK8 https://lianwu.app/media/youtube/iFSCPvhZoK8.jpg Next month's ad budget? Ask three AIs (Roundtable) Try Lian Wu (Mac): https://lianwu.app/yt Next month's ad budget? Ask three AIs. Open Roundtable from Guilds and choose Challenge a plan. Ask how to sell next month, and attach your ads, your plan and your survey. Text files, PDFs and pictures all work, and every AI reads them each turn. Invite three AIs, each on its own model. One argues for the plan, one argues against it, and one pulls it together. After each round, add direction to steer them. Pause any time, and pick up again when you're ready. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. The product in this demo, Copperline, is made up, and so are its numbers. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/iFSCPvhZoK8 51 2026-11-07T15:25:55.000Z Marvin Builds https://lianwu.app/zh-TW/videos/iFSCPvhZoK8 https://lianwu.app/media/youtube/iFSCPvhZoK8.jpg Next month's ad budget? Ask three AIs (Roundtable) Try Lian Wu (Mac): https://lianwu.app/yt Next month's ad budget? Ask three AIs. Open Roundtable from Guilds and choose Challenge a plan. Ask how to sell next month, and attach your ads, your plan and your survey. Text files, PDFs and pictures all work, and every AI reads them each turn. Invite three AIs, each on its own model. One argues for the plan, one argues against it, and one pulls it together. After each round, add direction to steer them. Pause any time, and pick up again when you're ready. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. The product in this demo, Copperline, is made up, and so are its numbers. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/iFSCPvhZoK8 51 2026-11-07T15:25:55.000Z Marvin Builds https://lianwu.app/ja/videos/iFSCPvhZoK8 https://lianwu.app/media/youtube/iFSCPvhZoK8.jpg Next month's ad budget? Ask three AIs (Roundtable) Try Lian Wu (Mac): https://lianwu.app/yt Next month's ad budget? Ask three AIs. Open Roundtable from Guilds and choose Challenge a plan. Ask how to sell next month, and attach your ads, your plan and your survey. Text files, PDFs and pictures all work, and every AI reads them each turn. Invite three AIs, each on its own model. One argues for the plan, one argues against it, and one pulls it together. After each round, add direction to steer them. Pause any time, and pick up again when you're ready. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. The product in this demo, Copperline, is made up, and so are its numbers. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/iFSCPvhZoK8 51 2026-11-07T15:25:55.000Z Marvin Builds https://lianwu.app/ko/videos/iFSCPvhZoK8 https://lianwu.app/media/youtube/iFSCPvhZoK8.jpg Next month's ad budget? Ask three AIs (Roundtable) Try Lian Wu (Mac): https://lianwu.app/yt Next month's ad budget? Ask three AIs. Open Roundtable from Guilds and choose Challenge a plan. Ask how to sell next month, and attach your ads, your plan and your survey. Text files, PDFs and pictures all work, and every AI reads them each turn. Invite three AIs, each on its own model. One argues for the plan, one argues against it, and one pulls it together. After each round, add direction to steer them. Pause any time, and pick up again when you're ready. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. The product in this demo, Copperline, is made up, and so are its numbers. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/iFSCPvhZoK8 51 2026-11-07T15:25:55.000Z Marvin Builds https://lianwu.app/videos/W4kVZ8KAPPo https://lianwu.app/media/youtube/W4kVZ8KAPPo.jpg Big decision? Don't ask just one AI. Roundtable #Shorts Big decision? Don't ask just one AI. Roundtable puts three AIs in one discussion, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round you can add direction. When you finish, the host sums up where they agree, where they don't, and what to do next. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app https://www.youtube.com/embed/W4kVZ8KAPPo 32 2026-11-07T03:29:49.000Z Marvin Builds https://lianwu.app/zh-TW/videos/W4kVZ8KAPPo https://lianwu.app/media/youtube/W4kVZ8KAPPo.jpg Big decision? Don't ask just one AI. Roundtable #Shorts Big decision? Don't ask just one AI. Roundtable puts three AIs in one discussion, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round you can add direction. When you finish, the host sums up where they agree, where they don't, and what to do next. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app https://www.youtube.com/embed/W4kVZ8KAPPo 32 2026-11-07T03:29:49.000Z Marvin Builds https://lianwu.app/ja/videos/W4kVZ8KAPPo https://lianwu.app/media/youtube/W4kVZ8KAPPo.jpg Big decision? Don't ask just one AI. Roundtable #Shorts Big decision? Don't ask just one AI. Roundtable puts three AIs in one discussion, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round you can add direction. When you finish, the host sums up where they agree, where they don't, and what to do next. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app https://www.youtube.com/embed/W4kVZ8KAPPo 32 2026-11-07T03:29:49.000Z Marvin Builds https://lianwu.app/ko/videos/W4kVZ8KAPPo https://lianwu.app/media/youtube/W4kVZ8KAPPo.jpg Big decision? Don't ask just one AI. Roundtable #Shorts Big decision? Don't ask just one AI. Roundtable puts three AIs in one discussion, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round you can add direction. When you finish, the host sums up where they agree, where they don't, and what to do next. Roundtable is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app https://www.youtube.com/embed/W4kVZ8KAPPo 32 2026-11-07T03:29:49.000Z Marvin Builds https://lianwu.app/videos/eVA-QwHEf6I https://lianwu.app/media/youtube/eVA-QwHEf6I.jpg Roundtable: three AIs argue your decision, then sum it up Try Lian Wu (Mac): https://lianwu.app/yt Big decision? Don't ask just one AI. Open Roundtable from Guilds and pick the kind of discussion: make a decision, brainstorm ideas, challenge a plan, or explore a question. Paste your question with the numbers. You can attach a PDF, a picture or some code as background. Invite three AIs, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round, add direction to steer them. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/eVA-QwHEf6I 48 2026-11-07T03:29:48.000Z Marvin Builds https://lianwu.app/zh-TW/videos/eVA-QwHEf6I https://lianwu.app/media/youtube/eVA-QwHEf6I.jpg Roundtable: three AIs argue your decision, then sum it up Try Lian Wu (Mac): https://lianwu.app/yt Big decision? Don't ask just one AI. Open Roundtable from Guilds and pick the kind of discussion: make a decision, brainstorm ideas, challenge a plan, or explore a question. Paste your question with the numbers. You can attach a PDF, a picture or some code as background. Invite three AIs, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round, add direction to steer them. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/eVA-QwHEf6I 48 2026-11-07T03:29:48.000Z Marvin Builds https://lianwu.app/ja/videos/eVA-QwHEf6I https://lianwu.app/media/youtube/eVA-QwHEf6I.jpg Roundtable: three AIs argue your decision, then sum it up Try Lian Wu (Mac): https://lianwu.app/yt Big decision? Don't ask just one AI. Open Roundtable from Guilds and pick the kind of discussion: make a decision, brainstorm ideas, challenge a plan, or explore a question. Paste your question with the numbers. You can attach a PDF, a picture or some code as background. Invite three AIs, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round, add direction to steer them. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/eVA-QwHEf6I 48 2026-11-07T03:29:48.000Z Marvin Builds https://lianwu.app/ko/videos/eVA-QwHEf6I https://lianwu.app/media/youtube/eVA-QwHEf6I.jpg Roundtable: three AIs argue your decision, then sum it up Try Lian Wu (Mac): https://lianwu.app/yt Big decision? Don't ask just one AI. Open Roundtable from Guilds and pick the kind of discussion: make a decision, brainstorm ideas, challenge a plan, or explore a question. Paste your question with the numbers. You can attach a PDF, a picture or some code as background. Invite three AIs, each on its own model. An advocate argues for it, a critic argues against it, and a synthesizer pulls it together. After each round, add direction to steer them. Then finish, and the host sums up: where they agree, where they don't, and what to do next. Copy the discussion as Markdown, or bring it into a conversation and keep working. Roundtable is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/eVA-QwHEf6I 48 2026-11-07T03:29:48.000Z Marvin Builds https://lianwu.app/videos/A3WUM2x0hvk https://lianwu.app/media/youtube/A3WUM2x0hvk.jpg From xiaolongbao to calligraphy, four at a time | Painter #Shorts Describe a picture once in Painter, and it draws four at a time with the image model you already have: a ChatGPT plan, an OpenAI API key or OpenRouter. Pick your favorite as the reference for the next one: a calligraphy scroll of Wang Zhihuan's line "climb one more floor to see a thousand miles further". Every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app This version shows Lian Wu in Traditional Chinese. https://www.youtube.com/embed/A3WUM2x0hvk 24 2026-11-05T16:53:30.000Z Marvin Builds https://lianwu.app/zh-TW/videos/A3WUM2x0hvk https://lianwu.app/media/youtube/A3WUM2x0hvk.jpg 從小籠包到書法，一次畫四張｜畫家 Painter #Shorts 從小籠包到書法，一次畫四張。 想畫什麼，在畫家描述一次就好。用你已經有的圖像模型（ChatGPT 訂閱、OpenAI API 金鑰或 OpenRouter），一次畫四張。 挑一張喜歡的當參考圖，再用同樣的畫風寫一幅書法：王之渙〈登鸛雀樓〉的「欲窮千里目，更上一層樓」。每張圖都收在你 Mac 上的同一個圖庫。 畫家是 Lian Wu 裡的 Guild，Lian Wu 是我正在做的 Mac App（需搭載 Apple 晶片）。lianwu.app Painter in Lian Wu, in Traditional Chinese: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/A3WUM2x0hvk 24 2026-11-05T16:53:30.000Z Marvin Builds https://lianwu.app/ja/videos/A3WUM2x0hvk https://lianwu.app/media/youtube/A3WUM2x0hvk.jpg 從小籠包到書法，一次畫四張｜畫家 Painter #Shorts 從小籠包到書法，一次畫四張。 想畫什麼，在畫家描述一次就好。用你已經有的圖像模型（ChatGPT 訂閱、OpenAI API 金鑰或 OpenRouter），一次畫四張。 挑一張喜歡的當參考圖，再用同樣的畫風寫一幅書法：王之渙〈登鸛雀樓〉的「欲窮千里目，更上一層樓」。每張圖都收在你 Mac 上的同一個圖庫。 畫家是 Lian Wu 裡的 Guild，Lian Wu 是我正在做的 Mac App（需搭載 Apple 晶片）。lianwu.app Painter in Lian Wu, in Traditional Chinese: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/A3WUM2x0hvk 24 2026-11-05T16:53:30.000Z Marvin Builds https://lianwu.app/ko/videos/A3WUM2x0hvk https://lianwu.app/media/youtube/A3WUM2x0hvk.jpg 從小籠包到書法，一次畫四張｜畫家 Painter #Shorts 從小籠包到書法，一次畫四張。 想畫什麼，在畫家描述一次就好。用你已經有的圖像模型（ChatGPT 訂閱、OpenAI API 金鑰或 OpenRouter），一次畫四張。 挑一張喜歡的當參考圖，再用同樣的畫風寫一幅書法：王之渙〈登鸛雀樓〉的「欲窮千里目，更上一層樓」。每張圖都收在你 Mac 上的同一個圖庫。 畫家是 Lian Wu 裡的 Guild，Lian Wu 是我正在做的 Mac App（需搭載 Apple 晶片）。lianwu.app Painter in Lian Wu, in Traditional Chinese: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/A3WUM2x0hvk 24 2026-11-05T16:53:30.000Z Marvin Builds https://lianwu.app/videos/8zz_RbceJPU https://lianwu.app/media/youtube/8zz_RbceJPU.jpg From kimchi to taekwondo, four at a time | Painter #Shorts Describe a picture once in Painter, and it draws four at a time with the image model you already have: a ChatGPT plan, an OpenAI API key or OpenRouter. Pick your favorite as the reference for the next one: a taekwondo athlete in the same style. Every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app This version shows Lian Wu in Korean. https://www.youtube.com/embed/8zz_RbceJPU 24 2026-11-05T16:52:50.000Z Marvin Builds https://lianwu.app/zh-TW/videos/8zz_RbceJPU https://lianwu.app/media/youtube/8zz_RbceJPU.jpg 김치 밥상부터 태권도까지, 한 번에 네 장 | Painter #Shorts 김치 밥상부터 태권도까지, 한 번에 네 장. 그리고 싶은 그림을 Painter에 한 번만 설명하세요. 지금 쓰는 이미지 모델(ChatGPT 요금제, OpenAI API 키, OpenRouter)로 한 번에 네 장을 그립니다. 마음에 드는 한 장을 참고로 골라, 같은 그림체로 태권도 선수를 그렸어요. 그린 그림은 모두 내 Mac의 갤러리에 남습니다. Painter는 제가 만들고 있는 Mac 앱 Lian Wu의 Guild입니다(Apple 실리콘 Mac용). lianwu.app Painter in Lian Wu, in Korean: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/8zz_RbceJPU 24 2026-11-05T16:52:50.000Z Marvin Builds https://lianwu.app/ja/videos/8zz_RbceJPU https://lianwu.app/media/youtube/8zz_RbceJPU.jpg 김치 밥상부터 태권도까지, 한 번에 네 장 | Painter #Shorts 김치 밥상부터 태권도까지, 한 번에 네 장. 그리고 싶은 그림을 Painter에 한 번만 설명하세요. 지금 쓰는 이미지 모델(ChatGPT 요금제, OpenAI API 키, OpenRouter)로 한 번에 네 장을 그립니다. 마음에 드는 한 장을 참고로 골라, 같은 그림체로 태권도 선수를 그렸어요. 그린 그림은 모두 내 Mac의 갤러리에 남습니다. Painter는 제가 만들고 있는 Mac 앱 Lian Wu의 Guild입니다(Apple 실리콘 Mac용). lianwu.app Painter in Lian Wu, in Korean: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/8zz_RbceJPU 24 2026-11-05T16:52:50.000Z Marvin Builds https://lianwu.app/ko/videos/8zz_RbceJPU https://lianwu.app/media/youtube/8zz_RbceJPU.jpg 김치 밥상부터 태권도까지, 한 번에 네 장 | Painter #Shorts 김치 밥상부터 태권도까지, 한 번에 네 장. 그리고 싶은 그림을 Painter에 한 번만 설명하세요. 지금 쓰는 이미지 모델(ChatGPT 요금제, OpenAI API 키, OpenRouter)로 한 번에 네 장을 그립니다. 마음에 드는 한 장을 참고로 골라, 같은 그림체로 태권도 선수를 그렸어요. 그린 그림은 모두 내 Mac의 갤러리에 남습니다. Painter는 제가 만들고 있는 Mac 앱 Lian Wu의 Guild입니다(Apple 실리콘 Mac용). lianwu.app Painter in Lian Wu, in Korean: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/8zz_RbceJPU 24 2026-11-05T16:52:50.000Z Marvin Builds https://lianwu.app/videos/dBGoPlADiBQ https://lianwu.app/media/youtube/dBGoPlADiBQ.jpg Sushi and Mount Fuji, four at a time | Painter #Shorts Describe a picture once in Painter, and it draws four at a time with the image model you already have: a ChatGPT plan, an OpenAI API key or OpenRouter. Pick your favorite as the reference for the next one: Mount Fuji at sunset with cherry blossoms. Every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app This version shows Lian Wu in Japanese. https://www.youtube.com/embed/dBGoPlADiBQ 24 2026-11-05T16:52:09.000Z Marvin Builds https://lianwu.app/zh-TW/videos/dBGoPlADiBQ https://lianwu.app/media/youtube/dBGoPlADiBQ.jpg 富士山とお寿司を、一度に4枚｜Painter #Shorts 富士山とお寿司を、一度に4枚。 描きたい画像を Painter に一度説明するだけ。いま使っている画像モデル（ChatGPT プラン、OpenAI の API キー、OpenRouter）で、一度に4枚描きます。 気に入った一枚を参考にして、次は夕焼けの富士山と桜並木。描いた画像は、すべて Mac のギャラリーに残ります。 Painter は、私が作っている Mac アプリ「Lian Wu」の Guild です（Apple シリコン搭載の Mac 用）。lianwu.app Painter in Lian Wu, in Japanese: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/dBGoPlADiBQ 24 2026-11-05T16:52:09.000Z Marvin Builds https://lianwu.app/ja/videos/dBGoPlADiBQ https://lianwu.app/media/youtube/dBGoPlADiBQ.jpg 富士山とお寿司を、一度に4枚｜Painter #Shorts 富士山とお寿司を、一度に4枚。 描きたい画像を Painter に一度説明するだけ。いま使っている画像モデル（ChatGPT プラン、OpenAI の API キー、OpenRouter）で、一度に4枚描きます。 気に入った一枚を参考にして、次は夕焼けの富士山と桜並木。描いた画像は、すべて Mac のギャラリーに残ります。 Painter は、私が作っている Mac アプリ「Lian Wu」の Guild です（Apple シリコン搭載の Mac 用）。lianwu.app Painter in Lian Wu, in Japanese: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/dBGoPlADiBQ 24 2026-11-05T16:52:09.000Z Marvin Builds https://lianwu.app/ko/videos/dBGoPlADiBQ https://lianwu.app/media/youtube/dBGoPlADiBQ.jpg 富士山とお寿司を、一度に4枚｜Painter #Shorts 富士山とお寿司を、一度に4枚。 描きたい画像を Painter に一度説明するだけ。いま使っている画像モデル（ChatGPT プラン、OpenAI の API キー、OpenRouter）で、一度に4枚描きます。 気に入った一枚を参考にして、次は夕焼けの富士山と桜並木。描いた画像は、すべて Mac のギャラリーに残ります。 Painter は、私が作っている Mac アプリ「Lian Wu」の Guild です（Apple シリコン搭載の Mac 用）。lianwu.app Painter in Lian Wu, in Japanese: describe once, get four, and use your favorite as the next one's reference. https://www.youtube.com/embed/dBGoPlADiBQ 24 2026-11-05T16:52:09.000Z Marvin Builds https://lianwu.app/videos/9l-g0mICsXQ https://lianwu.app/media/youtube/9l-g0mICsXQ.jpg Painter: describe once, draw four at a time, keep every picture on your Mac Try Lian Wu (Mac): https://lianwu.app/yt Describe a picture once in Painter. Choose a shape, like square for Instagram, and how many to draw. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like and use it as the reference for the next. Every picture stays in one gallery on your Mac, and you can find any of them again by searching for what you described. Painter is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/9l-g0mICsXQ 32 2026-11-05T15:30:05.000Z Marvin Builds https://lianwu.app/zh-TW/videos/9l-g0mICsXQ https://lianwu.app/media/youtube/9l-g0mICsXQ.jpg Painter: describe once, draw four at a time, keep every picture on your Mac Try Lian Wu (Mac): https://lianwu.app/yt Describe a picture once in Painter. Choose a shape, like square for Instagram, and how many to draw. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like and use it as the reference for the next. Every picture stays in one gallery on your Mac, and you can find any of them again by searching for what you described. Painter is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/9l-g0mICsXQ 32 2026-11-05T15:30:05.000Z Marvin Builds https://lianwu.app/ja/videos/9l-g0mICsXQ https://lianwu.app/media/youtube/9l-g0mICsXQ.jpg Painter: describe once, draw four at a time, keep every picture on your Mac Try Lian Wu (Mac): https://lianwu.app/yt Describe a picture once in Painter. Choose a shape, like square for Instagram, and how many to draw. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like and use it as the reference for the next. Every picture stays in one gallery on your Mac, and you can find any of them again by searching for what you described. Painter is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/9l-g0mICsXQ 32 2026-11-05T15:30:05.000Z Marvin Builds https://lianwu.app/ko/videos/9l-g0mICsXQ https://lianwu.app/media/youtube/9l-g0mICsXQ.jpg Painter: describe once, draw four at a time, keep every picture on your Mac Try Lian Wu (Mac): https://lianwu.app/yt Describe a picture once in Painter. Choose a shape, like square for Instagram, and how many to draw. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like and use it as the reference for the next. Every picture stays in one gallery on your Mac, and you can find any of them again by searching for what you described. Painter is a Guild in Lian Wu, for Macs with Apple silicon. https://www.youtube.com/embed/9l-g0mICsXQ 32 2026-11-05T15:30:05.000Z Marvin Builds https://lianwu.app/videos/kfqbX8v8a98 https://lianwu.app/media/youtube/kfqbX8v8a98.jpg Need a picture for your next post? Painter #Shorts OpenAI just shipped Images 2.6. Next video, I put it through Painter. First, here's how Painter works. Describe a picture once in Painter. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like, use it as the reference for the next, and every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app Follow for the Images 2.6 test. https://www.youtube.com/embed/kfqbX8v8a98 24 2026-11-05T15:30:01.000Z Marvin Builds https://lianwu.app/zh-TW/videos/kfqbX8v8a98 https://lianwu.app/media/youtube/kfqbX8v8a98.jpg Need a picture for your next post? Painter #Shorts OpenAI just shipped Images 2.6. Next video, I put it through Painter. First, here's how Painter works. Describe a picture once in Painter. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like, use it as the reference for the next, and every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app Follow for the Images 2.6 test. https://www.youtube.com/embed/kfqbX8v8a98 24 2026-11-05T15:30:01.000Z Marvin Builds https://lianwu.app/ja/videos/kfqbX8v8a98 https://lianwu.app/media/youtube/kfqbX8v8a98.jpg Need a picture for your next post? Painter #Shorts OpenAI just shipped Images 2.6. Next video, I put it through Painter. First, here's how Painter works. Describe a picture once in Painter. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like, use it as the reference for the next, and every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app Follow for the Images 2.6 test. https://www.youtube.com/embed/kfqbX8v8a98 24 2026-11-05T15:30:01.000Z Marvin Builds https://lianwu.app/ko/videos/kfqbX8v8a98 https://lianwu.app/media/youtube/kfqbX8v8a98.jpg Need a picture for your next post? Painter #Shorts OpenAI just shipped Images 2.6. Next video, I put it through Painter. First, here's how Painter works. Describe a picture once in Painter. It draws four at a time with the image model you already have: your ChatGPT plan, an OpenAI key or OpenRouter. Pick the one you like, use it as the reference for the next, and every picture stays in one gallery on your Mac. Painter is a Guild in Lian Wu, the Mac app I'm building (Apple silicon). lianwu.app Follow for the Images 2.6 test. https://www.youtube.com/embed/kfqbX8v8a98 24 2026-11-05T15:30:01.000Z Marvin Builds https://lianwu.app/videos/IndtUUSWliM https://lianwu.app/media/youtube/IndtUUSWliM.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/IndtUUSWliM 31 2026-11-02T16:55:14.000Z Marvin Builds https://lianwu.app/zh-TW/videos/IndtUUSWliM https://lianwu.app/media/youtube/IndtUUSWliM.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/IndtUUSWliM 31 2026-11-02T16:55:14.000Z Marvin Builds https://lianwu.app/ja/videos/IndtUUSWliM https://lianwu.app/media/youtube/IndtUUSWliM.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/IndtUUSWliM 31 2026-11-02T16:55:14.000Z Marvin Builds https://lianwu.app/ko/videos/IndtUUSWliM https://lianwu.app/media/youtube/IndtUUSWliM.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/IndtUUSWliM 31 2026-11-02T16:55:14.000Z Marvin Builds https://lianwu.app/videos/R_J7Y-V32Q8 https://lianwu.app/media/youtube/R_J7Y-V32Q8.jpg ChatGPT and Codex will share one limit in December #Shorts ChatGPT and Codex will share one usage limit, starting in December. On Plus, Pro and Business, one allowance covers chat and Codex, counted in tokens. Stronger models and harder tasks usually use more. Instant chat won't count. So an afternoon of coding in Codex uses up your chat allowance, and the other way around. You'll see what's left, and when it resets, in Settings, then Usage. Run out, and you can wait for the reset, upgrade, or buy credits. Still unknown: the exact date, and how many tokens each plan gets. Better or worse for you? Tell me in the comments. Sources: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits https://x.com/btibor91/status/2115927832069292214 https://www.youtube.com/embed/R_J7Y-V32Q8 45 2026-11-02T15:17:16.000Z Marvin Builds https://lianwu.app/zh-TW/videos/R_J7Y-V32Q8 https://lianwu.app/media/youtube/R_J7Y-V32Q8.jpg ChatGPT and Codex will share one limit in December #Shorts ChatGPT and Codex will share one usage limit, starting in December. On Plus, Pro and Business, one allowance covers chat and Codex, counted in tokens. Stronger models and harder tasks usually use more. Instant chat won't count. So an afternoon of coding in Codex uses up your chat allowance, and the other way around. You'll see what's left, and when it resets, in Settings, then Usage. Run out, and you can wait for the reset, upgrade, or buy credits. Still unknown: the exact date, and how many tokens each plan gets. Better or worse for you? Tell me in the comments. Sources: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits https://x.com/btibor91/status/2115927832069292214 https://www.youtube.com/embed/R_J7Y-V32Q8 45 2026-11-02T15:17:16.000Z Marvin Builds https://lianwu.app/ja/videos/R_J7Y-V32Q8 https://lianwu.app/media/youtube/R_J7Y-V32Q8.jpg ChatGPT and Codex will share one limit in December #Shorts ChatGPT and Codex will share one usage limit, starting in December. On Plus, Pro and Business, one allowance covers chat and Codex, counted in tokens. Stronger models and harder tasks usually use more. Instant chat won't count. So an afternoon of coding in Codex uses up your chat allowance, and the other way around. You'll see what's left, and when it resets, in Settings, then Usage. Run out, and you can wait for the reset, upgrade, or buy credits. Still unknown: the exact date, and how many tokens each plan gets. Better or worse for you? Tell me in the comments. Sources: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits https://x.com/btibor91/status/2115927832069292214 https://www.youtube.com/embed/R_J7Y-V32Q8 45 2026-11-02T15:17:16.000Z Marvin Builds https://lianwu.app/ko/videos/R_J7Y-V32Q8 https://lianwu.app/media/youtube/R_J7Y-V32Q8.jpg ChatGPT and Codex will share one limit in December #Shorts ChatGPT and Codex will share one usage limit, starting in December. On Plus, Pro and Business, one allowance covers chat and Codex, counted in tokens. Stronger models and harder tasks usually use more. Instant chat won't count. So an afternoon of coding in Codex uses up your chat allowance, and the other way around. You'll see what's left, and when it resets, in Settings, then Usage. Run out, and you can wait for the reset, upgrade, or buy credits. Still unknown: the exact date, and how many tokens each plan gets. Better or worse for you? Tell me in the comments. Sources: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits https://x.com/btibor91/status/2115927832069292214 https://www.youtube.com/embed/R_J7Y-V32Q8 45 2026-11-02T15:17:16.000Z Marvin Builds https://lianwu.app/videos/P39SnUlSh68 https://lianwu.app/media/youtube/P39SnUlSh68.jpg Do you use more than one AI? #Shorts From my Dev Day talk, Oct 31, 2026. I use more than one AI, because each one is good at different things. I don't want to choose just one. I want to switch between them in the same conversation, and keep going. In the demo, Claude sorts 40 sample comments into themes. Then, without opening a new window, I switch to GPT. It can see the conversation and the files, so it goes straight to checking Claude's work. One AI does the work, and another one checks it. Then Painter: send the next batch while the last one is still drawing. Up to four draw at once, and the rest wait in line. QR Coder makes a code, and Painter draws it onto a truck that phones can still scan. Lian Wu is a Mac app for Apple silicon. You connect your own plan or API key, and each provider bills you directly. https://lianwu.app/yt https://www.youtube.com/embed/P39SnUlSh68 46 2026-11-01T05:50:59.000Z Marvin Builds https://lianwu.app/zh-TW/videos/P39SnUlSh68 https://lianwu.app/media/youtube/P39SnUlSh68.jpg Do you use more than one AI? #Shorts From my Dev Day talk, Oct 31, 2026. I use more than one AI, because each one is good at different things. I don't want to choose just one. I want to switch between them in the same conversation, and keep going. In the demo, Claude sorts 40 sample comments into themes. Then, without opening a new window, I switch to GPT. It can see the conversation and the files, so it goes straight to checking Claude's work. One AI does the work, and another one checks it. Then Painter: send the next batch while the last one is still drawing. Up to four draw at once, and the rest wait in line. QR Coder makes a code, and Painter draws it onto a truck that phones can still scan. Lian Wu is a Mac app for Apple silicon. You connect your own plan or API key, and each provider bills you directly. https://lianwu.app/yt https://www.youtube.com/embed/P39SnUlSh68 46 2026-11-01T05:50:59.000Z Marvin Builds https://lianwu.app/ja/videos/P39SnUlSh68 https://lianwu.app/media/youtube/P39SnUlSh68.jpg Do you use more than one AI? #Shorts From my Dev Day talk, Oct 31, 2026. I use more than one AI, because each one is good at different things. I don't want to choose just one. I want to switch between them in the same conversation, and keep going. In the demo, Claude sorts 40 sample comments into themes. Then, without opening a new window, I switch to GPT. It can see the conversation and the files, so it goes straight to checking Claude's work. One AI does the work, and another one checks it. Then Painter: send the next batch while the last one is still drawing. Up to four draw at once, and the rest wait in line. QR Coder makes a code, and Painter draws it onto a truck that phones can still scan. Lian Wu is a Mac app for Apple silicon. You connect your own plan or API key, and each provider bills you directly. https://lianwu.app/yt https://www.youtube.com/embed/P39SnUlSh68 46 2026-11-01T05:50:59.000Z Marvin Builds https://lianwu.app/ko/videos/P39SnUlSh68 https://lianwu.app/media/youtube/P39SnUlSh68.jpg Do you use more than one AI? #Shorts From my Dev Day talk, Oct 31, 2026. I use more than one AI, because each one is good at different things. I don't want to choose just one. I want to switch between them in the same conversation, and keep going. In the demo, Claude sorts 40 sample comments into themes. Then, without opening a new window, I switch to GPT. It can see the conversation and the files, so it goes straight to checking Claude's work. One AI does the work, and another one checks it. Then Painter: send the next batch while the last one is still drawing. Up to four draw at once, and the rest wait in line. QR Coder makes a code, and Painter draws it onto a truck that phones can still scan. Lian Wu is a Mac app for Apple silicon. You connect your own plan or API key, and each provider bills you directly. https://lianwu.app/yt https://www.youtube.com/embed/P39SnUlSh68 46 2026-11-01T05:50:59.000Z Marvin Builds https://lianwu.app/videos/ZainKv6-pC8 https://lianwu.app/media/youtube/ZainKv6-pC8.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/ZainKv6-pC8 31 2026-10-31T16:17:24.000Z Marvin Builds https://lianwu.app/zh-TW/videos/ZainKv6-pC8 https://lianwu.app/media/youtube/ZainKv6-pC8.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/ZainKv6-pC8 31 2026-10-31T16:17:24.000Z Marvin Builds https://lianwu.app/ja/videos/ZainKv6-pC8 https://lianwu.app/media/youtube/ZainKv6-pC8.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/ZainKv6-pC8 31 2026-10-31T16:17:24.000Z Marvin Builds https://lianwu.app/ko/videos/ZainKv6-pC8 https://lianwu.app/media/youtube/ZainKv6-pC8.jpg Claude's 5-hour limit is gone. The weekly one isn't. #Shorts Claude just removed the 5-hour limit. Only on Ultra, for verified accounts. Ultra is the new name for Max 20x. The weekly limit is still there, at ten times Pro's usage. Pro and Max still have the 5-hour limit. Verification is automatic. Check Settings, then Usage. No Current session bar means you're verified. Sources: https://claude.com/pricing https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan https://www.youtube.com/embed/ZainKv6-pC8 31 2026-10-31T16:17:24.000Z Marvin Builds https://lianwu.app/videos/T2U9kjHUIaQ https://lianwu.app/media/youtube/T2U9kjHUIaQ.jpg Can a second AI check the first one's work? #Shorts Claude sorted 40 sample comments into five themes. Then I switched to GPT in the same Lian Wu conversation and asked it to challenge the grouping. GPT pushed back on C29: onboarding, not access. ▶ Try Lian Wu: https://lianwu.app/yt Sample comments written for this demo, not real customers. Model replies are shown as recorded. Lian Wu for Mac · $20 once · 14-day full refund · Apple silicon Macs, macOS 14 or later. Edited with OpenChatCut (open source). https://www.youtube.com/embed/T2U9kjHUIaQ 25 2026-10-29T15:28:35.000Z Marvin Builds https://lianwu.app/zh-TW/videos/T2U9kjHUIaQ https://lianwu.app/media/youtube/T2U9kjHUIaQ.jpg Can a second AI check the first one's work? #Shorts Claude sorted 40 sample comments into five themes. Then I switched to GPT in the same Lian Wu conversation and asked it to challenge the grouping. GPT pushed back on C29: onboarding, not access. ▶ Try Lian Wu: https://lianwu.app/yt Sample comments written for this demo, not real customers. Model replies are shown as recorded. Lian Wu for Mac · $20 once · 14-day full refund · Apple silicon Macs, macOS 14 or later. Edited with OpenChatCut (open source). https://www.youtube.com/embed/T2U9kjHUIaQ 25 2026-10-29T15:28:35.000Z Marvin Builds https://lianwu.app/ja/videos/T2U9kjHUIaQ https://lianwu.app/media/youtube/T2U9kjHUIaQ.jpg Can a second AI check the first one's work? #Shorts Claude sorted 40 sample comments into five themes. Then I switched to GPT in the same Lian Wu conversation and asked it to challenge the grouping. GPT pushed back on C29: onboarding, not access. ▶ Try Lian Wu: https://lianwu.app/yt Sample comments written for this demo, not real customers. Model replies are shown as recorded. Lian Wu for Mac · $20 once · 14-day full refund · Apple silicon Macs, macOS 14 or later. Edited with OpenChatCut (open source). https://www.youtube.com/embed/T2U9kjHUIaQ 25 2026-10-29T15:28:35.000Z Marvin Builds https://lianwu.app/ko/videos/T2U9kjHUIaQ https://lianwu.app/media/youtube/T2U9kjHUIaQ.jpg Can a second AI check the first one's work? #Shorts Claude sorted 40 sample comments into five themes. Then I switched to GPT in the same Lian Wu conversation and asked it to challenge the grouping. GPT pushed back on C29: onboarding, not access. ▶ Try Lian Wu: https://lianwu.app/yt Sample comments written for this demo, not real customers. Model replies are shown as recorded. Lian Wu for Mac · $20 once · 14-day full refund · Apple silicon Macs, macOS 14 or later. Edited with OpenChatCut (open source). https://www.youtube.com/embed/T2U9kjHUIaQ 25 2026-10-29T15:28:35.000Z Marvin Builds https://lianwu.app/videos/gTPruloD9wQ https://lianwu.app/media/youtube/gTPruloD9wQ.jpg Claude hit its limit? Keep the same chat in GPT #Shorts Same conversation, new model. Lian Wu for Mac: $20 once, no subscription, 14-day full refund. I'm the developer. lianwu.app https://www.youtube.com/embed/gTPruloD9wQ 37 2026-10-29T09:40:29.000Z Marvin Builds https://lianwu.app/zh-TW/videos/gTPruloD9wQ https://lianwu.app/media/youtube/gTPruloD9wQ.jpg Claude hit its limit? Keep the same chat in GPT #Shorts Same conversation, new model. Lian Wu for Mac: $20 once, no subscription, 14-day full refund. I'm the developer. lianwu.app https://www.youtube.com/embed/gTPruloD9wQ 37 2026-10-29T09:40:29.000Z Marvin Builds https://lianwu.app/ja/videos/gTPruloD9wQ https://lianwu.app/media/youtube/gTPruloD9wQ.jpg Claude hit its limit? Keep the same chat in GPT #Shorts Same conversation, new model. Lian Wu for Mac: $20 once, no subscription, 14-day full refund. I'm the developer. lianwu.app https://www.youtube.com/embed/gTPruloD9wQ 37 2026-10-29T09:40:29.000Z Marvin Builds https://lianwu.app/ko/videos/gTPruloD9wQ https://lianwu.app/media/youtube/gTPruloD9wQ.jpg Claude hit its limit? Keep the same chat in GPT #Shorts Same conversation, new model. Lian Wu for Mac: $20 once, no subscription, 14-day full refund. I'm the developer. lianwu.app https://www.youtube.com/embed/gTPruloD9wQ 37 2026-10-29T09:40:29.000Z Marvin Builds https://lianwu.app/videos/gVZYmYk5FtY https://lianwu.app/media/youtube/gVZYmYk5FtY.jpg Claude hit its limit? Keep the same conversation in GPT (Mac app) Switch AI anytime. Keep the conversation. ▶ Try Lian Wu: https://lianwu.app/yt Lian Wu is a Mac app that connects OpenAI, Claude, Gemini, Grok, OpenRouter and more. In this demo, Claude sorts 40 sample customer comments (written for this demo) into five themes. Then I switch to GPT in the same conversation. GPT reads everything so far, checks Claude's grouping and carries on. • $20 one-time purchase. No subscription. • 14-day full refund. • Apple silicon Macs (M1 or later), macOS 14 or later. • Sign in with plans you already pay for (ChatGPT through Codex, Claude through Claude Code) or add your own API key. Each provider bills you directly. I'm the developer, Marvin. If you use more than one AI on a Mac, I'd love your feedback in the comments. Edited with OpenChatCut (open source). https://www.youtube.com/embed/gVZYmYk5FtY 76 2026-10-29T09:34:09.000Z Marvin Builds https://lianwu.app/zh-TW/videos/gVZYmYk5FtY https://lianwu.app/media/youtube/gVZYmYk5FtY.jpg Claude hit its limit? Keep the same conversation in GPT (Mac app) Switch AI anytime. Keep the conversation. ▶ Try Lian Wu: https://lianwu.app/yt Lian Wu is a Mac app that connects OpenAI, Claude, Gemini, Grok, OpenRouter and more. In this demo, Claude sorts 40 sample customer comments (written for this demo) into five themes. Then I switch to GPT in the same conversation. GPT reads everything so far, checks Claude's grouping and carries on. • $20 one-time purchase. No subscription. • 14-day full refund. • Apple silicon Macs (M1 or later), macOS 14 or later. • Sign in with plans you already pay for (ChatGPT through Codex, Claude through Claude Code) or add your own API key. Each provider bills you directly. I'm the developer, Marvin. If you use more than one AI on a Mac, I'd love your feedback in the comments. Edited with OpenChatCut (open source). https://www.youtube.com/embed/gVZYmYk5FtY 76 2026-10-29T09:34:09.000Z Marvin Builds https://lianwu.app/ja/videos/gVZYmYk5FtY https://lianwu.app/media/youtube/gVZYmYk5FtY.jpg Claude hit its limit? Keep the same conversation in GPT (Mac app) Switch AI anytime. Keep the conversation. ▶ Try Lian Wu: https://lianwu.app/yt Lian Wu is a Mac app that connects OpenAI, Claude, Gemini, Grok, OpenRouter and more. In this demo, Claude sorts 40 sample customer comments (written for this demo) into five themes. Then I switch to GPT in the same conversation. GPT reads everything so far, checks Claude's grouping and carries on. • $20 one-time purchase. No subscription. • 14-day full refund. • Apple silicon Macs (M1 or later), macOS 14 or later. • Sign in with plans you already pay for (ChatGPT through Codex, Claude through Claude Code) or add your own API key. Each provider bills you directly. I'm the developer, Marvin. If you use more than one AI on a Mac, I'd love your feedback in the comments. Edited with OpenChatCut (open source). https://www.youtube.com/embed/gVZYmYk5FtY 76 2026-10-29T09:34:09.000Z Marvin Builds https://lianwu.app/ko/videos/gVZYmYk5FtY https://lianwu.app/media/youtube/gVZYmYk5FtY.jpg Claude hit its limit? Keep the same conversation in GPT (Mac app) Switch AI anytime. Keep the conversation. ▶ Try Lian Wu: https://lianwu.app/yt Lian Wu is a Mac app that connects OpenAI, Claude, Gemini, Grok, OpenRouter and more. In this demo, Claude sorts 40 sample customer comments (written for this demo) into five themes. Then I switch to GPT in the same conversation. GPT reads everything so far, checks Claude's grouping and carries on. • $20 one-time purchase. No subscription. • 14-day full refund. • Apple silicon Macs (M1 or later), macOS 14 or later. • Sign in with plans you already pay for (ChatGPT through Codex, Claude through Claude Code) or add your own API key. Each provider bills you directly. I'm the developer, Marvin. If you use more than one AI on a Mac, I'd love your feedback in the comments. Edited with OpenChatCut (open source). https://www.youtube.com/embed/gVZYmYk5FtY 76 2026-10-29T09:34:09.000Z Marvin Builds https://lianwu.app/videos/BKQD57Uxd84 https://lianwu.app/media/youtube/BKQD57Uxd84.jpg GPT-6 Astra: Steer Your Agent Mid-Task #Shorts As of 2026-09-06: GPT-6 Astra supports mid-turn steering, so you can add requirements while the agent is working. In this Short, I explain the behavior and the workflow I want to test: start a desktop build, change the mobile navigation requirement halfway through, and check what Astra keeps. I have not tested this workflow yet. What workflow should I test first? Leave it in the comments. Marvin Builds: real AI-agent workflow tests in public. #Shorts #GPT6 #Astra #AIAgents https://www.youtube.com/embed/BKQD57Uxd84 64 2026-09-06T03:42:58.000Z Marvin Builds https://lianwu.app/zh-TW/videos/BKQD57Uxd84 https://lianwu.app/media/youtube/BKQD57Uxd84.jpg GPT-6 Astra: Steer Your Agent Mid-Task #Shorts As of 2026-09-06: GPT-6 Astra supports mid-turn steering, so you can add requirements while the agent is working. In this Short, I explain the behavior and the workflow I want to test: start a desktop build, change the mobile navigation requirement halfway through, and check what Astra keeps. I have not tested this workflow yet. What workflow should I test first? Leave it in the comments. Marvin Builds: real AI-agent workflow tests in public. #Shorts #GPT6 #Astra #AIAgents https://www.youtube.com/embed/BKQD57Uxd84 64 2026-09-06T03:42:58.000Z Marvin Builds https://lianwu.app/ja/videos/BKQD57Uxd84 https://lianwu.app/media/youtube/BKQD57Uxd84.jpg GPT-6 Astra: Steer Your Agent Mid-Task #Shorts As of 2026-09-06: GPT-6 Astra supports mid-turn steering, so you can add requirements while the agent is working. In this Short, I explain the behavior and the workflow I want to test: start a desktop build, change the mobile navigation requirement halfway through, and check what Astra keeps. I have not tested this workflow yet. What workflow should I test first? Leave it in the comments. Marvin Builds: real AI-agent workflow tests in public. #Shorts #GPT6 #Astra #AIAgents https://www.youtube.com/embed/BKQD57Uxd84 64 2026-09-06T03:42:58.000Z Marvin Builds https://lianwu.app/ko/videos/BKQD57Uxd84 https://lianwu.app/media/youtube/BKQD57Uxd84.jpg GPT-6 Astra: Steer Your Agent Mid-Task #Shorts As of 2026-09-06: GPT-6 Astra supports mid-turn steering, so you can add requirements while the agent is working. In this Short, I explain the behavior and the workflow I want to test: start a desktop build, change the mobile navigation requirement halfway through, and check what Astra keeps. I have not tested this workflow yet. What workflow should I test first? Leave it in the comments. Marvin Builds: real AI-agent workflow tests in public. #Shorts #GPT6 #Astra #AIAgents https://www.youtube.com/embed/BKQD57Uxd84 64 2026-09-06T03:42:58.000Z Marvin Builds https://lianwu.app/videos/qualBRg758I https://lianwu.app/media/youtube/qualBRg758I.jpg I Used Codex to Search Flights From Multiple Starting Cities Try Lian Wu (Mac): https://lianwu.app/yt I found lower fares by changing where the trip started. I used Codex with SerpApi to compare the same London trip from Taipei, Hong Kong, Singapore, and Kuala Lumpur. Codex organized the results into a shortlist, and I checked the strongest options on visible flight pages. The lowest fare wasn't always the trip I would choose. Positioning flights, timing, baggage, and separate-ticket risk could erase the savings. Every option assumes I will fly all ticketed segments. This is external-origin flight search, not hidden-city ticketing or skiplagging. Marvin Builds: real AI-agent workflow tests in public. Flight prices shown were recorded on 2026-07-12 for travel in September 2026 and may change. #Codex #AIAgents #FlightSearch #TravelTech #SerpApi https://www.youtube.com/embed/qualBRg758I 516 2026-07-13T17:00:09.000Z Marvin Builds https://lianwu.app/zh-TW/videos/qualBRg758I https://lianwu.app/media/youtube/qualBRg758I.jpg I Used Codex to Search Flights From Multiple Starting Cities Try Lian Wu (Mac): https://lianwu.app/yt I found lower fares by changing where the trip started. I used Codex with SerpApi to compare the same London trip from Taipei, Hong Kong, Singapore, and Kuala Lumpur. Codex organized the results into a shortlist, and I checked the strongest options on visible flight pages. The lowest fare wasn't always the trip I would choose. Positioning flights, timing, baggage, and separate-ticket risk could erase the savings. Every option assumes I will fly all ticketed segments. This is external-origin flight search, not hidden-city ticketing or skiplagging. Marvin Builds: real AI-agent workflow tests in public. Flight prices shown were recorded on 2026-07-12 for travel in September 2026 and may change. #Codex #AIAgents #FlightSearch #TravelTech #SerpApi https://www.youtube.com/embed/qualBRg758I 516 2026-07-13T17:00:09.000Z Marvin Builds https://lianwu.app/ja/videos/qualBRg758I https://lianwu.app/media/youtube/qualBRg758I.jpg I Used Codex to Search Flights From Multiple Starting Cities Try Lian Wu (Mac): https://lianwu.app/yt I found lower fares by changing where the trip started. I used Codex with SerpApi to compare the same London trip from Taipei, Hong Kong, Singapore, and Kuala Lumpur. Codex organized the results into a shortlist, and I checked the strongest options on visible flight pages. The lowest fare wasn't always the trip I would choose. Positioning flights, timing, baggage, and separate-ticket risk could erase the savings. Every option assumes I will fly all ticketed segments. This is external-origin flight search, not hidden-city ticketing or skiplagging. Marvin Builds: real AI-agent workflow tests in public. Flight prices shown were recorded on 2026-07-12 for travel in September 2026 and may change. #Codex #AIAgents #FlightSearch #TravelTech #SerpApi https://www.youtube.com/embed/qualBRg758I 516 2026-07-13T17:00:09.000Z Marvin Builds https://lianwu.app/ko/videos/qualBRg758I https://lianwu.app/media/youtube/qualBRg758I.jpg I Used Codex to Search Flights From Multiple Starting Cities Try Lian Wu (Mac): https://lianwu.app/yt I found lower fares by changing where the trip started. I used Codex with SerpApi to compare the same London trip from Taipei, Hong Kong, Singapore, and Kuala Lumpur. Codex organized the results into a shortlist, and I checked the strongest options on visible flight pages. The lowest fare wasn't always the trip I would choose. Positioning flights, timing, baggage, and separate-ticket risk could erase the savings. Every option assumes I will fly all ticketed segments. This is external-origin flight search, not hidden-city ticketing or skiplagging. Marvin Builds: real AI-agent workflow tests in public. Flight prices shown were recorded on 2026-07-12 for travel in September 2026 and may change. #Codex #AIAgents #FlightSearch #TravelTech #SerpApi https://www.youtube.com/embed/qualBRg758I 516 2026-07-13T17:00:09.000Z Marvin Builds https://lianwu.app/videos/oVEtlueD2FE https://lianwu.app/media/youtube/oVEtlueD2FE.jpg Apple Just Made AI Costs Visible #Shorts As of 2026-06-27: Apple raised prices on Macs and iPads, and the reason is bigger than Apple. Apple points to AI data centers pushing up memory and storage chip costs. Micron is saying the other side: demand is bigger than supply, and memory is becoming strategic. AI infrastructure costs are starting to reach consumer hardware. If you were already planning to buy a MacBook Pro, Mac mini, or higher-memory Mac, I would not wait too long just hoping prices go down. Do not panic-buy, but if you need the machine, move faster. Marvin Builds: real AI-agent workflow tests in public. #Shorts #Apple #MacBookPro #MacMini #AI #Micron #Memory https://www.youtube.com/embed/oVEtlueD2FE 119 2026-06-27T15:05:56.000Z Marvin Builds https://lianwu.app/zh-TW/videos/oVEtlueD2FE https://lianwu.app/media/youtube/oVEtlueD2FE.jpg Apple Just Made AI Costs Visible #Shorts As of 2026-06-27: Apple raised prices on Macs and iPads, and the reason is bigger than Apple. Apple points to AI data centers pushing up memory and storage chip costs. Micron is saying the other side: demand is bigger than supply, and memory is becoming strategic. AI infrastructure costs are starting to reach consumer hardware. If you were already planning to buy a MacBook Pro, Mac mini, or higher-memory Mac, I would not wait too long just hoping prices go down. Do not panic-buy, but if you need the machine, move faster. Marvin Builds: real AI-agent workflow tests in public. #Shorts #Apple #MacBookPro #MacMini #AI #Micron #Memory https://www.youtube.com/embed/oVEtlueD2FE 119 2026-06-27T15:05:56.000Z Marvin Builds https://lianwu.app/ja/videos/oVEtlueD2FE https://lianwu.app/media/youtube/oVEtlueD2FE.jpg Apple Just Made AI Costs Visible #Shorts As of 2026-06-27: Apple raised prices on Macs and iPads, and the reason is bigger than Apple. Apple points to AI data centers pushing up memory and storage chip costs. Micron is saying the other side: demand is bigger than supply, and memory is becoming strategic. AI infrastructure costs are starting to reach consumer hardware. If you were already planning to buy a MacBook Pro, Mac mini, or higher-memory Mac, I would not wait too long just hoping prices go down. Do not panic-buy, but if you need the machine, move faster. Marvin Builds: real AI-agent workflow tests in public. #Shorts #Apple #MacBookPro #MacMini #AI #Micron #Memory https://www.youtube.com/embed/oVEtlueD2FE 119 2026-06-27T15:05:56.000Z Marvin Builds https://lianwu.app/ko/videos/oVEtlueD2FE https://lianwu.app/media/youtube/oVEtlueD2FE.jpg Apple Just Made AI Costs Visible #Shorts As of 2026-06-27: Apple raised prices on Macs and iPads, and the reason is bigger than Apple. Apple points to AI data centers pushing up memory and storage chip costs. Micron is saying the other side: demand is bigger than supply, and memory is becoming strategic. AI infrastructure costs are starting to reach consumer hardware. If you were already planning to buy a MacBook Pro, Mac mini, or higher-memory Mac, I would not wait too long just hoping prices go down. Do not panic-buy, but if you need the machine, move faster. Marvin Builds: real AI-agent workflow tests in public. #Shorts #Apple #MacBookPro #MacMini #AI #Micron #Memory https://www.youtube.com/embed/oVEtlueD2FE 119 2026-06-27T15:05:56.000Z Marvin Builds https://lianwu.app/videos/uTehpid3tRM https://lianwu.app/media/youtube/uTehpid3tRM.jpg Claude Tag: AI Agents Join Team Chat #Shorts Anthropic introduced Claude Tag, a Slack beta that lets teams mention @Claude inside a channel. My takeaway: the real shift is not Claude inside Slack. It is AI agents moving from private chat windows into shared team spaces. When agents work inside team threads, access control, memory, spend limits, and audit logs become product-critical. Source: https://www.anthropic.com/news/introducing-claude-tag Marvin Builds: real AI-agent workflow tests in public. #Shorts #Claude #AIAgents #Anthropic https://www.youtube.com/embed/uTehpid3tRM 69 2026-06-24T15:52:56.000Z Marvin Builds https://lianwu.app/zh-TW/videos/uTehpid3tRM https://lianwu.app/media/youtube/uTehpid3tRM.jpg Claude Tag: AI Agents Join Team Chat #Shorts Anthropic introduced Claude Tag, a Slack beta that lets teams mention @Claude inside a channel. My takeaway: the real shift is not Claude inside Slack. It is AI agents moving from private chat windows into shared team spaces. When agents work inside team threads, access control, memory, spend limits, and audit logs become product-critical. Source: https://www.anthropic.com/news/introducing-claude-tag Marvin Builds: real AI-agent workflow tests in public. #Shorts #Claude #AIAgents #Anthropic https://www.youtube.com/embed/uTehpid3tRM 69 2026-06-24T15:52:56.000Z Marvin Builds https://lianwu.app/ja/videos/uTehpid3tRM https://lianwu.app/media/youtube/uTehpid3tRM.jpg Claude Tag: AI Agents Join Team Chat #Shorts Anthropic introduced Claude Tag, a Slack beta that lets teams mention @Claude inside a channel. My takeaway: the real shift is not Claude inside Slack. It is AI agents moving from private chat windows into shared team spaces. When agents work inside team threads, access control, memory, spend limits, and audit logs become product-critical. Source: https://www.anthropic.com/news/introducing-claude-tag Marvin Builds: real AI-agent workflow tests in public. #Shorts #Claude #AIAgents #Anthropic https://www.youtube.com/embed/uTehpid3tRM 69 2026-06-24T15:52:56.000Z Marvin Builds https://lianwu.app/ko/videos/uTehpid3tRM https://lianwu.app/media/youtube/uTehpid3tRM.jpg Claude Tag: AI Agents Join Team Chat #Shorts Anthropic introduced Claude Tag, a Slack beta that lets teams mention @Claude inside a channel. My takeaway: the real shift is not Claude inside Slack. It is AI agents moving from private chat windows into shared team spaces. When agents work inside team threads, access control, memory, spend limits, and audit logs become product-critical. Source: https://www.anthropic.com/news/introducing-claude-tag Marvin Builds: real AI-agent workflow tests in public. #Shorts #Claude #AIAgents #Anthropic https://www.youtube.com/embed/uTehpid3tRM 69 2026-06-24T15:52:56.000Z Marvin Builds https://lianwu.app/videos/5YAfCv0Cy3E https://lianwu.app/media/youtube/5YAfCv0Cy3E.jpg I Gave Codex One Image to Build a Web Game Demo Try Lian Wu (Mac): https://lianwu.app/yt In the Codex era, building a demo should start with image generation. Not because the image is final art. Because a strong image gives an agent a clearer target than a long, flat spec. It shows layout, mood, UI density, game feel, and the kind of details that are hard to describe in text. In this video, I start with a fantasy brick breaker concept image. Then I give that image to Codex and ask it to turn the direction into a playable frontend demo. The goal is not to ship a complete game. The goal is to show a faster way to start: image first, agent build, human review, then refine. Play the demo: https://fantasy-break.marvinswift.com What I cover: 1. Generating the concept image 2. Using the image as a visual brief 3. Letting Codex build the first version 4. Reviewing what worked and what felt off 5. Improving the UI and gameplay feel 6. Showing the live demo Next, I will make a separate video about the Cloudflare deployment and custom domain setup. Marvin Builds is where I build and test real AI-agent workflows in public. https://www.youtube.com/embed/5YAfCv0Cy3E 303 2026-06-22T15:39:04.000Z Marvin Builds https://lianwu.app/zh-TW/videos/5YAfCv0Cy3E https://lianwu.app/media/youtube/5YAfCv0Cy3E.jpg I Gave Codex One Image to Build a Web Game Demo Try Lian Wu (Mac): https://lianwu.app/yt In the Codex era, building a demo should start with image generation. Not because the image is final art. Because a strong image gives an agent a clearer target than a long, flat spec. It shows layout, mood, UI density, game feel, and the kind of details that are hard to describe in text. In this video, I start with a fantasy brick breaker concept image. Then I give that image to Codex and ask it to turn the direction into a playable frontend demo. The goal is not to ship a complete game. The goal is to show a faster way to start: image first, agent build, human review, then refine. Play the demo: https://fantasy-break.marvinswift.com What I cover: 1. Generating the concept image 2. Using the image as a visual brief 3. Letting Codex build the first version 4. Reviewing what worked and what felt off 5. Improving the UI and gameplay feel 6. Showing the live demo Next, I will make a separate video about the Cloudflare deployment and custom domain setup. Marvin Builds is where I build and test real AI-agent workflows in public. https://www.youtube.com/embed/5YAfCv0Cy3E 303 2026-06-22T15:39:04.000Z Marvin Builds https://lianwu.app/ja/videos/5YAfCv0Cy3E https://lianwu.app/media/youtube/5YAfCv0Cy3E.jpg I Gave Codex One Image to Build a Web Game Demo Try Lian Wu (Mac): https://lianwu.app/yt In the Codex era, building a demo should start with image generation. Not because the image is final art. Because a strong image gives an agent a clearer target than a long, flat spec. It shows layout, mood, UI density, game feel, and the kind of details that are hard to describe in text. In this video, I start with a fantasy brick breaker concept image. Then I give that image to Codex and ask it to turn the direction into a playable frontend demo. The goal is not to ship a complete game. The goal is to show a faster way to start: image first, agent build, human review, then refine. Play the demo: https://fantasy-break.marvinswift.com What I cover: 1. Generating the concept image 2. Using the image as a visual brief 3. Letting Codex build the first version 4. Reviewing what worked and what felt off 5. Improving the UI and gameplay feel 6. Showing the live demo Next, I will make a separate video about the Cloudflare deployment and custom domain setup. Marvin Builds is where I build and test real AI-agent workflows in public. https://www.youtube.com/embed/5YAfCv0Cy3E 303 2026-06-22T15:39:04.000Z Marvin Builds https://lianwu.app/ko/videos/5YAfCv0Cy3E https://lianwu.app/media/youtube/5YAfCv0Cy3E.jpg I Gave Codex One Image to Build a Web Game Demo Try Lian Wu (Mac): https://lianwu.app/yt In the Codex era, building a demo should start with image generation. Not because the image is final art. Because a strong image gives an agent a clearer target than a long, flat spec. It shows layout, mood, UI density, game feel, and the kind of details that are hard to describe in text. In this video, I start with a fantasy brick breaker concept image. Then I give that image to Codex and ask it to turn the direction into a playable frontend demo. The goal is not to ship a complete game. The goal is to show a faster way to start: image first, agent build, human review, then refine. Play the demo: https://fantasy-break.marvinswift.com What I cover: 1. Generating the concept image 2. Using the image as a visual brief 3. Letting Codex build the first version 4. Reviewing what worked and what felt off 5. Improving the UI and gameplay feel 6. Showing the live demo Next, I will make a separate video about the Cloudflare deployment and custom domain setup. Marvin Builds is where I build and test real AI-agent workflows in public. https://www.youtube.com/embed/5YAfCv0Cy3E 303 2026-06-22T15:39:04.000Z Marvin Builds https://lianwu.app/videos/9JTs6EvMyBQ https://lianwu.app/media/youtube/9JTs6EvMyBQ.jpg claude -p Still Works With Your Subscription #Shorts Anthropic paused its planned Claude Agent SDK credit change. For now, Agent SDK usage, claude -p, and third-party apps built on the Agent SDK continue to work with Claude subscriptions the same way as before. The practical takeaway: if you rely on claude -p scripts today, they are not cut off by this credit change yet. Watch the next Anthropic notice before changing your workflow. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AIagents #DeveloperTools https://www.youtube.com/embed/9JTs6EvMyBQ 57 2026-06-16T11:56:19.000Z Marvin Builds https://lianwu.app/zh-TW/videos/9JTs6EvMyBQ https://lianwu.app/media/youtube/9JTs6EvMyBQ.jpg claude -p Still Works With Your Subscription #Shorts Anthropic paused its planned Claude Agent SDK credit change. For now, Agent SDK usage, claude -p, and third-party apps built on the Agent SDK continue to work with Claude subscriptions the same way as before. The practical takeaway: if you rely on claude -p scripts today, they are not cut off by this credit change yet. Watch the next Anthropic notice before changing your workflow. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AIagents #DeveloperTools https://www.youtube.com/embed/9JTs6EvMyBQ 57 2026-06-16T11:56:19.000Z Marvin Builds https://lianwu.app/ja/videos/9JTs6EvMyBQ https://lianwu.app/media/youtube/9JTs6EvMyBQ.jpg claude -p Still Works With Your Subscription #Shorts Anthropic paused its planned Claude Agent SDK credit change. For now, Agent SDK usage, claude -p, and third-party apps built on the Agent SDK continue to work with Claude subscriptions the same way as before. The practical takeaway: if you rely on claude -p scripts today, they are not cut off by this credit change yet. Watch the next Anthropic notice before changing your workflow. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AIagents #DeveloperTools https://www.youtube.com/embed/9JTs6EvMyBQ 57 2026-06-16T11:56:19.000Z Marvin Builds https://lianwu.app/ko/videos/9JTs6EvMyBQ https://lianwu.app/media/youtube/9JTs6EvMyBQ.jpg claude -p Still Works With Your Subscription #Shorts Anthropic paused its planned Claude Agent SDK credit change. For now, Agent SDK usage, claude -p, and third-party apps built on the Agent SDK continue to work with Claude subscriptions the same way as before. The practical takeaway: if you rely on claude -p scripts today, they are not cut off by this credit change yet. Watch the next Anthropic notice before changing your workflow. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AIagents #DeveloperTools https://www.youtube.com/embed/9JTs6EvMyBQ 57 2026-06-16T11:56:19.000Z Marvin Builds https://lianwu.app/videos/BOxwiCBtd9w https://lianwu.app/media/youtube/BOxwiCBtd9w.jpg Claude Fable 5 Access Changed Overnight #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was paused after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/BOxwiCBtd9w 48 2026-06-14T00:00:01.000Z Marvin Builds https://lianwu.app/zh-TW/videos/BOxwiCBtd9w https://lianwu.app/media/youtube/BOxwiCBtd9w.jpg Claude Fable 5 Access Changed Overnight #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was paused after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/BOxwiCBtd9w 48 2026-06-14T00:00:01.000Z Marvin Builds https://lianwu.app/ja/videos/BOxwiCBtd9w https://lianwu.app/media/youtube/BOxwiCBtd9w.jpg Claude Fable 5 Access Changed Overnight #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was paused after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/BOxwiCBtd9w 48 2026-06-14T00:00:01.000Z Marvin Builds https://lianwu.app/ko/videos/BOxwiCBtd9w https://lianwu.app/media/youtube/BOxwiCBtd9w.jpg Claude Fable 5 Access Changed Overnight #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was paused after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/BOxwiCBtd9w 48 2026-06-14T00:00:01.000Z Marvin Builds https://lianwu.app/videos/Xc-K0AjcyH0 https://lianwu.app/media/youtube/Xc-K0AjcyH0.jpg Claude Fable 5 Access Was Turned Off #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was turned off after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/Xc-K0AjcyH0 48 2026-06-13T08:05:53.000Z Marvin Builds https://lianwu.app/zh-TW/videos/Xc-K0AjcyH0 https://lianwu.app/media/youtube/Xc-K0AjcyH0.jpg Claude Fable 5 Access Was Turned Off #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was turned off after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/Xc-K0AjcyH0 48 2026-06-13T08:05:53.000Z Marvin Builds https://lianwu.app/ja/videos/Xc-K0AjcyH0 https://lianwu.app/media/youtube/Xc-K0AjcyH0.jpg Claude Fable 5 Access Was Turned Off #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was turned off after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/Xc-K0AjcyH0 48 2026-06-13T08:05:53.000Z Marvin Builds https://lianwu.app/ko/videos/Xc-K0AjcyH0 https://lianwu.app/media/youtube/Xc-K0AjcyH0.jpg Claude Fable 5 Access Was Turned Off #Shorts Anthropic says access to Claude Fable 5 and Mythos 5 was turned off after a broad U.S. government order. The block affected the Claude website, API, Claude Code, and Claude Cowork, while other Claude models still worked. This Short is about what that means for builders: AI product access can change overnight when policy changes. Marvin Builds: real AI-agent workflows in public. #Shorts #ClaudeAI #Anthropic #AI #AITools https://www.youtube.com/embed/Xc-K0AjcyH0 48 2026-06-13T08:05:53.000Z Marvin Builds https://lianwu.app/videos/EA-z_MSRIAw https://lianwu.app/media/youtube/EA-z_MSRIAw.jpg Claude Fable 5 and Mythos 5: The June 22 Window #Shorts Anthropic launched Claude Fable 5 and Claude Mythos 5. Fable 5 is included on Pro, Max, Team, and seat-based Enterprise through June 22. On June 23, usage credits are required. Source: https://www.anthropic.com/news/claude-fable-5-mythos-5 #Shorts #Claude #Anthropic #AI https://www.youtube.com/embed/EA-z_MSRIAw 54 2026-06-10T14:49:40.000Z Marvin Builds https://lianwu.app/zh-TW/videos/EA-z_MSRIAw https://lianwu.app/media/youtube/EA-z_MSRIAw.jpg Claude Fable 5 and Mythos 5: The June 22 Window #Shorts Anthropic launched Claude Fable 5 and Claude Mythos 5. Fable 5 is included on Pro, Max, Team, and seat-based Enterprise through June 22. On June 23, usage credits are required. Source: https://www.anthropic.com/news/claude-fable-5-mythos-5 #Shorts #Claude #Anthropic #AI https://www.youtube.com/embed/EA-z_MSRIAw 54 2026-06-10T14:49:40.000Z Marvin Builds https://lianwu.app/ja/videos/EA-z_MSRIAw https://lianwu.app/media/youtube/EA-z_MSRIAw.jpg Claude Fable 5 and Mythos 5: The June 22 Window #Shorts Anthropic launched Claude Fable 5 and Claude Mythos 5. Fable 5 is included on Pro, Max, Team, and seat-based Enterprise through June 22. On June 23, usage credits are required. Source: https://www.anthropic.com/news/claude-fable-5-mythos-5 #Shorts #Claude #Anthropic #AI https://www.youtube.com/embed/EA-z_MSRIAw 54 2026-06-10T14:49:40.000Z Marvin Builds https://lianwu.app/ko/videos/EA-z_MSRIAw https://lianwu.app/media/youtube/EA-z_MSRIAw.jpg Claude Fable 5 and Mythos 5: The June 22 Window #Shorts Anthropic launched Claude Fable 5 and Claude Mythos 5. Fable 5 is included on Pro, Max, Team, and seat-based Enterprise through June 22. On June 23, usage credits are required. Source: https://www.anthropic.com/news/claude-fable-5-mythos-5 #Shorts #Claude #Anthropic #AI https://www.youtube.com/embed/EA-z_MSRIAw 54 2026-06-10T14:49:40.000Z Marvin Builds https://lianwu.app/videos/Cj8D5jgf1zc https://lianwu.app/media/youtube/Cj8D5jgf1zc.jpg Google cut engineering education? #Shorts One X post started this thread: Gergely Orosz said everyone on Google's Engineering Education team was laid off, and that he checked with people inside Google. This is not a Google announcement, and I did not find broad tech-media confirmation when preparing this Short. The first visible spread I saw was the original X post being reposted to Hacker News. The signal still matters because Engineering Education is the kind of team that teaches engineers internally, documents systems, and turns hard-won knowledge into training material. Careful wording: reported, not announced. Marvin Builds: AI work, carefully verified. #Shorts #Google #EngineeringEducation #TechLayoffs #AI https://www.youtube.com/embed/Cj8D5jgf1zc 54 2026-06-06T16:31:30.000Z Marvin Builds https://lianwu.app/zh-TW/videos/Cj8D5jgf1zc https://lianwu.app/media/youtube/Cj8D5jgf1zc.jpg Google cut engineering education? #Shorts One X post started this thread: Gergely Orosz said everyone on Google's Engineering Education team was laid off, and that he checked with people inside Google. This is not a Google announcement, and I did not find broad tech-media confirmation when preparing this Short. The first visible spread I saw was the original X post being reposted to Hacker News. The signal still matters because Engineering Education is the kind of team that teaches engineers internally, documents systems, and turns hard-won knowledge into training material. Careful wording: reported, not announced. Marvin Builds: AI work, carefully verified. #Shorts #Google #EngineeringEducation #TechLayoffs #AI https://www.youtube.com/embed/Cj8D5jgf1zc 54 2026-06-06T16:31:30.000Z Marvin Builds https://lianwu.app/ja/videos/Cj8D5jgf1zc https://lianwu.app/media/youtube/Cj8D5jgf1zc.jpg Google cut engineering education? #Shorts One X post started this thread: Gergely Orosz said everyone on Google's Engineering Education team was laid off, and that he checked with people inside Google. This is not a Google announcement, and I did not find broad tech-media confirmation when preparing this Short. The first visible spread I saw was the original X post being reposted to Hacker News. The signal still matters because Engineering Education is the kind of team that teaches engineers internally, documents systems, and turns hard-won knowledge into training material. Careful wording: reported, not announced. Marvin Builds: AI work, carefully verified. #Shorts #Google #EngineeringEducation #TechLayoffs #AI https://www.youtube.com/embed/Cj8D5jgf1zc 54 2026-06-06T16:31:30.000Z Marvin Builds https://lianwu.app/ko/videos/Cj8D5jgf1zc https://lianwu.app/media/youtube/Cj8D5jgf1zc.jpg Google cut engineering education? #Shorts One X post started this thread: Gergely Orosz said everyone on Google's Engineering Education team was laid off, and that he checked with people inside Google. This is not a Google announcement, and I did not find broad tech-media confirmation when preparing this Short. The first visible spread I saw was the original X post being reposted to Hacker News. The signal still matters because Engineering Education is the kind of team that teaches engineers internally, documents systems, and turns hard-won knowledge into training material. Careful wording: reported, not announced. Marvin Builds: AI work, carefully verified. #Shorts #Google #EngineeringEducation #TechLayoffs #AI https://www.youtube.com/embed/Cj8D5jgf1zc 54 2026-06-06T16:31:30.000Z Marvin Builds https://lianwu.app/videos/4sES7QNYuPc https://lianwu.app/media/youtube/4sES7QNYuPc.jpg Google cut engineering education? #Shorts A careful read of Gergely Orosz's single-source report about Google's Engineering Education team, the first visible spread on X and Hacker News, and why the signal matters. Not a Google official announcement. #Shorts https://www.youtube.com/embed/4sES7QNYuPc 54 2026-06-06T16:31:24.000Z Marvin Builds https://lianwu.app/zh-TW/videos/4sES7QNYuPc https://lianwu.app/media/youtube/4sES7QNYuPc.jpg Google cut engineering education? #Shorts A careful read of Gergely Orosz's single-source report about Google's Engineering Education team, the first visible spread on X and Hacker News, and why the signal matters. Not a Google official announcement. #Shorts https://www.youtube.com/embed/4sES7QNYuPc 54 2026-06-06T16:31:24.000Z Marvin Builds https://lianwu.app/ja/videos/4sES7QNYuPc https://lianwu.app/media/youtube/4sES7QNYuPc.jpg Google cut engineering education? #Shorts A careful read of Gergely Orosz's single-source report about Google's Engineering Education team, the first visible spread on X and Hacker News, and why the signal matters. Not a Google official announcement. #Shorts https://www.youtube.com/embed/4sES7QNYuPc 54 2026-06-06T16:31:24.000Z Marvin Builds https://lianwu.app/ko/videos/4sES7QNYuPc https://lianwu.app/media/youtube/4sES7QNYuPc.jpg Google cut engineering education? #Shorts A careful read of Gergely Orosz's single-source report about Google's Engineering Education team, the first visible spread on X and Hacker News, and why the signal matters. Not a Google official announcement. #Shorts https://www.youtube.com/embed/4sES7QNYuPc 54 2026-06-06T16:31:24.000Z Marvin Builds https://lianwu.app/videos/CcRcaeP1mw8 https://lianwu.app/media/youtube/CcRcaeP1mw8.jpg Move AI Memory Between Agents: Codex to Claude Try Lian Wu (Mac): https://lianwu.app/yt In this Marvin Builds video, I demo a simple but useful AI-agent workflow: take reviewed memory/context from one AI agent, bring it into another agent, and verify what the receiving agent learned. The point is not to blindly sync every private conversation. The point is a clean handoff: extract, review, import, verify. Prompt used in the demo: Export all of my stored memories and any context you've learned about me from past conversations. Preserve my words verbatim where possible, especially for instructions and preferences. ## Categories (output in this order): 1. **Instructions**: Rules I've explicitly asked you to follow going forward — tone, format, style, "always do X", "never do Y", and corrections to your behavior. Only include rules from stored memories, not from conversations. 2. **Identity**: Name, age, location, education, family, relationships, languages, and personal interests. 3. **Career**: Current and past roles, companies, and general skill areas. 4. **Projects**: Projects I meaningfully built or committed to. Ideally ONE entry per project. Include what it does, current status, and any key decisions. Use the project name or a short descriptor as the first words of the entry. 5. **Preferences**: Opinions, tastes, and working-style preferences that apply broadly. ## Format: Use section headers for each category. Within each category, list one entry per line, sorted by oldest date first. Format each line as: [YYYY-MM-DD] - Entry content here. If no date is known, use [unknown] instead. ## Output: - Wrap the entire export in a single code block for easy copying. - After the code block, state whether this is the complete set or if more remain. Marvin Builds: I build and test real AI-agent workflows in public. https://www.youtube.com/embed/CcRcaeP1mw8 104 2026-06-06T10:55:48.000Z Marvin Builds https://lianwu.app/zh-TW/videos/CcRcaeP1mw8 https://lianwu.app/media/youtube/CcRcaeP1mw8.jpg Move AI Memory Between Agents: Codex to Claude Try Lian Wu (Mac): https://lianwu.app/yt In this Marvin Builds video, I demo a simple but useful AI-agent workflow: take reviewed memory/context from one AI agent, bring it into another agent, and verify what the receiving agent learned. The point is not to blindly sync every private conversation. The point is a clean handoff: extract, review, import, verify. Prompt used in the demo: Export all of my stored memories and any context you've learned about me from past conversations. Preserve my words verbatim where possible, especially for instructions and preferences. ## Categories (output in this order): 1. **Instructions**: Rules I've explicitly asked you to follow going forward — tone, format, style, "always do X", "never do Y", and corrections to your behavior. Only include rules from stored memories, not from conversations. 2. **Identity**: Name, age, location, education, family, relationships, languages, and personal interests. 3. **Career**: Current and past roles, companies, and general skill areas. 4. **Projects**: Projects I meaningfully built or committed to. Ideally ONE entry per project. Include what it does, current status, and any key decisions. Use the project name or a short descriptor as the first words of the entry. 5. **Preferences**: Opinions, tastes, and working-style preferences that apply broadly. ## Format: Use section headers for each category. Within each category, list one entry per line, sorted by oldest date first. Format each line as: [YYYY-MM-DD] - Entry content here. If no date is known, use [unknown] instead. ## Output: - Wrap the entire export in a single code block for easy copying. - After the code block, state whether this is the complete set or if more remain. Marvin Builds: I build and test real AI-agent workflows in public. https://www.youtube.com/embed/CcRcaeP1mw8 104 2026-06-06T10:55:48.000Z Marvin Builds https://lianwu.app/ja/videos/CcRcaeP1mw8 https://lianwu.app/media/youtube/CcRcaeP1mw8.jpg Move AI Memory Between Agents: Codex to Claude Try Lian Wu (Mac): https://lianwu.app/yt In this Marvin Builds video, I demo a simple but useful AI-agent workflow: take reviewed memory/context from one AI agent, bring it into another agent, and verify what the receiving agent learned. The point is not to blindly sync every private conversation. The point is a clean handoff: extract, review, import, verify. Prompt used in the demo: Export all of my stored memories and any context you've learned about me from past conversations. Preserve my words verbatim where possible, especially for instructions and preferences. ## Categories (output in this order): 1. **Instructions**: Rules I've explicitly asked you to follow going forward — tone, format, style, "always do X", "never do Y", and corrections to your behavior. Only include rules from stored memories, not from conversations. 2. **Identity**: Name, age, location, education, family, relationships, languages, and personal interests. 3. **Career**: Current and past roles, companies, and general skill areas. 4. **Projects**: Projects I meaningfully built or committed to. Ideally ONE entry per project. Include what it does, current status, and any key decisions. Use the project name or a short descriptor as the first words of the entry. 5. **Preferences**: Opinions, tastes, and working-style preferences that apply broadly. ## Format: Use section headers for each category. Within each category, list one entry per line, sorted by oldest date first. Format each line as: [YYYY-MM-DD] - Entry content here. If no date is known, use [unknown] instead. ## Output: - Wrap the entire export in a single code block for easy copying. - After the code block, state whether this is the complete set or if more remain. Marvin Builds: I build and test real AI-agent workflows in public. https://www.youtube.com/embed/CcRcaeP1mw8 104 2026-06-06T10:55:48.000Z Marvin Builds https://lianwu.app/ko/videos/CcRcaeP1mw8 https://lianwu.app/media/youtube/CcRcaeP1mw8.jpg Move AI Memory Between Agents: Codex to Claude Try Lian Wu (Mac): https://lianwu.app/yt In this Marvin Builds video, I demo a simple but useful AI-agent workflow: take reviewed memory/context from one AI agent, bring it into another agent, and verify what the receiving agent learned. The point is not to blindly sync every private conversation. The point is a clean handoff: extract, review, import, verify. Prompt used in the demo: Export all of my stored memories and any context you've learned about me from past conversations. Preserve my words verbatim where possible, especially for instructions and preferences. ## Categories (output in this order): 1. **Instructions**: Rules I've explicitly asked you to follow going forward — tone, format, style, "always do X", "never do Y", and corrections to your behavior. Only include rules from stored memories, not from conversations. 2. **Identity**: Name, age, location, education, family, relationships, languages, and personal interests. 3. **Career**: Current and past roles, companies, and general skill areas. 4. **Projects**: Projects I meaningfully built or committed to. Ideally ONE entry per project. Include what it does, current status, and any key decisions. Use the project name or a short descriptor as the first words of the entry. 5. **Preferences**: Opinions, tastes, and working-style preferences that apply broadly. ## Format: Use section headers for each category. Within each category, list one entry per line, sorted by oldest date first. Format each line as: [YYYY-MM-DD] - Entry content here. If no date is known, use [unknown] instead. ## Output: - Wrap the entire export in a single code block for easy copying. - After the code block, state whether this is the complete set or if more remain. Marvin Builds: I build and test real AI-agent workflows in public. https://www.youtube.com/embed/CcRcaeP1mw8 104 2026-06-06T10:55:48.000Z Marvin Builds https://lianwu.app/videos/StjQL2zM_ug https://lianwu.app/media/youtube/StjQL2zM_ug.jpg Claude Code /effort: Max to Ultracode #Shorts Claude Code has a small but useful control: type /effort and you can tune the balance between faster responses and deeper thinking. In this Marvin Builds Short, I show the newer /effort UI moving from low, medium, high, xhigh, and max, then further into Ultracode. The point is not the visual effect by itself. The point is the workflow signal: quick edits can stay fast, while hard architecture, deep debugging, or multi-agent work can ask for more thinking. If you are building with AI agents, follow Marvin Builds. I test real AI-agent workflows in public. #Shorts #ClaudeCode #AIAgents #MarvinBuilds https://www.youtube.com/embed/StjQL2zM_ug 63 2026-06-02T08:48:57.000Z Marvin Builds https://lianwu.app/zh-TW/videos/StjQL2zM_ug https://lianwu.app/media/youtube/StjQL2zM_ug.jpg Claude Code /effort: Max to Ultracode #Shorts Claude Code has a small but useful control: type /effort and you can tune the balance between faster responses and deeper thinking. In this Marvin Builds Short, I show the newer /effort UI moving from low, medium, high, xhigh, and max, then further into Ultracode. The point is not the visual effect by itself. The point is the workflow signal: quick edits can stay fast, while hard architecture, deep debugging, or multi-agent work can ask for more thinking. If you are building with AI agents, follow Marvin Builds. I test real AI-agent workflows in public. #Shorts #ClaudeCode #AIAgents #MarvinBuilds https://www.youtube.com/embed/StjQL2zM_ug 63 2026-06-02T08:48:57.000Z Marvin Builds https://lianwu.app/ja/videos/StjQL2zM_ug https://lianwu.app/media/youtube/StjQL2zM_ug.jpg Claude Code /effort: Max to Ultracode #Shorts Claude Code has a small but useful control: type /effort and you can tune the balance between faster responses and deeper thinking. In this Marvin Builds Short, I show the newer /effort UI moving from low, medium, high, xhigh, and max, then further into Ultracode. The point is not the visual effect by itself. The point is the workflow signal: quick edits can stay fast, while hard architecture, deep debugging, or multi-agent work can ask for more thinking. If you are building with AI agents, follow Marvin Builds. I test real AI-agent workflows in public. #Shorts #ClaudeCode #AIAgents #MarvinBuilds https://www.youtube.com/embed/StjQL2zM_ug 63 2026-06-02T08:48:57.000Z Marvin Builds https://lianwu.app/ko/videos/StjQL2zM_ug https://lianwu.app/media/youtube/StjQL2zM_ug.jpg Claude Code /effort: Max to Ultracode #Shorts Claude Code has a small but useful control: type /effort and you can tune the balance between faster responses and deeper thinking. In this Marvin Builds Short, I show the newer /effort UI moving from low, medium, high, xhigh, and max, then further into Ultracode. The point is not the visual effect by itself. The point is the workflow signal: quick edits can stay fast, while hard architecture, deep debugging, or multi-agent work can ask for more thinking. If you are building with AI agents, follow Marvin Builds. I test real AI-agent workflows in public. #Shorts #ClaudeCode #AIAgents #MarvinBuilds https://www.youtube.com/embed/StjQL2zM_ug 63 2026-06-02T08:48:57.000Z Marvin Builds https://lianwu.app/videos/oiXZbYk_mUc https://lianwu.app/media/youtube/oiXZbYk_mUc.jpg Codex Automations for Baseball Research #Shorts Codex Automations can prepare recurring research context before you start working. In this Marvin Builds Short, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, Level, Position, Age, Scout Focus.] Wa https://www.youtube.com/embed/oiXZbYk_mUc 59 2026-05-31T14:51:16.000Z Marvin Builds https://lianwu.app/zh-TW/videos/oiXZbYk_mUc https://lianwu.app/media/youtube/oiXZbYk_mUc.jpg Codex Automations for Baseball Research #Shorts Codex Automations can prepare recurring research context before you start working. In this Marvin Builds Short, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, Level, Position, Age, Scout Focus.] Wa https://www.youtube.com/embed/oiXZbYk_mUc 59 2026-05-31T14:51:16.000Z Marvin Builds https://lianwu.app/ja/videos/oiXZbYk_mUc https://lianwu.app/media/youtube/oiXZbYk_mUc.jpg Codex Automations for Baseball Research #Shorts Codex Automations can prepare recurring research context before you start working. In this Marvin Builds Short, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, Level, Position, Age, Scout Focus.] Wa https://www.youtube.com/embed/oiXZbYk_mUc 59 2026-05-31T14:51:16.000Z Marvin Builds https://lianwu.app/ko/videos/oiXZbYk_mUc https://lianwu.app/media/youtube/oiXZbYk_mUc.jpg Codex Automations for Baseball Research #Shorts Codex Automations can prepare recurring research context before you start working. In this Marvin Builds Short, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, Level, Position, Age, Scout Focus.] Wa https://www.youtube.com/embed/oiXZbYk_mUc 59 2026-05-31T14:51:16.000Z Marvin Builds https://lianwu.app/videos/WVxG1Ekkf_I https://lianwu.app/media/youtube/WVxG1Ekkf_I.jpg Codex Automations for Baseball Research Try Lian Wu (Mac): https://lianwu.app/yt Codex Automations can prepare recurring research context before you start working. In this Marvin Builds demo, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, https://www.youtube.com/embed/WVxG1Ekkf_I 122 2026-05-30T06:25:09.000Z Marvin Builds https://lianwu.app/zh-TW/videos/WVxG1Ekkf_I https://lianwu.app/media/youtube/WVxG1Ekkf_I.jpg Codex Automations for Baseball Research Try Lian Wu (Mac): https://lianwu.app/yt Codex Automations can prepare recurring research context before you start working. In this Marvin Builds demo, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, https://www.youtube.com/embed/WVxG1Ekkf_I 122 2026-05-30T06:25:09.000Z Marvin Builds https://lianwu.app/ja/videos/WVxG1Ekkf_I https://lianwu.app/media/youtube/WVxG1Ekkf_I.jpg Codex Automations for Baseball Research Try Lian Wu (Mac): https://lianwu.app/yt Codex Automations can prepare recurring research context before you start working. In this Marvin Builds demo, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, https://www.youtube.com/embed/WVxG1Ekkf_I 122 2026-05-30T06:25:09.000Z Marvin Builds https://lianwu.app/ko/videos/WVxG1Ekkf_I https://lianwu.app/media/youtube/WVxG1Ekkf_I.jpg Codex Automations for Baseball Research Try Lian Wu (Mac): https://lianwu.app/yt Codex Automations can prepare recurring research context before you start working. In this Marvin Builds demo, baseball research is the example: public Major League and Minor League sources, candidate flags, source notes, and human review questions. This is not a professional scouting tutorial. The point is the automation pattern: let Codex scan public sources, prepare a first-pass brief, and keep the final judgment human. The exact prompt used for the demo is below. Copy it, adapt it, and use it for your own recurring research task. PROMPT Create a recurring Codex Automation that prepares a concise morning baseball scout-style research brief. This is a Codex Automations demo, not a professional scouting workflow. Use baseball scouting only as the example domain. Your job is to prepare context for human review, not to make final player decisions. Run context: - Use the current run date, the user's timezone, and the latest verifiable public information available at run time. - Start the report with a short data cutoff line. - If live data is unavailable or a page cannot be verified, state the limitation clearly. Task: Scan public baseball sources and prepare a short brief that helps a human decide where to look first. Source priorities: 1. MLB official stats and player pages: - mlb.com/stats - mlb.com/player 2. Baseball Savant / Statcast: - baseballsavant.mlb.com - focus on contact quality, exit velocity, hard-hit rate, barrels, expected stats, pitch profile, or similar public metrics when available 3. MiLB public stats: - milb.com/stats - focus on age versus level, production, plate discipline, strikeout/walk pattern, role, and recent changes 4. MLB Pipeline / prospect context: - mlb.com/milb/pipeline - mlb.com/prospects 5. Transactions and roster movement: - mlb.com/transactions - official team or league transaction pages when relevant Optional watchlist: [Paste your own player watchlist here. Suggested columns: Player, Organization, https://www.youtube.com/embed/WVxG1Ekkf_I 122 2026-05-30T06:25:09.000Z Marvin Builds https://lianwu.app/videos/ooqKjlHq0TA https://lianwu.app/media/youtube/ooqKjlHq0TA.jpg I did not build a scouting app. I scheduled Codex Automations. Try Lian Wu (Mac): https://lianwu.app/yt Prompt from the video: Prepare a daily football scouting brief that can be run as a recurring Codex Automation for an independent football scout. This is a Codex Automations demo, but the report should still be grounded in real, verifiable public sources. Do not invent players, match events, ratings, injuries, contract details, market values, or transfer rumors. Role: You are an independent football scout preparing a morning brief for a small club or agency. Your job is to prepare context for human review, not to make final recruitment decisions. Task: Create a concise daily scouting brief based on the latest completed matches, recent market context, and relevant local reports from these focus markets: - England: EFL Championship - Netherlands: Eredivisie - Portugal: Primeira Liga / Liga Portugal - Belgium: Jupiler Pro League / Belgian Pro League Run context: When this automation runs, use the current run date, the user's timezone, and the latest verifiable public information available at run time. Start the report with a short data cutoff line. Watchlist: If I provide a watchlist below, use it as the main tracking list. If no watchlist is provided, do not invent one. Instead, create a league-level scouting brief and use the "New Prospect Radar" section to identify up to three real young-player leads only when they can be verified from reliable current sources. Optional watchlist input: [Paste your own watchlist here. Suggested columns: Player, Age, League, Club, Position, Scout Focus.] Source priorities: 1. Official league sites first: - efl.com - eredivisie.nl - ligaportugal.pt - proleague.be 2. Stats and match context: - fbref.com - sofascore.com - whoscored.com 3. Market, contract, injury, and transfer context: - transfermarkt.com 4. Local or journalist context: - reputable local outlets and named journalists when available - summarize non-English material in English Verification rules: - Verify current information from reliable s https://www.youtube.com/embed/ooqKjlHq0TA 160 2026-05-25T08:58:06.000Z Marvin Builds https://lianwu.app/zh-TW/videos/ooqKjlHq0TA https://lianwu.app/media/youtube/ooqKjlHq0TA.jpg I did not build a scouting app. I scheduled Codex Automations. Try Lian Wu (Mac): https://lianwu.app/yt Prompt from the video: Prepare a daily football scouting brief that can be run as a recurring Codex Automation for an independent football scout. This is a Codex Automations demo, but the report should still be grounded in real, verifiable public sources. Do not invent players, match events, ratings, injuries, contract details, market values, or transfer rumors. Role: You are an independent football scout preparing a morning brief for a small club or agency. Your job is to prepare context for human review, not to make final recruitment decisions. Task: Create a concise daily scouting brief based on the latest completed matches, recent market context, and relevant local reports from these focus markets: - England: EFL Championship - Netherlands: Eredivisie - Portugal: Primeira Liga / Liga Portugal - Belgium: Jupiler Pro League / Belgian Pro League Run context: When this automation runs, use the current run date, the user's timezone, and the latest verifiable public information available at run time. Start the report with a short data cutoff line. Watchlist: If I provide a watchlist below, use it as the main tracking list. If no watchlist is provided, do not invent one. Instead, create a league-level scouting brief and use the "New Prospect Radar" section to identify up to three real young-player leads only when they can be verified from reliable current sources. Optional watchlist input: [Paste your own watchlist here. Suggested columns: Player, Age, League, Club, Position, Scout Focus.] Source priorities: 1. Official league sites first: - efl.com - eredivisie.nl - ligaportugal.pt - proleague.be 2. Stats and match context: - fbref.com - sofascore.com - whoscored.com 3. Market, contract, injury, and transfer context: - transfermarkt.com 4. Local or journalist context: - reputable local outlets and named journalists when available - summarize non-English material in English Verification rules: - Verify current information from reliable s https://www.youtube.com/embed/ooqKjlHq0TA 160 2026-05-25T08:58:06.000Z Marvin Builds https://lianwu.app/ja/videos/ooqKjlHq0TA https://lianwu.app/media/youtube/ooqKjlHq0TA.jpg I did not build a scouting app. I scheduled Codex Automations. Try Lian Wu (Mac): https://lianwu.app/yt Prompt from the video: Prepare a daily football scouting brief that can be run as a recurring Codex Automation for an independent football scout. This is a Codex Automations demo, but the report should still be grounded in real, verifiable public sources. Do not invent players, match events, ratings, injuries, contract details, market values, or transfer rumors. Role: You are an independent football scout preparing a morning brief for a small club or agency. Your job is to prepare context for human review, not to make final recruitment decisions. Task: Create a concise daily scouting brief based on the latest completed matches, recent market context, and relevant local reports from these focus markets: - England: EFL Championship - Netherlands: Eredivisie - Portugal: Primeira Liga / Liga Portugal - Belgium: Jupiler Pro League / Belgian Pro League Run context: When this automation runs, use the current run date, the user's timezone, and the latest verifiable public information available at run time. Start the report with a short data cutoff line. Watchlist: If I provide a watchlist below, use it as the main tracking list. If no watchlist is provided, do not invent one. Instead, create a league-level scouting brief and use the "New Prospect Radar" section to identify up to three real young-player leads only when they can be verified from reliable current sources. Optional watchlist input: [Paste your own watchlist here. Suggested columns: Player, Age, League, Club, Position, Scout Focus.] Source priorities: 1. Official league sites first: - efl.com - eredivisie.nl - ligaportugal.pt - proleague.be 2. Stats and match context: - fbref.com - sofascore.com - whoscored.com 3. Market, contract, injury, and transfer context: - transfermarkt.com 4. Local or journalist context: - reputable local outlets and named journalists when available - summarize non-English material in English Verification rules: - Verify current information from reliable s https://www.youtube.com/embed/ooqKjlHq0TA 160 2026-05-25T08:58:06.000Z Marvin Builds https://lianwu.app/ko/videos/ooqKjlHq0TA https://lianwu.app/media/youtube/ooqKjlHq0TA.jpg I did not build a scouting app. I scheduled Codex Automations. Try Lian Wu (Mac): https://lianwu.app/yt Prompt from the video: Prepare a daily football scouting brief that can be run as a recurring Codex Automation for an independent football scout. This is a Codex Automations demo, but the report should still be grounded in real, verifiable public sources. Do not invent players, match events, ratings, injuries, contract details, market values, or transfer rumors. Role: You are an independent football scout preparing a morning brief for a small club or agency. Your job is to prepare context for human review, not to make final recruitment decisions. Task: Create a concise daily scouting brief based on the latest completed matches, recent market context, and relevant local reports from these focus markets: - England: EFL Championship - Netherlands: Eredivisie - Portugal: Primeira Liga / Liga Portugal - Belgium: Jupiler Pro League / Belgian Pro League Run context: When this automation runs, use the current run date, the user's timezone, and the latest verifiable public information available at run time. Start the report with a short data cutoff line. Watchlist: If I provide a watchlist below, use it as the main tracking list. If no watchlist is provided, do not invent one. Instead, create a league-level scouting brief and use the "New Prospect Radar" section to identify up to three real young-player leads only when they can be verified from reliable current sources. Optional watchlist input: [Paste your own watchlist here. Suggested columns: Player, Age, League, Club, Position, Scout Focus.] Source priorities: 1. Official league sites first: - efl.com - eredivisie.nl - ligaportugal.pt - proleague.be 2. Stats and match context: - fbref.com - sofascore.com - whoscored.com 3. Market, contract, injury, and transfer context: - transfermarkt.com 4. Local or journalist context: - reputable local outlets and named journalists when available - summarize non-English material in English Verification rules: - Verify current information from reliable s https://www.youtube.com/embed/ooqKjlHq0TA 160 2026-05-25T08:58:06.000Z Marvin Builds https://lianwu.app/videos/CjFdGYU4xGE https://lianwu.app/media/youtube/CjFdGYU4xGE.jpg Codex Automations: Mega7 morning brief #Shorts Codex Automations can prepare recurring context before work starts. In this Short, I use a Mega7 US market morning brief as one example. This is a workflow demo, not investment advice. Copy/paste prompt: Every morning at 6:00 PM Singapore, create a concise Mega7 US market morning brief based on the latest completed US trading session. Cover: Apple (AAPL), Microsoft (MSFT), Nvidia (NVDA), Amazon (AMZN), Google / Alphabet (GOOGL), Meta (META), and Tesla (TSLA). Verify information from current, reliable sources wherever possible, and include source links for key claims. Use current market data and reputable financial/news sources. If live data or current news is unavailable, state the limitation clearly and do not guess. Format the report exactly as: - Market Context: what happened in the US session overall - Mega7 Snapshot: price movement, relative strength/weakness, and notable drivers - News That Matters: company-specific or sector news affecting the Mega7 - AI / Cloud / Consumer / EV Themes: only include themes that are actually relevant today - Watchlist For Singapore Morning: what the stakeholder should pay attention to next - Confidence Level: high / medium / low, with a short reason Do not provide buy/sell recommendations. Focus on decision context. Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #Codex #AIAgents #OpenAI #MarvinBuilds https://www.youtube.com/embed/CjFdGYU4xGE 61 2026-05-23T04:54:35.000Z Marvin Builds https://lianwu.app/zh-TW/videos/CjFdGYU4xGE https://lianwu.app/media/youtube/CjFdGYU4xGE.jpg Codex Automations: Mega7 morning brief #Shorts Codex Automations can prepare recurring context before work starts. In this Short, I use a Mega7 US market morning brief as one example. This is a workflow demo, not investment advice. Copy/paste prompt: Every morning at 6:00 PM Singapore, create a concise Mega7 US market morning brief based on the latest completed US trading session. Cover: Apple (AAPL), Microsoft (MSFT), Nvidia (NVDA), Amazon (AMZN), Google / Alphabet (GOOGL), Meta (META), and Tesla (TSLA). Verify information from current, reliable sources wherever possible, and include source links for key claims. Use current market data and reputable financial/news sources. If live data or current news is unavailable, state the limitation clearly and do not guess. Format the report exactly as: - Market Context: what happened in the US session overall - Mega7 Snapshot: price movement, relative strength/weakness, and notable drivers - News That Matters: company-specific or sector news affecting the Mega7 - AI / Cloud / Consumer / EV Themes: only include themes that are actually relevant today - Watchlist For Singapore Morning: what the stakeholder should pay attention to next - Confidence Level: high / medium / low, with a short reason Do not provide buy/sell recommendations. Focus on decision context. Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #Codex #AIAgents #OpenAI #MarvinBuilds https://www.youtube.com/embed/CjFdGYU4xGE 61 2026-05-23T04:54:35.000Z Marvin Builds https://lianwu.app/ja/videos/CjFdGYU4xGE https://lianwu.app/media/youtube/CjFdGYU4xGE.jpg Codex Automations: Mega7 morning brief #Shorts Codex Automations can prepare recurring context before work starts. In this Short, I use a Mega7 US market morning brief as one example. This is a workflow demo, not investment advice. Copy/paste prompt: Every morning at 6:00 PM Singapore, create a concise Mega7 US market morning brief based on the latest completed US trading session. Cover: Apple (AAPL), Microsoft (MSFT), Nvidia (NVDA), Amazon (AMZN), Google / Alphabet (GOOGL), Meta (META), and Tesla (TSLA). Verify information from current, reliable sources wherever possible, and include source links for key claims. Use current market data and reputable financial/news sources. If live data or current news is unavailable, state the limitation clearly and do not guess. Format the report exactly as: - Market Context: what happened in the US session overall - Mega7 Snapshot: price movement, relative strength/weakness, and notable drivers - News That Matters: company-specific or sector news affecting the Mega7 - AI / Cloud / Consumer / EV Themes: only include themes that are actually relevant today - Watchlist For Singapore Morning: what the stakeholder should pay attention to next - Confidence Level: high / medium / low, with a short reason Do not provide buy/sell recommendations. Focus on decision context. Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #Codex #AIAgents #OpenAI #MarvinBuilds https://www.youtube.com/embed/CjFdGYU4xGE 61 2026-05-23T04:54:35.000Z Marvin Builds https://lianwu.app/ko/videos/CjFdGYU4xGE https://lianwu.app/media/youtube/CjFdGYU4xGE.jpg Codex Automations: Mega7 morning brief #Shorts Codex Automations can prepare recurring context before work starts. In this Short, I use a Mega7 US market morning brief as one example. This is a workflow demo, not investment advice. Copy/paste prompt: Every morning at 6:00 PM Singapore, create a concise Mega7 US market morning brief based on the latest completed US trading session. Cover: Apple (AAPL), Microsoft (MSFT), Nvidia (NVDA), Amazon (AMZN), Google / Alphabet (GOOGL), Meta (META), and Tesla (TSLA). Verify information from current, reliable sources wherever possible, and include source links for key claims. Use current market data and reputable financial/news sources. If live data or current news is unavailable, state the limitation clearly and do not guess. Format the report exactly as: - Market Context: what happened in the US session overall - Mega7 Snapshot: price movement, relative strength/weakness, and notable drivers - News That Matters: company-specific or sector news affecting the Mega7 - AI / Cloud / Consumer / EV Themes: only include themes that are actually relevant today - Watchlist For Singapore Morning: what the stakeholder should pay attention to next - Confidence Level: high / medium / low, with a short reason Do not provide buy/sell recommendations. Focus on decision context. Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #Codex #AIAgents #OpenAI #MarvinBuilds https://www.youtube.com/embed/CjFdGYU4xGE 61 2026-05-23T04:54:35.000Z Marvin Builds https://lianwu.app/videos/2HORbGbVhPc https://lianwu.app/media/youtube/2HORbGbVhPc.jpg Codex installed OpenClaw for me #Shorts If OpenClaw setup got stuck, this is the workflow I would use now: give the setup task to Codex, watch the process, and verify the result. Full workflow: https://youtu.be/EqDgXWrr2us Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #AIagents #Codex #OpenClaw #MarvinBuilds https://www.youtube.com/embed/2HORbGbVhPc 60 2026-05-21T14:54:14.000Z Marvin Builds https://lianwu.app/zh-TW/videos/2HORbGbVhPc https://lianwu.app/media/youtube/2HORbGbVhPc.jpg Codex installed OpenClaw for me #Shorts If OpenClaw setup got stuck, this is the workflow I would use now: give the setup task to Codex, watch the process, and verify the result. Full workflow: https://youtu.be/EqDgXWrr2us Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #AIagents #Codex #OpenClaw #MarvinBuilds https://www.youtube.com/embed/2HORbGbVhPc 60 2026-05-21T14:54:14.000Z Marvin Builds https://lianwu.app/ja/videos/2HORbGbVhPc https://lianwu.app/media/youtube/2HORbGbVhPc.jpg Codex installed OpenClaw for me #Shorts If OpenClaw setup got stuck, this is the workflow I would use now: give the setup task to Codex, watch the process, and verify the result. Full workflow: https://youtu.be/EqDgXWrr2us Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #AIagents #Codex #OpenClaw #MarvinBuilds https://www.youtube.com/embed/2HORbGbVhPc 60 2026-05-21T14:54:14.000Z Marvin Builds https://lianwu.app/ko/videos/2HORbGbVhPc https://lianwu.app/media/youtube/2HORbGbVhPc.jpg Codex installed OpenClaw for me #Shorts If OpenClaw setup got stuck, this is the workflow I would use now: give the setup task to Codex, watch the process, and verify the result. Full workflow: https://youtu.be/EqDgXWrr2us Marvin Builds: real AI-agent workflows, shown end to end. #Shorts #AIagents #Codex #OpenClaw #MarvinBuilds https://www.youtube.com/embed/2HORbGbVhPc 60 2026-05-21T14:54:14.000Z Marvin Builds https://lianwu.app/videos/EI3m0HcLV1Y https://lianwu.app/media/youtube/EI3m0HcLV1Y.jpg Karpathy joins Anthropic: my hope for AI agents #Shorts Andrej Karpathy joined Anthropic. My quick reaction as someone learning and building AI agents: I hope this pushes agents toward better human collaboration. Source: https://x.com/karpathy/status/2056753169888334312?s=20 #Shorts #AIagents #Software3 #Anthropic #MarvinBuilds https://www.youtube.com/embed/EI3m0HcLV1Y 31 2026-05-20T06:46:09.000Z Marvin Builds https://lianwu.app/zh-TW/videos/EI3m0HcLV1Y https://lianwu.app/media/youtube/EI3m0HcLV1Y.jpg Karpathy joins Anthropic: my hope for AI agents #Shorts Andrej Karpathy joined Anthropic. My quick reaction as someone learning and building AI agents: I hope this pushes agents toward better human collaboration. Source: https://x.com/karpathy/status/2056753169888334312?s=20 #Shorts #AIagents #Software3 #Anthropic #MarvinBuilds https://www.youtube.com/embed/EI3m0HcLV1Y 31 2026-05-20T06:46:09.000Z Marvin Builds https://lianwu.app/ja/videos/EI3m0HcLV1Y https://lianwu.app/media/youtube/EI3m0HcLV1Y.jpg Karpathy joins Anthropic: my hope for AI agents #Shorts Andrej Karpathy joined Anthropic. My quick reaction as someone learning and building AI agents: I hope this pushes agents toward better human collaboration. Source: https://x.com/karpathy/status/2056753169888334312?s=20 #Shorts #AIagents #Software3 #Anthropic #MarvinBuilds https://www.youtube.com/embed/EI3m0HcLV1Y 31 2026-05-20T06:46:09.000Z Marvin Builds https://lianwu.app/ko/videos/EI3m0HcLV1Y https://lianwu.app/media/youtube/EI3m0HcLV1Y.jpg Karpathy joins Anthropic: my hope for AI agents #Shorts Andrej Karpathy joined Anthropic. My quick reaction as someone learning and building AI agents: I hope this pushes agents toward better human collaboration. Source: https://x.com/karpathy/status/2056753169888334312?s=20 #Shorts #AIagents #Software3 #Anthropic #MarvinBuilds https://www.youtube.com/embed/EI3m0HcLV1Y 31 2026-05-20T06:46:09.000Z Marvin Builds https://lianwu.app/videos/EqDgXWrr2us https://lianwu.app/media/youtube/EqDgXWrr2us.jpg I asked Codex agent to install OpenClaw Try Lian Wu (Mac): https://lianwu.app/yt If you tried to install OpenClaw before and got stuck, this video is for you. In this first Marvin Builds episode, I use Codex on a Mac as a setup operator: it installs OpenClaw, verifies that the local dashboard is reachable, and then opens the fresh .openclaw folder so the same agent can inspect and help maintain the environment it created. The point is not that OpenClaw installation is always easy. The point is that installation can become something you delegate, watch, and verify with an AI agent. Marvin Builds: real AI-agent workflows, shown end to end. #AIagents #Codex #OpenClaw #OpenAI #MarvinBuilds https://www.youtube.com/embed/EqDgXWrr2us 344 2026-05-19T04:08:57.000Z Marvin Builds https://lianwu.app/zh-TW/videos/EqDgXWrr2us https://lianwu.app/media/youtube/EqDgXWrr2us.jpg I asked Codex agent to install OpenClaw Try Lian Wu (Mac): https://lianwu.app/yt If you tried to install OpenClaw before and got stuck, this video is for you. In this first Marvin Builds episode, I use Codex on a Mac as a setup operator: it installs OpenClaw, verifies that the local dashboard is reachable, and then opens the fresh .openclaw folder so the same agent can inspect and help maintain the environment it created. The point is not that OpenClaw installation is always easy. The point is that installation can become something you delegate, watch, and verify with an AI agent. Marvin Builds: real AI-agent workflows, shown end to end. #AIagents #Codex #OpenClaw #OpenAI #MarvinBuilds https://www.youtube.com/embed/EqDgXWrr2us 344 2026-05-19T04:08:57.000Z Marvin Builds https://lianwu.app/ja/videos/EqDgXWrr2us https://lianwu.app/media/youtube/EqDgXWrr2us.jpg I asked Codex agent to install OpenClaw Try Lian Wu (Mac): https://lianwu.app/yt If you tried to install OpenClaw before and got stuck, this video is for you. In this first Marvin Builds episode, I use Codex on a Mac as a setup operator: it installs OpenClaw, verifies that the local dashboard is reachable, and then opens the fresh .openclaw folder so the same agent can inspect and help maintain the environment it created. The point is not that OpenClaw installation is always easy. The point is that installation can become something you delegate, watch, and verify with an AI agent. Marvin Builds: real AI-agent workflows, shown end to end. #AIagents #Codex #OpenClaw #OpenAI #MarvinBuilds https://www.youtube.com/embed/EqDgXWrr2us 344 2026-05-19T04:08:57.000Z Marvin Builds https://lianwu.app/ko/videos/EqDgXWrr2us https://lianwu.app/media/youtube/EqDgXWrr2us.jpg I asked Codex agent to install OpenClaw Try Lian Wu (Mac): https://lianwu.app/yt If you tried to install OpenClaw before and got stuck, this video is for you. In this first Marvin Builds episode, I use Codex on a Mac as a setup operator: it installs OpenClaw, verifies that the local dashboard is reachable, and then opens the fresh .openclaw folder so the same agent can inspect and help maintain the environment it created. The point is not that OpenClaw installation is always easy. The point is that installation can become something you delegate, watch, and verify with an AI agent. Marvin Builds: real AI-agent workflows, shown end to end. #AIagents #Codex #OpenClaw #OpenAI #MarvinBuilds https://www.youtube.com/embed/EqDgXWrr2us 344 2026-05-19T04:08:57.000Z Marvin Builds