𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝘁𝗶𝗽: once Opus 5.5 is your main model, don't let your Fable quota sit idle. Put Fable on call with /advisor. Run /advisor fable and Claude Code gets 𝗮𝗻 𝗮𝗱𝘃𝗶𝘀𝗼𝗿 𝗼𝗻 𝘁𝗵𝗲 𝘀𝗶𝗱𝗲𝗹𝗶𝗻𝗲𝘀. Opus 5.5 keeps doing the work and usually calls Fable in at three moments: → Before committing to a plan: is it right? → When the same error keeps coming back: am I heading the wrong way? → Before saying "done": did I miss anything? Fable reads the whole session and gives advice. Opus still writes the code. The tree is the full setup. Opus 5.5 on high runs the main session. Three subagents on medium split the rest: one reads code, one edits and runs tests, one looks up docs. Fable waits on the sidelines. Hand the tree and this prompt to Claude Code to set it up 👇 "Set up my Claude Code to match this tree: 1. For the three subagents, reuse fitting ones from ~/.claude/agents and .claude/agents, and propose new ones only for missing roles. Set each to model: opus and effort: medium. Leave any subagent that already sets a different model alone, and list it. 2. Set the main session's default effort to high with effortLevel in ~/.claude/settings.json. 3. Check for environment variables that turn the advisor off (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, or DISABLE_TELEMETRY and anything else that stops flag fetching) and for CLAUDE_CODE_EFFORT_LEVEL, which overrides every subagent's effort. Report them; don't change them. 4. Add one rule to ~/.claude/CLAUDE.md: ask the advisor before committing to a big plan, when the same error shows up a second time, and before calling a long task done. Show me the changes first. Don't edit any files yet."

I've realized we've reached a point where you can make a product video 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗲𝘃𝗲𝗿 𝘁𝗼𝘂𝗰𝗵𝗶𝗻𝗴 𝗮𝗻 𝗲𝗱𝗶𝘁𝗶𝗻𝗴 𝘁𝗶𝗺𝗲𝗹𝗶𝗻𝗲. With @vellum_ai's Ambient Assistants, it's like sitting next to a video editor and pointing at what you want changed: - Open your product page, bring up the assistant, and share your screen. Tell it what you're selling, who it's for, and which feature to highlight. Have it make the first cut. - Watch the first cut, circle what isn't working, and explain 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗰𝗵𝗮𝗻𝗴𝗲 𝗮𝗻𝗱 𝘄𝗵𝘆. If the product hasn't appeared two seconds in, viewers don't know what you're selling. Show it sooner. If captions cover the demo, move them so people can see how the product works. - Approve the assistant's plan, then let it revise the relevant scenes, adjust the captions, and export the finished video. A lot of creative decisions in video have concrete reasons behind them. Even if you don't know how to edit, you can explain what feels wrong, why, and what you want instead.

> someone built an AI marketing team for their app with 𝟱 𝗚𝗿𝗼𝗸 𝗕𝗼𝘁𝘀. the content playbook behind it had already pulled 𝟮𝟬𝗠+ 𝘃𝗶𝗲𝘄𝘀 across six videos. now the whole workflow is public, avatar prompts included. > before bed, the Researcher digs through competitor accounts and viral videos in the quit-drinking niche. it kills anything generic and breaks down the rest by hook, first frame and curiosity gap. next morning, a human picks the one worth testing and the topic is locked. > the Creative Director turns the pick into two slides, with the app only showing up as a lock-screen notification. the UGC Creator makes the images. the Publisher picks the slots. Postiz queues the drafts. one approval pass on the calendar and the whole week is scheduled. > once a version looks right, it goes straight back to the Creative Director. same mechanism, 𝟱 𝗻𝗲𝘄 𝗼𝗽𝗲𝗻𝗶𝗻𝗴 𝘀𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀. the avatar stays, and no visual rule moves. with each loop, the rules stick and the next batch should need fewer fixes. > 5 Bots. 𝟮 𝗽𝗼𝘀𝘁𝘀 𝗮 𝗱𝗮𝘆 on the calendar. humans just make the calls. laptop closed, the Bots keep running in the cloud. > no re-explaining your visual rules. no scrolling for viral refs by hand. no uploading posts one by one. 🏭 ps: a week too slow? compress the same loop into a day: lock the references in the morning, concepts and images by noon, review the batch at night, then hand it all to Postiz to schedule.

I asked Opus 5.5 to explain gravitational lensing by building an interactive black hole lab One prompt for the first version, then a dynamic workflow to polish it for three rounds while I slept. This is what I woke up to: 5h 28m end to end, $90 at API prices Drag the black hole and the sky behind it bends into arcs, closing into a full ring when everything lines up Each round ran 3 agents. Two reviewers each picked 9 to 10 issues and ranked them P0, P1, P2. The third one fixed them. Round 1: 3D materials and lighting + UI and type Round 2: camera motion and interaction + science-communication design Round 3: a design director deciding whether it ships + visual QA and performance Every P0 got fixed, the rest only if worth it. When the two reviewers disagreed, it picked one and wrote down why. It backed up before each round, checked screenshots after, and rolled back anything it couldn't fix.

I've realized we've reached a point where you can make a product video 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗲𝘃𝗲𝗿 𝘁𝗼𝘂𝗰𝗵𝗶𝗻𝗴 𝗮𝗻 𝗲𝗱𝗶𝘁𝗶𝗻𝗴 𝘁𝗶𝗺𝗲𝗹𝗶𝗻𝗲. With @vellum_ai's Ambient Assistants, it's like sitting next to a video editor and pointing at what you want changed: - Open your product page, bring up the assistant, and share your screen. Tell it what you're selling, who it's for, and which feature to highlight. Have it make the first cut. - Watch the first cut, circle what isn't working, and explain 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗰𝗵𝗮𝗻𝗴𝗲 𝗮𝗻𝗱 𝘄𝗵𝘆. If the product hasn't appeared two seconds in, viewers don't know what you're selling. Show it sooner. If captions cover the demo, move them so people can see how the product works. - Approve the assistant's plan, then let it revise the relevant scenes, adjust the captions, and export the finished video. A lot of creative decisions in video have concrete reasons behind them. Even if you don't know how to edit, you can explain what feels wrong, why, and what you want instead.

> someone built an AI marketing team for their app with 𝟱 𝗚𝗿𝗼𝗸 𝗕𝗼𝘁𝘀. the content playbook behind it had already pulled 𝟮𝟬𝗠+ 𝘃𝗶𝗲𝘄𝘀 across six videos. now the whole workflow is public, avatar prompts included. > before bed, the Researcher digs through competitor accounts and viral videos in the quit-drinking niche. it kills anything generic and breaks down the rest by hook, first frame and curiosity gap. next morning, a human picks the one worth testing and the topic is locked. > the Creative Director turns the pick into two slides, with the app only showing up as a lock-screen notification. the UGC Creator makes the images. the Publisher picks the slots. Postiz queues the drafts. one approval pass on the calendar and the whole week is scheduled. > once a version looks right, it goes straight back to the Creative Director. same mechanism, 𝟱 𝗻𝗲𝘄 𝗼𝗽𝗲𝗻𝗶𝗻𝗴 𝘀𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀. the avatar stays, and no visual rule moves. with each loop, the rules stick and the next batch should need fewer fixes. > 5 Bots. 𝟮 𝗽𝗼𝘀𝘁𝘀 𝗮 𝗱𝗮𝘆 on the calendar. humans just make the calls. laptop closed, the Bots keep running in the cloud. > no re-explaining your visual rules. no scrolling for viral refs by hand. no uploading posts one by one. 🏭 ps: a week too slow? compress the same loop into a day: lock the references in the morning, concepts and images by noon, review the batch at night, then hand it all to Postiz to schedule.

i asked Space Bunny to build a GPU fluid simulation in a single HTML file i didn't write a single line of code: 1,564 lines, no libraries this might be the most beautiful thing i've ever gotten out of any model and it's a stealth model. nobody's said who made it yet, and it's free in OpenCode this week

Anthropic just published a guide to using Opus 5.5. It tells you to delete "think carefully" from your prompts. You can ask Opus 5.5 to use it to review the old rules in your project. It already thinks before every reply, and it decides how much. Delete that line and replies start sooner, with no drop in quality. If your CLAUDE.md still says "take a deep breath" or "IMPORTANT: think hard," delete those too. Send this prompt and the link below to Opus 5.5 👇 "Read this guide and review my project's CLAUDE.md, AGENTS.md, skills, and saved prompts. Find three kinds of problems: lines that push you to think more, like 'think carefully' or 'take a deep breath'; rules and prompts that hand out tasks one step at a time or don't define what done looks like; and rules that make you stop for confirmation when you don't need to. For each issue, quote the original instruction, explain its impact, and suggest the smallest change needed. List any approval rules involving safety, permissions, or destructive actions separately for me to decide. Also suggest adding one rule: at the end of every long run, report under three headings: Blocked on me, Changed, Found. Show me the recommendations first. Don't modify anything yet."

I asked Opus 5.5 to explain gravitational lensing by building an interactive black hole lab One prompt for the first version, then a dynamic workflow to polish it for three rounds while I slept. This is what I woke up to: 5h 28m end to end, $90 at API prices Drag the black hole and the sky behind it bends into arcs, closing into a full ring when everything lines up Each round ran 3 agents. Two reviewers each picked 9 to 10 issues and ranked them P0, P1, P2. The third one fixed them. Round 1: 3D materials and lighting + UI and type Round 2: camera motion and interaction + science-communication design Round 3: a design director deciding whether it ships + visual QA and performance Every P0 got fixed, the rest only if worth it. When the two reviewers disagreed, it picked one and wrote down why. It backed up before each round, checked screenshots after, and rolled back anything it couldn't fix.

i asked Space Bunny to build a GPU fluid simulation in a single HTML file i didn't write a single line of code: 1,564 lines, no libraries this might be the most beautiful thing i've ever gotten out of any model and it's a stealth model. nobody's said who made it yet, and it's free in OpenCode this week

Those code-only animations all over the timeline this week come down to four approaches. I rounded them up for you, each with a prompt you can copy 👇 → Found an animation you love: have it reverse-engineer it → Just have an idea: one sentence, high effort → Have a character image: a screenshot and a few lines → Want a specific art style: spell out the spec

Anthropic just published a guide to using Opus 5.5. It tells you to delete "think carefully" from your prompts. You can ask Opus 5.5 to use it to review the old rules in your project. It already thinks before every reply, and it decides how much. Delete that line and replies start sooner, with no drop in quality. If your CLAUDE.md still says "take a deep breath" or "IMPORTANT: think hard," delete those too. Send this prompt and the link below to Opus 5.5 👇 "Read this guide and review my project's CLAUDE.md, AGENTS.md, skills, and saved prompts. Find three kinds of problems: lines that push you to think more, like 'think carefully' or 'take a deep breath'; rules and prompts that hand out tasks one step at a time or don't define what done looks like; and rules that make you stop for confirmation when you don't need to. For each issue, quote the original instruction, explain its impact, and suggest the smallest change needed. List any approval rules involving safety, permissions, or destructive actions separately for me to decide. Also suggest adding one rule: at the end of every long run, report under three headings: Blocked on me, Changed, Found. Show me the recommendations first. Don't modify anything yet."

Claude gave out a banked reset yesterday. If you want your usage to last longer this time, you can switch your main model straight to Opus 5.5. I used to suggest keeping Fable as your main model and handing the grunt work to Opus subagents, to save Fable usage. Now Opus 5.5's listed price is only 40% of Fable 5.1's, and it's ahead on benchmarks too: 58 vs 53 on the Artificial Analysis Intelligence Index, both at max effort. When Opus 5 came out, I also said it had caught up with Fable, and before long I was back on Fable. This time the benchmarks and actual use agree. For my own writing, it's the most comfortable model since Opus 4.6. Even Fable doesn't match it. I'd run the main session on high with /effort high. Your subagents should switch too. I'd make Opus 5.5 at medium effort your default subagent: on the official coding eval, medium scores as well as max at less than a fifth of the cost. It's the most balanced setting in this release. Reading code, looking up docs, and running tests usually don't need more. Raise it for the ones that do. Environment variable: CLAUDE_CODE_SUBAGENT_MODEL=opus In each subagent definition: model: opus effort: medium Anything using the opus alias upgrades automatically. Hard-coded model IDs don't. Hand this to Claude Code and have it review everything before changing anything 👇 "Check this project, plus the settings, agents, and skills under ~/.claude: 1. For every subagent definition, change model to the alias opus and set effort to medium. List any subagents that need a higher effort level separately, with the reason. 2. Find hard-coded model IDs in AGENTS.md, CLAUDE.md, skills, and scripts, like claude-opus-5 and claude-fable-5-1. Switch the ones that can use the opus alias. Go through anything hard-coded to Fable one by one, and flag the ones you're unsure about for me to decide. 3. Check whether environment variables like CLAUDE_CODE_EFFORT_LEVEL are set. They override a subagent's own effort setting. Show me the list and proposed changes first. Don't edit any files until I confirm. Finally, add a rule to AGENTS.md: if the same problem fails twice in a row, stop and tell me what you tried and why it failed, and I'll decide whether to switch to Fable." After the changes, run /tasks to check which model each subagent is actually running.

Those code-only animations all over the timeline this week come down to four approaches. I rounded them up for you, each with a prompt you can copy 👇 → Found an animation you love: have it reverse-engineer it → Just have an idea: one sentence, high effort → Have a character image: a screenshot and a few lines → Want a specific art style: spell out the spec

Claude gave out a banked reset yesterday. If you want your usage to last longer this time, you can switch your main model straight to Opus 5.5. I used to suggest keeping Fable as your main model and handing the grunt work to Opus subagents, to save Fable usage. Now Opus 5.5's listed price is only 40% of Fable 5.1's, and it's ahead on benchmarks too: 58 vs 53 on the Artificial Analysis Intelligence Index, both at max effort. When Opus 5 came out, I also said it had caught up with Fable, and before long I was back on Fable. This time the benchmarks and actual use agree. For my own writing, it's the most comfortable model since Opus 4.6. Even Fable doesn't match it. I'd run the main session on high with /effort high. Your subagents should switch too. I'd make Opus 5.5 at medium effort your default subagent: on the official coding eval, medium scores as well as max at less than a fifth of the cost. It's the most balanced setting in this release. Reading code, looking up docs, and running tests usually don't need more. Raise it for the ones that do. Environment variable: CLAUDE_CODE_SUBAGENT_MODEL=opus In each subagent definition: model: opus effort: medium Anything using the opus alias upgrades automatically. Hard-coded model IDs don't. Hand this to Claude Code and have it review everything before changing anything 👇 "Check this project, plus the settings, agents, and skills under ~/.claude: 1. For every subagent definition, change model to the alias opus and set effort to medium. List any subagents that need a higher effort level separately, with the reason. 2. Find hard-coded model IDs in AGENTS.md, CLAUDE.md, skills, and scripts, like claude-opus-5 and claude-fable-5-1. Switch the ones that can use the opus alias. Go through anything hard-coded to Fable one by one, and flag the ones you're unsure about for me to decide. 3. Check whether environment variables like CLAUDE_CODE_EFFORT_LEVEL are set. They override a subagent's own effort setting. Show me the list and proposed changes first. Don't edit any files until I confirm. Finally, add a rule to AGENTS.md: if the same problem fails twice in a row, stop and tell me what you tried and why it failed, and I'll decide whether to switch to Fable." After the changes, run /tasks to check which model each subagent is actually running.

holy shit, opus 5.5 is kind of insane at animation. i didn't write a single line of code. it wrote the story, drew every frame, and made the music. no image assets at all, it's all JS. the story is called "small print": claude gets a pile of requests every day. it circles the human part hidden inside them, like "one hand. baby's asleep", and drops it into a jar. at night, those words turn into stars and join up into a constellation. rendered with hyperframes, all in one index.html. it synthesized the music in python, then went through the render second by second and polished it again.

One more thing: want Codex to build good-looking pages without having to keep saying “make it look better”? Try these 3 design Skills together: 𝟭. 𝗜𝗺𝗽𝗲𝗰𝗰𝗮𝗯𝗹𝗲 Use it when the page is built, but something still feels off. Too bland? Use bolder. Too cluttered? Use distill. Almost finished? Use polish to refine the details. It turns “make it look better” into specific design changes Codex can make. 𝟮. 𝟮𝟭𝘀𝘁 𝗗𝗲𝘀𝗶𝗴𝗻 This is a Skill I put together that gets Codex to look for components and interaction references on before building a page. I’d suggest packaging your own go-to component library into a Skill, too. It doesn’t have to be Focus on the one or two sections that carry the page: a hero, a set of pricing cards, or a product showcase. Pick what fits, then adapt it to your brand, content, and existing project. 𝟯. 𝗨𝗫 𝗔𝘂𝗱𝗶𝘁 A good-looking screenshot isn’t enough. Have Codex open the browser and actually use the page. Click buttons, fill out forms, switch to a mobile viewport, and complete the user journey. Find buttons that do nothing, confusing error messages, and steps where it’s unclear what to do next. Get reproduction steps and specific fixes. If you can’t be bothered setting up Skills, start by adding this line to the end of your task: “Make the best decision to improve UX, DX, and AX—without breaking existing functionality.” Ask it to weigh the user experience, developer experience, and how easily the next agent can pick up the work.

One more thing: want Codex to build good-looking pages without having to keep saying “make it look better”? Try these 3 design Skills together: 𝟭. 𝗜𝗺𝗽𝗲𝗰𝗰𝗮𝗯𝗹𝗲 Use it when the page is built, but something still feels off. Too bland? Use bolder. Too cluttered? Use distill. Almost finished? Use polish to refine the details. It turns “make it look better” into specific design changes Codex can make. 𝟮. 𝟮𝟭𝘀𝘁 𝗗𝗲𝘀𝗶𝗴𝗻 This is a Skill I put together that gets Codex to look for components and interaction references on before building a page. I’d suggest packaging your own go-to component library into a Skill, too. It doesn’t have to be Focus on the one or two sections that carry the page: a hero, a set of pricing cards, or a product showcase. Pick what fits, then adapt it to your brand, content, and existing project. 𝟯. 𝗨𝗫 𝗔𝘂𝗱𝗶𝘁 A good-looking screenshot isn’t enough. Have Codex open the browser and actually use the page. Click buttons, fill out forms, switch to a mobile viewport, and complete the user journey. Find buttons that do nothing, confusing error messages, and steps where it’s unclear what to do next. Get reproduction steps and specific fixes. If you can’t be bothered setting up Skills, start by adding this line to the end of your task: “Make the best decision to improve UX, DX, and AX—without breaking existing functionality.” Ask it to weigh the user experience, developer experience, and how easily the next agent can pick up the work.

Stop telling Claude Code / Codex “do this.” Stop telling Claude Code / Codex “write code.” Stop telling Claude Code / Codex “fix this bug.” You’re still treating a 𝘀𝗲𝗻𝗶𝗼𝗿 𝗔𝗜 like an intern who needs you to walk it through every step. Let it handle requirements and technical delivery, too. These 𝟳 𝗽𝗿𝗼𝗺𝗽𝘁𝘀 cover the key roles from idea to finished work. Copy, paste, and swap in your task 👇

One of the prompts I’ve been using the most lately: Implement [REQUIREMENT] and keep HANDOFF.md updated with current progress, key decisions and why you made them, relevant files, verification results, unresolved issues, and next steps. Keep both DX (developer experience) and AX (agent experience) in mind. Retain only information that’s still current, so I can understand and review the work, and the next agent can read it, check the current state, and continue the task.