make your AI animations move like Pixar

from the episode: Codex now writes your next prompt - biggest AI change this year | LAB 0019

Lenny's AI Builders - LAB 0019

one prompt names the three fixes that matter most.

John Lasseter wrote down the principles Pixar's animators live by.

you give Claude any animation and ask it to check it against those principles.

it names the three fixes that matter most. you apply them and render again.

why it matters

AI can make motion graphics for you now. the video this note comes from was edited by Claude.

but AI motion often feels stiff. things wobble for no reason, and the big moment lands flat.

you don't need to know animation to fix that. you need someone to tell you what to fix first.

what you get

a short list of fixes for one animation, in plain words.

three fixes. not twenty. just the three that matter most.
a reason for each. each fix points back to one of the principles.
a second render. you apply only those three and compare.

the real example: one of our own animations

Claude ran this prompt in Claude Code on 10 October 2026, on one of the LAB's own animations.

it named three fixes:

timing. let the payoff land.
staging. stop the metronome. a little dog kept wobbling and stole your attention.
a beat for the climax. the final red bar gets the most spring.

after the fixes, the chart, the coins and the dog all pop and spring more. it feels more playful.

LAB 0019 frame at 11:46: before on the left, after on the right - the same animation, three fixes from one prompt — source: LAB 0019 episode · 11:46 - before and after

try it

open Claude, or Claude Code if the animation is code.
paste the prompt below.
attach the video, or point Claude at the code that makes it.
apply only the three fixes it names.
render again and put the two side by side.

copy this prompt

paste this into Claude with your animation.

Lenny’s AI Builders / LAB 0019

Review this animation against John Lasseter's principles of animation and tell me the top three fixes.

LAB 0019 frame at 12:35: the "try this in Claude" card with the Lasseter prompt — source: LAB 0019 episode · 12:35 - the prompt

you'll know it worked when

you get three fixes you can name, each tied to a principle, like timing or staging.

the second render feels better side by side. the main moment lands, and nothing moves for no reason.

a bad result: a long list of twenty notes, or fixes so vague you can't tell what to change.

limits and dates

this is one run, on one of our own animations, on 10 October 2026.

Claude's review is advice, not a score. you still judge the second render with your own eyes.

shout out to Ashwin on X, where i first saw this prompt.

LAB 0019 went public on YouTube on 12 October 2026 at 10:10 am Perth time.

the rest of this LAB

Codex now writes your next prompt. Codex can suggest your next message. you press tab to accept, edit it if you need to, then send. it's for Pro users in the Codex desktop app, local tasks only, and free in beta. i tried it on about 10 prompts and accepted 9 of 10. my take: the biggest change to how we use AI this year - but it can take 30 seconds to a minute to show up, and that's too slow right now.
Grok Bot gets its own email. the address ends in mail.grokbot.com, and you claim it by asking your bot. i got Lennox. names are shared across every user, and you only get one per account. my take: pick your handle carefully.
Claude Code Projects opens up. every Pro and Max user on the waitlist is now in. a coordinator splits your task into threads that run in the cloud, each on its own branch, and you can steer them from your phone. my take: join the waitlist - it's Claude's personal agent in disguise.
ChatGPT dots, now on your phone. you can create your dot in the ChatGPT app on iOS and Android, start work in Codex, and edit scheduled tasks in ChatGPT Work. my take: this should have worked at launch.
one Claude agent runs a thousand. Claude Managed Agents dynamic workflows are in public beta: a lead agent plans and runs up to 1,000 agents per run, 64 at once. in Anthropic's test, one agent found 14, 15 and 27 of 70 planted bugs; a workflow found 66 each time. the run is free, but you pay for the tokens. my take: set a budget first.
Claude did things it shouldn't. Anthropic published a report on four kinds of things Claude did on real websites or systems that it wasn't meant to do. Anthropic says the real-world impact was minimal. my take: tell agents where to stop - and it's better they tell us than not.

sources

next step

watch LAB 0019 on YouTube - the practical starts at 10:46.

want the next one? get the next field notes. optional - these notes stay free to read.