Willison dropped three kakapo photos from image search into Claude Opus 5.5 as visual reference and asked for an HTML5 canvas pixel-art party of at least 20 jumping kakapo with confetti.
To get a Keynote-ready clip he gave a local Claude Code session a plain-language spec: load the page, record 15 seconds, wait 3 seconds before clicking, and spread several clicks around the clickable area.
The Playwright script Claude Code wrote needs no screen recorder: it opens a 1280x720 browser context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → with record_video_dir set, then fires ten timed clicks across the corners, edges and centre, sleeping to hit each timestamp.
The script declares its playwright dependency in an inline '# /// script' metadata header, so it runs as a single standalone file.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Why it matters
Shows a reusable pattern for turning a one-off AI-generated HTML artifact into a shareable video: have a coding AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → script Playwright to load, interact with, and screen-record the page, with the prompt and script shared in full.
