Someone built a free open source tool that lets your AI agent make polished screen recordings, so you can get a product demo video from one prompt. You ask your agent something like "make a video showing how human handoff works." It then: → Explores your site or desktop... app and writes the scenario → Films it with a smooth cursor, auto zooms, callouts, captions and an optional voice-over → Gives you an MP4, and works with Claude Code, Codex, Cursor or any MCP client It has no AI of its own and needs no model key, because your agent does the thinking. README read, not run.show more

3 callouts, 5‑second demo feels slick

The part nobody checks in an agent-made demo: the VO. Screen tools nail the visuals, then the voice reads "v2.1.0" as "v two point one point oh" and the whole thing sounds fake. Record one line with a version number or an email in it, and listen to that line alone.

Tried a similar flow last week, agent nails the clicks, but pacing still needs a human pass. Without that, demos feel like a speedrun and the aha moment dies

Demo video from one prompt is nice. Tell it to use Stagger and Spring on the UI too, recording looks way less stiff

one prompt to a demo video sounds great until the app asks for a login. wonder how it handles the part where it has to wait on a slow page

now the demo works and the app doesnt

This is a smart use of AI agents because it turns a product demo from a manual editing task into a repeatable workflow.

every build I ship already gets a demo gif in the readme. gonna try this and see if a real walkthrough video beats it. Really cool find.

Recording is the easy part. The hard part is making the demo deterministic, same seed data and same clicks, so you can regenerate it after every UI change.