Screen Recording

Coming soon

Show your coding agent what happened

Describing a UI bug in a prompt is slow and lossy. Record it instead. A recording carries what you clicked, what broke, and what the error said, and the transcript carries your narration, word for word.

The problem with typing out a repro

An agent fixing a UI bug needs the same things a teammate would. It needs the exact steps, the exact error text, and the context you'd say out loud. Typing that into a prompt loses information at every step. You paraphrase the error, forget a step, and skip the aside ("it only happens when the cart has a subscription item") that identifies the bug.

The workflow

  1. Start a recording of the window where the bug happens.
  2. Reproduce it once, narrating as you go. Say what you're doing and what you expected. Don't worry about stumbles. You can cut them from the transcript afterward.
  3. Stop. The transcript is already done, because it was generated on your Mac while you recorded.
  4. Copy the transcript with timestamps in one click, and export the clip as an MP4.
  5. Paste both into the agent chat. The transcript gives the agent your narration as text it can reason over, keyed to timestamps in the video, as in this example.
> fix the checkout crash, recording attached

00:04  I add the discount code and press Pay.
00:09  The button spins forever, and the console shows a 500
       from /api/checkout.
00:17  It only happens when the cart has a subscription item.

Why this fits agent workflows

The same clip works for people

The same clip drops into a pull request, a bug tracker, or Slack. For issues a person will triage first, see bug reports and QA.

Video

This page's video shows the workflow end to end. The steps it records are written up in the coding-agents video script.