Generating a PR with an agent is cheap now. Checking that it didn't break a user flow still takes a person and a browser.

That second job is where a lot of engineering time goes now.

@ContextQa released Ship, an AI agent that works as an automated QA engineer:

> It reads the PR
> works out which user flows the change affects, and runs those flows against the deployed app.
> reproduces bugs reported in Slack or Linear, finds the likely cause, and hands the evidence to a developer or a coding agent like Claude or Codex to fix.

If agents are going to ship more code, something other than a human has to exercise the result.