An AI-written Pest browser test caught 3 of 7 deliberate breaks, and an equivalent HTTP test caught a different 3; one break slipped past both.
A prompt that asked for the server-rejection path and told the agent to break its own page produced tests that caught all 7 and ran stable 10 of 10 times.
Browser tests ran about six times slower than HTTP tests, and they failed only when the page took longer than Pest's 5-second default wait.