Also counter intuitively, testing remains a somewhat ignored part of the AI SDLC. We may feel a false sense of confidence by adding 1000s of agent-generated unit tests but high velocity code changes need reliable integration tests because most systems fail at the seams. And tools like playwright etc are woefully slow and brittle for this purpose. They were barely helpful in traditional sdlc. But with today's speed of writing code, they are horribly slow.
...
integration test Playwright
...
Why is it that will spend an insane amount of compute on AI but Lynch Wen spending it on running integration tests?