I haven't seen it happen. They're the bottleneck now, and for some people, removing the bottleneck may mean removing the process. But there's probably a good reason why it's the bottleneck.

Some people think that as long as it passes all tests, that's all jolly - but it still needs a code review to check whether all the tests are still relevant and whether everything that needs to be tested is tested.

Even if AI gets the PRD 100% right, the PRD may be wrong. Getting the communications right is the essence of software engineering in the end.