Even in pre-AI world, there were no testers around me for the past many years, and those people who actually tested the final product before the release (their titles were probably something like "release engineer", not "Tester" or "QA") would file a bug if a problem is found, but they were never responsible for the bug I introduced. The commits had my name on them, and people would come to me with questions. If I could not answer the question because I forgot the code, I would sit down and figure out.

Now with AI, at least after a ~year of using AI for development, things haven't changed in terms of responsibility: my commits still have my name on them, and people will come to me with questions. If I cannot answer the question because I forgot the code or never wrote it in the first place, I will sit down and figure out.

As for the reviewer, I never saw anyone being grilled for stamping buggy code. People make mistakes all the time, including when reviewing someone's code. Code reviews are nice to have and catch real issues, but it's naive to think that they could protect from code breakage, regardless if it's post-AI or not.

Even if an AI agent generates the pr, the responsibility always lies with the human side who approves and merges it. An agent still cannot be held accountable.