I don't really believe (at least with the current state of models) in those stories were people have bot pick up a task from the backlog, automatically fix it and deploy and "works overnight". I mean, yes you can do that, but... I tried it. I gave it a list of bugs. It run for a while. Fixed all of them. Happily reported it's all fixed. And then I spent another day checking them and fixing them again because nothing was right. ** Maybe ** it makes sense for some small/trivial bugs, or in an easily testable environment, so LLM can verify it works properly.