Yeah, I get that there are startups trying to do it. But I work with Cursor quite a bit… there is no way I would trust an LLM code agent to take high-level direction and issue a PR on anything but the most trivial bug fix.

Last year they couldn’t even do a simple fix (they could add a null coalescing operator or an early return which didn’t make sense, that’s about it). Now I’m getting hundreds of LOC of functionality with multiple kLOC of tests out of the agent mode. No way it gets in without a few iterations, but it’s sooo much better than last April.