AI writes the code. I still have to babysit it.
I've noticed this has become the actual bottleneck for me.
The AI isn't slow anymore. It'll finish tasks way faster than I would. But I still can't just give it a list of things and walk away.
It finishes something, I check it, and maybe the UI moved a little. Or it looks completely fine until I actually click around and realize something broke. Sometimes everything compiles, all the tests pass, but the feature still isn't really usable.
So I fix it, give it the next task, wait again, check again... and before I know it it's 3am and I've spent the whole day basically babysitting it.
At this point the coding isn't what takes most of my time.
It's verifying that what it built actually works the way a real user would expect.
That shift from writing code to verifying behavior feels like a meaningful change in where the work actually happens.
Has that changed how you think about what an AI coding tool ultimately needs to be good at, or do you still see implementation as the primary problem to solve?