No they can't. They can get a clanker to crap out code at a pace that matches or beats yours, but I assume your code is better, and know that you have a better understanding of your code than the clanker user has of the clanker code.
As a test I implemented a relatively complex full stack app using AI without any input.
I told it what I want, it made all the decisions on how.
The first few iterations of the app worked well, and implemented relatively fast.
I was impressed.
However, as the features increased, AI got slower and the app also got slower. Bug fixes would end up breaking other things that were working before. Soon it was a mess of spaghetti. AI couldn't maintain it, and I sure as hell wasn't going to.
For the second test, I decided to guide it from the get go, tell it which direction to take, how to model, what tech to use. It took much longer to generate things but the app was much more performant and maintainable. It also felt like I was fighting against it some times, having to tell it to undo stupid things it did.
Did I mention it was slow?
Way too slow, waiting 5 minutes for each iteration is a major flow killer, even trying to run a few sessions concurrently to minimize my wait time didn't help since there were constant context switches.
I'm sure AI will get even better, but I really doubt it will ever be "matching the output of a human" in the full sense of the word.
AI and its use encouraged use, feels like gasoline poured on that existing mindset.
Once, my teacher in school told me that a software engineer needs to have some sadasim in it, because it will need to enjoy the pain of being stuck on a problem. I no longer believe this is true.
Most people are in it for the money, the status, the comfortable conditions, or maybe because this is the blue collar of the 21st century.
The people who are in it for the love of the game, are very little.
I do not like prompting coding agent at home, and still code as a hobby, but use coding agent extensively at work for various reasons: a feeling like better than good enough is not rewarded mainly.