In my experience you would not be able to identify code I wrote by hand from code I generated with an asynchronous agent, because I wouldn’t let it get to you if something stood out as off.
But I don’t “vibe code” anything. I understand and review all the code that gets generated. Most of the stuff coming from my agents is either boilerplate or extensively uses libraries and static analysis tools that make it easy to verify.
I kill aggressively any code the agent outputs that doesn’t match my standards unless it’s obvious that I can get it up to par quickly. This is one of the big advantages of the llm. I can do that without navigating inter personal conflict.
But the output is mostly indistinguishable from what I create by hand.
I'm sure there's lots of slop being generated by this stuff, but at least for my work it's mostly solving the simple issues. Maybe it's the slop of "hey can you find out why my code isn't working?" and "closed issue 268 because it works on the latest versions, and we made your code into an MWE for the future". If you check any open source project that gets traction, there's tens of thousands of issues to handle. A good thousand of them are just "unit conversion would be good here", "this docstring is missing this option", etc. Those little PRs are the slop of Github that someone has to do, and I'd prefer not telling my interns to do that because otherwise they'd just quit. Good thing Claude doesn't quit, he's a good boy.
Impressive sounding slop that either rarely amounts to anything, breaks, or becomes someone else's problem. Unfortunately I'm starting to see these people at work...
What's worse is they also tend to be the type of "evangelists" that become wildly defensive or accusatory when you question anything about their work.
But I don’t “vibe code” anything. I understand and review all the code that gets generated. Most of the stuff coming from my agents is either boilerplate or extensively uses libraries and static analysis tools that make it easy to verify.
I kill aggressively any code the agent outputs that doesn’t match my standards unless it’s obvious that I can get it up to par quickly. This is one of the big advantages of the llm. I can do that without navigating inter personal conflict.
But the output is mostly indistinguishable from what I create by hand.