Yeah, I was thinking it’s impossible to fake AI codegen with humans, for however terrible it is, it vomits the code it generates quickly.

I of course take issue with:

LLMs were already more than capable of generating high-quality code.

That claim will remain unjustified so long as LLMs equally generate imagined function calls and such as much as they like ‘real’ function calls.

But in any event, it’s also telling that despite ostensibly being a company focused around AI code gen, when it came to doing stupid in-house stuff that didn’t need doing they went with a bunch of human developers. And many of those are knock-offs of projects that should be pretty firmly in the territory of LLM codegen to let you ‘vibe code’.