I have, and we have engineers producing approximately 10x as many PRs with nearly as good of quality. Code Reviews are now the bottleneck, at some point we're going to trust it to just work and not review (or maybe we just sample, or maybe we review the code base quarterly to tidy up). In the same way most folks never check their TCP packets, or that their grocery bill was perfect (except in unexpected outcomes), people are just going to trust and accept the slight variance/misses.
The tools I'm referring to are well skilled Claude code / codex. I'm sure open weights are not far behind.
The bloom is off the AI rose and the consulting class doesn't have a good replacement magic bean yet. Expect more fumbling like this until they arrive at the next narrative.
Yeah, if even 5% of the work requires human input, that means that the maximum achievable productivity benefit is 20x. But in my experience, that style of work looks too much like 'dude in his bedroom doing whatever he wants' for management to allow such independence. So you get loaded up with meetings, admin tasks etc. which kill most of the gains.
The tools I'm referring to are well skilled Claude code / codex. I'm sure open weights are not far behind.