They generally have a hook into the accounts the employees use to access the provider. So you can burn tokens and look 'productive' and have it have nothing to do with your work.

A lot of the early adopters have already figured out that measuring employee usage just ends with big AI bills and essentially zero assurance that the AI provider was used productively. Those folks discontinued things like leaderboards and review based on tokens consumed.

Token usage, maybe? Just set up an agent with computer use to go talk to another agent. Problem solved.

You use your worker account to access The AI.

Repos moved from local infra to an "enterprise" github account a while back.

Management is sometimes 'true believer', and only after the bill comes do they decide that maybe their employees shouldn't burn as many tokens as possible anymore.

They all picture it like AI is a great things and their employees are just not tech savvy enough to appreciate it, so all they have to do is make everyone have a taste and obviously everyone will both love it and become amazingly effective. Turns out that people generally do have a decent impression, for better or for worse, and forced micro management of it doesn't boost morale nor productivity.

I’d just ask the AI if this (whatever project I’m working on) looks correct. Then take whatever feedback it gives, whether correct or not, apply it, save it, and ship it. Sounds like what they want, yes? Give it to them. Hahahah.

I thought about malicious compliance but I don't want the end-users to get a shitty product.