[valleyer]

> It used to be that they only had to say no to more junior engineers’ handwritten PRs, but now they have to say no to a barrage of AI-generated code, some of it generated by managers and VPs who are politically difficult to say no to.

Holy cow. I worked at a big tech firm but left the industry prior to the emergence of InstructGPT et al., so I haven't experienced LLM code generation from the inside. Is this really happening -- upper managers and VPs proposing code changes they generated with LLMs? I don't think I'd survive.

[rozap]

Yes, this is happening. A friend of mine was telling me about a 25k LoC PR that was submitted by a product manager that he had to content with. And the politics are real - you can't just be like "no", but it's pretty tough to meaningfully review a 25k line PR, let alone from someone who knows fuck all about what they're doing and can't answer questions you might have.

[simonw]

The CEO of Shopify is filing PRs against their public repos: https://github.com/Shopify/liquid/pull/2056

(To be fair, he did build liquid and much of Shopify himself at the start of the company so he's not exactly inexperienced, but still.)

[khuey]

I skimmed that PR briefly and it appears a) not to be slop and b) to be very reviewable as its structured as a series of small commits that each make one small change. This is far from the sort of management PR I'd fear.

[suttontom]

Am I going crazy? Is a PR with 94 commits that adds 1,600 LoC actually considered "very reviewable"? Please someone tell me if I'm crazy?