(Mostly posting this because of the hashtag, which I thought of All By Myself.)

@aparrish (obligatory disclaimer: I’m flailing in the dark with this stuff just like everybody else) I find myself spending much more of my dev time finding alternative ways to get that understanding without reading (all of) the code. LLM agents are good for that too, writing smaller scripts that pull out and visualise (deterministically!) important information. And also refactoring (more thoroughly than I would have time for without agents) to emphasise patterns that are easy to scan-and-agree.

@kyle @mattiem are you calling me a village?

@kyle @mattiem holy shit speaking as a parent: THIS

@jonny (I *hope* they’re not right for them either, it’s a bleak bloody future if this is the most effective way for anyone to ship something people use.)

@jonny Part of their choice is what “better” means. Producing better *code* requires using human judgement much more, which slows down shipping and costs more in human labour. The choice they’re making is to decide not to do that. Like I said earlier: I think they’re either betting on model improvements making it irrelevant, or the nature of the product keeping it irrelevant. I have no idea if they’re right for them; damn sure the same choice isn’t right for anything I’m close to.

@jonny sure that’s true for the things that directly involve user input and LLM input/output. But it’s an engineering choice how much you route through that versus how much you isolate in deterministic data flows. They’ve gone ALL IN on fast-and-loose, and the thing I find so distressing is that nothing yet has slapped them down for it.