[mahaekoh]

I’m in the same boat. I’ve been working with a more junior engineer that’s ecstatic about AI coding and I’m slowly settling into the position that, for those of us that have developed tons of opinions about how things should be done, trying to transfer all that experience to an AI through prompting is just not efficient, and I’ve grown comfortable with saying it’s easier for me to do it myself than repeated prompting and adjusting prompts. For a more junior engineer, though, it’s a lot easier to accept what the AI did, and as long as it’s functional, their opinions aren’t strong enough to spark the urge to keep adjusting. Theres just a different utility curve for different people.

Does that mean we’ll get worse (or less opinionated) code over time? Maybe. I used to tell my team that code should be written to be easily understood by maintainers, but if all the maintainers are AI and they don’t care, does it matter?

FWIW, I still reach for Claude once in a while, and I find its response useful maybe one out of ten times, particularly when dealing with code I don’t feel the need to learn or maintain in the long run. But if reviewing Claude’s code requires me to learn the code base properly, often might as well write it myself.