AI Coding Fatigue

It's the end of work, and it's late. The Seattle summer sun fools me into thinking that the day is young and there is still much to do. Although it's been 10 hours that I've been "coding", I don't feel accomplished. Instead I am exhausted and dreading to continue the very same repetitive, evaluative work that I have been exercising all day; reviewing code written by the Powerful AI program and having to make managerial decisions about the generated work in front of me.

I submit a pull request for review and let my teammate know that it's ready for a second pair of (human) eyes. I feel confident about the submission; I went back and forth with the Powerful AI Program on our plan file, specific implementation details, test coverage, smoke tests, integration tests—the amount of engineering rigor I can slap on now with less effort than before is tremendous. Sure, I worked on it while also managing several other sessions that I jumped back and forth on every few minutes (every few seconds sometimes?) but the Powerful AI Program has my back. It knows our coding styles, our engineering standards, and has direct access to documentation when needed for following design patterns and best practices. If the Powerful AI Program can disprove a central conjecture in discrete geometry, then surely following some simple instructions and rules across our repository should be a piece of cake?

Not quite; my teammate lets me know the PR looks good but he noticed a few "oddities" in the code. New database columns are being over-indexed without good reason, newly introduced interfaces and properties are using uncommon data types when more simple ones should do. I'm reading his review comments and my first thoughts are "yes! These are such obvious points I should have picked up on a quick glance!". I feel like an idiot—I can't see the forest from the trees and I'm frustrated to go back to the Powerful AI Program and question its reasoning. There will definitely be some valid explanation for making its decisions.

When I ask, it replies with its favorite catch phrase when being corrected: "You're absolutely right! The provided suggestions are better alternatives because...".

UGH. No good "reasoning" on why it averted following more common conventions that would have been obvious to a lowly brain-powered human. With head-in-hands, I get up from my computer. My brain is fried.