“Done!”

“Wait. This sentence doesn’t make any sense.”

“You’re totally right, my bad. Here’s a new version.”

“Wait. I asked you for examples of X, but this is Y.”

“My apologies, you’re right to push back. I didn’t check thoroughly enough and relied on a single source. Here’s a corrected version.”

“Wait…”

And on, and on, and on.

When you run a company, it’s hard to do better than hiring smart, competent people who care deeply about the work. In contrast, these days AI feels like hiring an extremely capable and productive person… who doesn’t give a sh*t about the work. Sometimes, its output will blow your mind. Other times, it will deliver—at record speed—sort of what you asked for, not caring at all if something in there is a lie, or if a bunch of nice-sounding words are a complete non sequitur. It’s as nonchalant about making mistakes as about being corrected, so it can play the game forever, or until you give up.

I know: maybe you can solve this with some more tokens. Having an LLM that reviews the work of the first one and yet another that reviews the review. So maybe it doesn’t matter that they don’t care: they can get the job done. And maybe, with AI progressing at a what-just-flew-by pace, a week from now it will act just as if it cared. Who knows. For now, I’d rather do my own imperfect writing—because I do care.

Disclosure: written by me, including the em dashes—my favorite punctuation mark; grammar effectively, if uncaringly, reviewed by Claude.