How I rebuilt my workflows around AI, and what stuck

I have pointed AI at a great deal of my own work over the past year, with more enthusiasm than discernment at first. Plenty of it was hype that did not survive contact with reality. Some of it stuck, hard, and changed how I operate. The interesting part is the line between the two.

What stuck was the repetitive, rules-based grind. Drafting first versions. Research and synthesis across more sources than I could hold in my head. Reformatting, summarizing, the structured transformations that used to eat hours of an afternoon and now take minutes. I even built a small system that tailors my job applications end to end, which is a story for another post.

What did not stick was anything that turned out to require real judgment or a real relationship. AI could draft the difficult message; it could not decide whether sending it was wise. It could surface options; it could not own the consequences of choosing one. Every time I tried to hand off the judgment along with the labor, the work got worse in ways that were not always obvious until later.

So the honest accounting is this: AI gave me back time and took nothing important away, precisely because I was careful about what I refused to delegate. It is a leverage multiplier on the parts of the work that were never the point, which frees you to spend more of yourself on the parts that were.

I think that is roughly where this settles for knowledge work, at least for now. The judgment stays human. The grind goes to the machine. The people who will get the most out of this are not the ones who hand it everything, and not the ones who refuse it on principle. They are the ones who think clearly about the boundary and keep redrawing it as the tools improve.