AI and: why I never let the tool win
My one rule for using AI in client work: additive to craft, never a replacement for it.
I use AI every day. I also have one rule about it: AI and, not AI instead.
That rule didn’t come from a manifesto. It came from watching what happens when you let the tool drive. Give it enough rope and it will happily build the whole site for you: fast, clean, grammatically spotless, and completely forgettable. The work comes out finished. It just doesn’t come out good.
The day the tool almost won
I’ve felt the pull. Early on, I let AI carry a homepage draft almost end to end. It read fine. Every section sat where it should, every sentence behaved. I nearly sent it.
Then I read it out loud, which is usually when you find out. Nobody was home in it. No point of view, no line you’d remember an hour later, nothing that sounded like the actual company whose name sat at the top. It was a very confident impression of a website.
So I rewrote it. Not all of it. The structure was sound, and plenty of what the tool gave me survived the second pass. But the parts that carry the weight, the ones a visitor actually feels, had to be made by a person who cared how they landed.
Where the tool earns its place
None of that turned me against AI. The fastest path through a blank page is rarely a blank page, and the tool is very good at getting you off zero. I use it to draft, to boil a pile of research down to something I can argue with, to throw up five directions when I’d otherwise marry the first one, and to catch the small mistakes a tired human misses at the end of a long day. That’s real leverage, and I’m not precious about taking it.
What matters is what I ask it for. I ask for raw material: speed, options, a rough cut. I never ask it for the verdict.
Where the human stays
Taste doesn’t come from a prompt. The parts you can feel belong to a person from the first draft to the last pass: the rhythm of a sentence, the restraint to leave a space empty, the judgment about what to cut. A model can hand you a hundred options. It can’t tell you which one is right for this client, on this page, for the single thing the page exists to do.
So I keep the final call. Every time. The line of copy, the spacing that finally feels settled, the choice that makes a page feel considered instead of generated: those are mine, on purpose.
Used this way, AI makes good work faster. It doesn’t make fast work good. Those are two different things, and keeping them straight is the whole job. The tool never wins, because I never hand it the one decision that would let it.