TAG

# AI workflow

2026-07-20

Why Does AI Keep Getting It Wrong? It's Not Dumb — You Didn't Say It Clearly

You ask AI to build something, and eight or nine times out of ten it comes back wrong, and you think "this AI is useless." But after a few months of doing this every day, I'm more and more sure of one thing: when AI gets it wrong, most of the time it isn't dumb — I didn't say it clearly. Same requirement, phrased differently, and it nails it on the first try. There's a very plain truth underneath this — AI can't read minds. It does what you said, not what you meant. This piece lays out the requirement-writing method I've worked out: the five parts of a requirement that lets AI get it right the first time, the five most common ways of not saying it clearly and how to fix each, and three moves that turn AI from a hand into a brain. All of it is copy-pasteable.

2026-07-16

I handed AI about half of my PM job, and there are a few things I still don't dare hand over

This year I handed AI roughly half of my day-to-day PM work — first drafts of documents, digging up competitor research, sorting hundreds of pieces of user feedback, turning meetings into action items, building clickable prototypes. It's fast and it never complains. But a few other things I haven't dared hand over, and I don't plan to. Not because AI can't do them — the opposite, some it does more smoothly than I do. It's because once you hand those off and they go wrong, you can't catch it, and by the time you do it's already too late. This piece lays out my 'hand over vs. keep' line, one item at a time, including the pits I nearly fell into after handing things off.

2026-07-14

A Day as an AI-Era PM: How I Turned One Sentence Into a Prototype You Can Actually Tap

One afternoon last week, I turned 'I want a little thing that tracks what I spend' into a prototype my coworker could actually tap on his phone — record a real expense, see a real pie chart. Not a single line of code. Everyone thinks the AI era means PMs have to go learn programming. It's actually the reverse: the skill that's worth money now is getting your words clear enough that AI gets it right on the first try. This isn't a lecture — it's exactly how I did it: how to make AI interrogate me first, how to change only one thing at a time, why the very first version should run on real data, and how to use 'can you tap through it?' as your acceptance line. Includes the potholes I stepped in.