Every week someone tells doctors, mechanics and analysts the same thing: learn AI or become irrelevant.
I think this advice is mostly wrong — and I say that as someone who builds with AI every day.
Two kinds of knowledge
There are two kinds of knowledge in every product.
The first is domain knowledge. A doctor knows what actually happens in a clinic between "patient walks in" and "patient gets better" — the insurance edge cases, the no-shows, the way a nurse really uses the schedule. A mechanic knows why customers come back angry, and it's almost never the repair itself. A business analyst can smell a broken process from one spreadsheet.
Programmers only guess at this. I've watched it for years: the hardest bugs in business software aren't in the code, they're in the assumptions. Someone guessed how a clinic works instead of knowing.
The second kind is build knowledge: how to turn a description into working software. This used to take a team and months. Now there are people who do it with AI tools in days — fast and well. Not because they're geniuses, but because they've spent their hours on exactly that skill.
The bad trade
Here's the trade the "learn AI" advice actually offers a specialist: spend a year of evenings becoming a mediocre prompt engineer, so you can build a mediocre version of the product you already understand perfectly.
That's a bad trade. Your scarce asset is twenty years of knowing what programmers only guess at. A year spent becoming average at someone else's craft is a year not spent deepening yours.
The rare thing was never the code. It's the expertise. Code is getting cheaper by the month; knowing what to build is not.
1 + 1
So my actual advice to specialists: don't learn to build. Find someone who's fast with AI, and bring the thing they can't get anywhere else — the truth about how your field really works.
In a good pairing, 1 + 1 isn't 2.
Page had ideas about links. Brin was fast with math. 1 + 1 = Google.
I'm curious about the other side of this: if you're a developer, have you ever built something with a domain expert who couldn't code at all? How did it go?
Top comments (0)