Vertical AI: the artificial intelligence that knows your sector
Generalist AI knows a little about everything. Vertical AI knows a great deal about one thing — yours. In a factory, that difference is everything.
Vertical artificial intelligence is AI built for one specific domain — a sector, a process, a problem — instead of for everyone at once. Against generalist AI, which knows a little about everything, vertical AI knows a great deal about a single thing: yours.
Sebastian Brau's blog
Enter your email and you will get an alert when something new is published. Unsubscribe anytime.
I first wrote about this in 2018, when it was a prediction. Today it's a description: the AI that is actually working in factories is the vertical kind, and the difference with the generalist kind decides almost everything.
Horizontal versus vertical
| Horizontal AI (generalist) | Vertical AI (domain) | |
|---|---|---|
| Knows about | a bit of everything | your process, in depth |
| Trained on | the whole internet | your line's data |
| The odd case | invents an answer fluently | recognizes it, or says it doesn't know |
| Built by | a lab, for millions | a team, with you inside |
| In the factory | impressive demos | sustained shifts |
Generalist AI is astonishing and has its place. But ask it about the correct shade of a ceramic glaze or the incipient leak on a test bench and it will answer with admirable confidence — never having seen your glaze or your bench.
Vertical AI trains on what's yours. At a bakery, the system that looks at every piece five times a second cannot translate poetry: it knows those two pieces are too close together and will stick in the oven. That's all it knows. That's why it works.
What vertical AI does with your people
The 2018 headline talked about "human capital", and here is the nuance that matters most to me.
Vertical AI doesn't come to replace your people's judgement: it comes to absorb the repeatable part of that judgement so the person can spend it where it's really needed. The veteran who spots the defect by eye doesn't disappear — his eye is now on the line twenty-four hours a day, and he now decides the cases the system isn't sure about.
The catch? This only goes well if the system is built with that veteran, not behind his back. A vertical AI trained without the people who know the process learns the wrong process — and earns the rejection of exactly the person it needed most.
The bridge to 2026
The practical consequence of all this is uncomfortable for anyone hoping to buy a product and forget about it: vertical AI isn't bought finished, it's built with you inside.
A generalist model gets downloaded. A vertical one needs your data, your process and weeks of somebody in your plant understanding what's normal and what isn't. That is exactly the work of industrial forward deployed engineering: the engineer inside the factory, on the real problem.
And out of that comes my rule for telling promises from systems: ask how many weeks they'll spend in your plant. If the answer is "none, the model is already trained", you're being sold horizontal AI with a vertical label stuck on top.