Manufacturing is one of the best places to deploy AI right now.
Durable demand, far from the tech centers
When I think about what holds value in an AI world, I keep coming back to the physical economy. However good the models get, we still need bolts and screws. We still need chairs. Add onshoring, and you've got durable demand with an actual tailwind behind it.
Manufacturing is also far from the tech centers. A $100M machine shop in Ohio doesn't have a software team, and it isn't getting one.
Agents change the math
We're not the first to see this. Palantir has been selling to the biggest manufacturers in the world for years; Airbus is the flagship example. Accenture and Deloitte built empires on the same insight. But that model took armies of forward-deployed engineers, at prices only an Airbus could pay.
Agents change that math. Custom software got cheap to build and cheap to maintain, so the mid-market can afford what used to be Fortune 500 only. And unlike in tech, cheap coding agents don't kill the vendor here. A machine shop has no one to hand the agents to.
The mid-market is where the best product gets built
Here's what I think people miss: the mid-market isn't the consolation prize below enterprise. It's where the best product gets built.
The beauty of early SaaS was the feedback loop. Ship, learn, ship again. Enterprise-first companies never get that loop. The sales cycle runs a year, the deployment runs another, and the feedback comes back filtered, because everyone attached to a seven-figure rollout needs it to look like it's working. Ego protects feedback.
A shop owner spending his own money protects nothing. He tells you in week two what's broken, and you fix it by week three. It's the same iteration speed early SaaS had, pointed at problems with real enterprise texture: messy data, thirty-year-old systems, actual operations.
Hard to enter, easy to iterate
It's not an easy market. Production lines don't tolerate downtime, the ERP predates the internet, and ITAR and CMMC decide who's even allowed in the room. We moved our whole stack into a government cloud to meet those requirements. Most startups our size won't do that.
That difficulty is a feature. It keeps the lane clear on the way in, and the feedback loop compounds the product once you're there. Hard to enter, easy to iterate.
