How AI actually sticks in operations.
Eleven lessons from putting AI to work on a real plant floor — not a vendor deck. Two run through all of them: AI rewards better questions, not more of them; and it runs on your data, not your prompts.
It began with a problem I couldn’t solve.
Not a course, not a platform — a seam-integrity problem I didn’t have the statistics for, and a hunch that AI was coming. Working that problem with AI taught me the whole framework below. Here’s that story.
Ten lessons, in order.
- The head start isn’t the tool — it’s the reps. Start where “AI is coming” meets a problem you can’t solve.
- Define your value narrowly. Hard, Capacity, Capability, or Decision — name which, or it’s cost, not value.
- Adoption has three populations. Builders, Users, Resisters — a few builders package what the many run.
- Your biggest opportunity — and biggest risk to relevance. Same amplifier, two sides. You don’t have to build, just use.
- AI relocates the gap — credentials to adoption. Require understanding, and the lift is real, not rented.
- Train the AI, not just the people. Data structure is how you onboard it to your world.
- Treat AI like a brilliant new hire. Confident even when wrong — supervise it against a source of truth.
- Scale with pull, not push. A proven, packaged win makes the next team ask to join.
- Don’t stall at “What.” Teach What / So What / Now What, and give the AI data that reaches past the What.
- The improvement has to beat the cost of changing. Aim at felt constraints, or people revert.
AI rewards the person who asks better questions — not more of them. The tool got cheap. The questions didn’t.
See it play out on the floor.
Field-tested notes on getting AI adopted in operations — the plants, the processes, and the people.
