AI Adoption Is Not a Tools Problem

When a business tells me its AI rollout has stalled, the conversation almost always starts with the tool. The tool is too slow, or the tool hallucinates, or the tool doesn’t connect to the system where the real work lives. Sometimes those complaints are accurate. They are almost never the reason the rollout stalled.

The reason is usually upstream of the software, in questions nobody wrote down: Who is allowed to be wrong now? Whose judgment does this replace, and did anyone say that out loud? What happens to the person whose expertise just became a prompt?

The surface complaint and the condition beneath

A tools problem is visible. It shows up in tickets, budgets, and vendor comparisons. A trust problem is a condition — it sits beneath the visible work and changes how everything above it behaves.

In my own business, the equivalent condition was not about AI at all. The visible operation looked healthy while the conditions beneath it — concentration risk, financing structure, assumptions about policy stability — were quietly shifting. Nobody inside the business was lying. We were answering the questions we could see.

What I am testing

My current working hypothesis has three parts:

1. Adoption stalls at the point where a tool’s output would override a person’s judgment, not at the point where the tool is technically weakest.
2. Leaders systematically underestimate how much of their operation runs on informal trust, because trust is invisible until you ask it to change shape.
3. The order of adoption matters more than the choice of vendor. A weaker tool introduced in the right sequence beats a stronger tool introduced against the grain of the org.

If the tool is the only thing you changed, the tool is the only thing you can blame.

I don’t think any of these are settled. The third one in particular could collapse under contact with more cases, which is partly why I publish this way — so the record shows what I believed before the evidence arrived.