Resistance is usually treated as a rollout problem.
Someone did not understand the tool. Someone needs training. Someone is attached to the old way. Someone is afraid of being replaced. All of those can be true. They are also the least interesting explanations.
The better question is: what does the resistance know?
People who refuse a new tool are often pointing at something the adoption plan skipped. They may know which part of the workflow carries legal, reputational, or customer risk. They may know which inputs are quietly unreliable. They may know that the work being automated is not one task but three judgments wearing the same label. They may know that the proposed output will create more review work than it removes.
Sometimes they are defending status. Sometimes they are right.
This is why "AI literacy" is too thin as the default diagnosis. A team can understand the tool perfectly and still decide not to use it. That decision might be irrational. It might also be an accurate reading of incentives, authority, or consequence.
If the tool is supposed to draft the report, who owns the report when it is wrong? If the tool is supposed to summarize the call, which details are allowed to become memory? If the tool is supposed to recommend the next action, who is allowed to ignore it? If the answer to those questions is vague, resistance is not friction. It is governance arriving early.
The useful rollout conversation starts there.
Not: how do we get people to adopt this?
First: what would make refusal rational?
Then: which part of that refusal should change, and which part should change the system?
The organizations that learn fastest from AI will not be the ones that remove resistance. They will be the ones that can tell the difference between resistance as fear, resistance as habit, and resistance as signal.