A VP of operations pulled up a dashboard for me and pointed at a row of red cells. Adoption by team. The tools had landed in spring, the training was mandatory, the licenses were already paid for, and a quarter into the year half his teams had barely logged in. He read the red as defiance, a pocket of people who had decided to wait him out, and he wanted to know how to force the issue.

I asked him something he was not expecting. What had those teams been told would happen to them once the tools did the job they were bought to do? He went quiet. The honest answer, the one his own all-hands messaging had been circling for months, was that the people who learned the software fastest would be fine and the people who did not would be managed out. The red cells were not an obedience problem. They were people acting sensibly inside a deal whose shape they could already see.

Sabotage is a strange word for self-preservation

A survey this year put a number on the thing leaders keep filing under resistance. It came from Writer, a company that sells enterprise AI software and has every reason to want adoption high, which polled twenty-four hundred executives and employees. Twenty-nine percent of employees admitted to actively undermining their own company’s AI strategy1, by refusing to touch the tools or ignoring what they produce. Among the youngest workers it ran closer to half. Three-quarters of the executives in the same study called that behavior a serious threat to the company’s future1. Sabotage is a heavy word. It puts the fault on the worker and treats the strategy as sound.

Then look at what those same executives said about the strategy. Ninety-two percent reported that they were actively cultivating a class of AI elite employees, the quick adopters who get the visibility and the headcount that survives1. Sixty percent said they were planning layoffs for the people who could not or would not keep pace1. More than three-quarters said anyone who failed to become proficient would be passed over for promotion and kept out of leadership1. A worker watching that does not need it spelled out. The terms are plain. Master the tool that exists to shrink the number of people like you, or be sorted into the group that gets cut.

The people carrying the risk are not the people authorizing it

This is the part the strategy decks tend to walk past. Enthusiasm for these tools is not spread evenly through a company, and it lines up almost exactly with who absorbs the harm when a rollout goes wrong.

The confidence sits with the people insulated from the consequences. The doubt sits with the people who will be standing closest to the wreck.

Pew found that only six percent of workers expect AI to bring them more opportunity over the long run, while about a third expect fewer2. The expectation of loss ran heaviest among lower and middle income workers, and the people most likely to wave it off as no real change to their prospects were the higher earners2. Set that beside an executive suite that lives in these tools for two hours a day and the trust gap explains itself. The people approving the rollout have already concluded it is good for them. The people who have to execute it ran their own numbers and landed somewhere colder.

A threat has a ceiling, and it is compliance

Coercion is a decent instrument for making someone open the software and a useless one for making them believe in it. You can compel the login. You cannot compel a person to bring judgment to the output, to catch where it is wrong, or to rebuild how they work around it, and those are the exact behaviors that separate the companies seeing real return from the ones presiding over dead pilots. Most workers do not have a manager actively making the case for these tools, and that is the one relationship with the power to move behavior. The mandate steps over it and drops a deadline and a consequence in its place.

I watched a different version of this years ago, in a place where the stakes were not a quarterly metric. In a combat zone, a team of five civilian contractors took a system throwing off more than eight hundred hostile incidents a month and brought it under fifty inside a year. Nobody marched them there. They were handed the outcome and trusted to own it, and they carried the result as theirs because it genuinely was. People reach for an outcome when they have been given the outcome to hold. They protect themselves when they have been handed a countdown.

The VP wanted the red cells turned green. The harder question, the one worth sitting in, is what he was really asking his people to take up. A tool, or the instrument of their own removal. Until that has an honest answer, every adoption dashboard is counting the wrong thing. So I will put it to you the way I put it to him. If the people closest to the work are declining the deal, what have they worked out that the strategy deck has not?