Alex had the board update almost ready, and that was what bothered him. The company had made harder choices than it ever had: where AI mattered, who owned the work, what the work actually required, what information the system could trust, and what foundation would keep the first capabilities from becoming fragile one-off builds.
Paper does not change behavior.
He had watched this twice before, in programs that had nothing to do with AI. Both times the technology worked. Both times the launch was well attended. And both times, eighteen months later, half the organization had quietly reconstructed the old process alongside the new one and was running both. Nobody had refused anything. They had simply kept doing what they knew worked while performing compliance with what they had been told to do.
So he called Simon, who had spent thirty years in human resources at companies going through something — a merger, a restructuring, a system that was going to change everything and mostly did not. Simon was the only person in Alex’s network who consistently answered a question about technology by talking about people, and was consistently right to do so.
Simon did not challenge the strategy, the technology plan, or the urgency. He challenged the gap.
“You have explained how the system will work. You have not explained why anyone will choose to work through it.”
The Woman Who Stopped
Before the board update, Alex ran three short conversations: one with managers, one with the people who would use the tools daily, and one with employees already experimenting on their own.
The frontline conversation was more complicated than resistance. Most of these people were already experimenting and were not against the tools. Some worried about their jobs, which was real and worth taking seriously. But as Alex listened, the fear got more specific. They were asking whether leadership understood what AI could and could not do, which uses made sense, what work would change, what judgment would remain theirs, and how mistakes would be handled.
One person said something he kept returning to. She had used the assistant for three weeks, found it genuinely useful, and stopped.
Not because it failed. Because she could not tell whether relying on it would be held against her if something went wrong. Nobody had told her it would. Nobody had told her it would not. In the absence of an answer, she chose the version of the job she could defend.
A supervisor gave the practical form of the same question. If the assistant recommends something and I disagree, am I expected to follow it, explain why I overrode it, or pretend I never saw it?
Neither of those is a training problem. No amount of instruction on prompts and features answers either one.
The Artifact That Took a Decade to Build and Ninety Seconds to Destroy
The manager conversation produced the sharpest finding.
They were supportive in principle and unsure what they were supposed to reinforce. Nobody had told them whether AI-assisted preparation was now expected in every sales review, whether a supervisor should challenge an employee who ignored a recommendation, or whether teams should be recognized for saving time, improving quality, reducing rework, or simply experimenting.
One put it plainly. “I can tell my team AI matters. But if I keep running the same meeting, asking for the same report, and recognizing the same behavior, they will know what actually matters.”
None of these managers were being difficult. Every one had been handed a new expectation without being told what it replaced. They were still accountable for the same numbers on the same cadence, and now they were also supposed to sponsor a change nobody had defined for them. Running the meeting the way they always had was not resistance. It was the rational choice.
Then one of them asked what would happen to the account summary itself, and the question mattered more than anything else in the session. The summary had been the artifact of preparation for a decade. It was how a manager knew someone had done the work. If the assistant produced it in ninety seconds, the artifact stopped proving anything, and the manager had lost the only instrument he had for judging effort. Nobody had given him a replacement.
That is the practical shape of the problem. Every routine a company is about to disrupt exists because it answers a question a manager needs answered. Remove the routine without answering that question differently and you have not modernized the manager. You have blinded him.
The Whole Arsenal, Not Part of It
Simon’s diagnosis was blunter than most change advice. More than any transformation he had been part of, this one needed the entire change management arsenal. Not selectively. All of it.
Before anything ships: stakeholder work, so the people whose jobs change hear about it from someone who knows them rather than from an announcement. Executives visibly using the tools themselves, because a leadership team that sponsors AI without touching it has told everyone exactly how important this is. Communication that continues past launch instead of peaking at it. And managers equipped to answer the question their people will actually ask, which is whether this affects their job — because a supervisor forced to improvise that answer will improvise reassurance and then spend a year walking it back.
Then the capability itself: hands-on, recurring training rather than a single course. Protected time to do it, on the calendar, defended like any other commitment. A network of named practitioners inside each team, further along than everyone else and available for the questions nobody raises in a classroom. And honest job redesign, because roles genuinely change and pretending otherwise insults the people doing the work.
Then the things that make it hold: career paths, since the old ladder assumed a mix of skills that is being reweighted. Changes to incentives, because people optimize for what they are measured on. A route for people to say this does not work for my situation, with visible evidence that saying it occasionally changes something. And an adoption measure someone owns and checks long after launch, because usage that spikes in week three and decays quietly is the most common outcome and the easiest to miss.
“Twelve things,” Simon said. “Every one of them is somebody’s job. That is the part executives skip. They approve the transformation and staff the technology.”
He added one asymmetry Alex had not considered. Apply it proportionately. Some people take to this immediately and spending energy there is wasted. Others need real support and will get there. And some will not change at all — not through open resistance, which is manageable, but through senior people who nod in meetings, quietly do not adopt, do not require it of their teams, and wait. Their teams read that signal accurately. If someone will not change, they cannot stay in a role where they block it.
Training Is the Line Item Everyone Underfunds
Conventional software training could run once, because the system stayed relatively stable. That model does not fit here. A technique that was necessary last quarter is now built in. Something the tool could not do in the spring is routine by the fall. Training delivered once teaches people a version of the tool that will not exist for long, and worse, teaches them a ceiling that is no longer there.
What works is recurring, hands-on practice on real work rather than demonstrations. Several of Simon’s peers had settled on something close to ninety minutes a week, part on the tools everyone uses and part on the specific processes their own teams are changing.
Alex asked where ninety minutes was supposed to come from.
“That is the whole test,” Simon said. “If it comes out of the day and the workload stays the same, you have not funded training. You have asked people to do it in their own time and called it a program. It has to be defended the way you would defend a customer commitment. The first time a manager cancels it because the quarter is tight, everyone learns what it was actually worth.”
He was equally blunt about who delivers it. A vendor can teach the interface. Only someone who does the work can teach the judgment.
Where the Saved Time Goes
One question decides whether any of this holds. If people save time with AI, what happens to the time? If AI only adds workload, people learn to hide efficiency.
If it creates capacity and managers fill every recovered hour with more volume, adoption feels like a punishment. The company has to be honest about where the capacity goes: better customer coverage, faster response, less rework, more coaching, improvement work, or higher-value decisions.
The same logic applies to measurement. A sales team cannot be told to use AI for better account insight while being measured only on calls completed. An operations team cannot be asked to improve exception quality while being rewarded only for clearing volume. When the new work conflicts with what the company still measures and recognizes, the old way wins.
So Alex added one requirement to every priority capability. Before it scales, show what will change around the work, so people know the new behavior matters. If you want people to use better evidence, managers have to ask for better evidence. If you want employees to challenge bad output, you have to make that safe. And if you want adoption to last, you have to stop protecting the habits that made the old way easier.
Adapted from Feasible, Not Achieved: Why AI Transformation Stalls and the NINE Decisions That Finish It, by Weston A. Jones, MBA and Jacob C. Roxon, PhD (JRNine Press).
The Jones-Roxon Nine™ is the Operator’s System for AI Transformation.