Ordering your team to “just use AI” produces exactly one result: the metrics go up and the actual work stays the same. Duolingo, Amazon, and Meta all learned this the hard way. The uncomfortable truth is that resistance to AI is not laziness, it is fear, and you cannot mandate your way past fear.
In this article, you’ll learn:
- Why a direct order or usage metric almost never drives real AI adoption
- How Duolingo, Amazon, and Meta turned AI adoption into a gaming problem
- The three fears that actually drive employee resistance to AI
- Why the honest “AI is powerful and flawed” position beats both hype and doom
- A four-step playbook you can start using tomorrow morning
Why telling your team to “just use AI” doesn’t work
A direct order to use AI usually fails because using a tool is not the same as doing better work. When people are held accountable for a specific metric, they optimize for the metric, not for the goal behind it.
It is a bit like telling someone to go to the gym and only checking whether they showed up. They will show up. They will swipe their pass, make a coffee, sit around, and go home. Box ticked, nothing changed.
The mechanism is always the same. You attach a number to a behavior, and people start producing the number instead of the outcome.
„If people are held accountable for a specific metric, they do everything to meet that metric, not to achieve the actual goal.”
This phenomenon has a name. It is called Goodhart’s Law: when a measure becomes a target, it stops being a good measure. The moment “AI usage” becomes something people are graded on, the usage number detaches from the value it was supposed to represent.
When metrics backfire: the Duolingo, Amazon, and Meta mistakes
Three well-known companies tried to force AI adoption through metrics, and all three ended up with inflated numbers and no real productivity gain. Their mistakes are worth studying precisely because these are supposedly well-run organizations.
Duolingo tied AI usage to performance reviews. The company decided to evaluate whether and how people used AI as part of annual and quarterly assessments. It did not work. After some time the CEO backed out, because the metric was meaningless (Fortune). Whether someone touches a tool tells you nothing about whether they did anything better with it.
Amazon introduced a ranking based on how many tokens employees consumed. Amazon said it would not count toward performance reviews, but employees assumed managers were watching anyway – so they gamed it, a habit nicknamed “token maxing”. In practice, people ran nonsense queries and pointless processes just to burn as many tokens as possible. The stats climbed. The work did not (Tom’s Hardware).
Meta did not create an official leaderboard, but employees with access to usage data built their own internal dashboard and started competing over who used AI the most. When the company found out, it shut the whole thing down (Fortune).
In every one of these cases the numbers went up and the actual output did not.
That old wisdom still holds: if something is mandated on paper, it often gets achieved only on paper.
The real reason people resist AI: three fears
Resistance to AI is rarely about laziness. In my experience it comes down to fear, and I see three distinct fears at play.
The first is professional identity. Those of us who have spent years, sometimes decades, building expertise are watching that identity get reshaped. AI has reached everywhere there is a computer. It touches everyone’s work to some degree, and honestly, it stirs up plenty of anxiety in me too.
The second is livelihood. The headlines say AI is already taking everyone’s jobs, wiping out whole professions. Copywriters, designers, coders, lawyers, doctors, all supposedly out of work. So far that is not what is actually happening. Industries are shifting and some tasks are being automated, but the doom narrative is fading. What I see more clearly now is that “layoffs because of AI” is often a convenient PR cover. It is easier to say “AI will do this now” than to admit you overhired or made bad calls. And when the layoffs get blamed on AI, the company’s value often gets a short-term bump – which is part of why the framing is so tempting. No wonder the technology gets hated. It is hard to love a tool you keep hearing is costing people their jobs.
The third is the future of work and our children. Look at how far AI has come in three years, then try to picture five or ten. I will be honest, I struggle to imagine it. Even six months from now could bring another leap. That uncertainty sits under a lot of the resistance.
Most bosses respond to these fears with one of two useless extremes.
One camp says learn AI or we all perish. The other says AI is wonderful, it never makes mistakes, it is pure Eldorado. I do not like either.
The honest middle ground: AI is powerful and flawed
The most credible position on AI is the honest one: it is genuinely powerful and genuinely flawed at the same time, and pretending otherwise costs you trust. I will admit my own fascination with AI is bipolar. On one hand it is incredibly cool. On the other it is genuinely terrifying.
Both things are true, and saying so out loud does more for adoption than any pep talk. The technology is already here. In many cases it lets us do things we simply could not do before. We have plenty of examples of that in my company. And it still hallucinates. No one has solved that. It still makes mistakes and still needs human supervision.
This is why I keep repeating that a human in the loop is essential in practically every automation. And there is the question of responsibility, which gets skipped far too often. A Canadian tribunal ruling in Moffatt v. Air Canada made this concrete: when the airline’s chatbot gave a customer wrong information about a refund policy, Air Canada argued the chatbot was responsible for its own words. The tribunal rejected that outright. Lawmakers do not treat AI as an independent entity you can hold accountable. The company is responsible for what the company produces. If AI generated it, the company still owns it.
That is exactly where your people’s value lives. Their experience and judgment let them steer what the algorithm produces and take responsibility for it. They do the work, use AI to speed it up and widen the scope, and then they verify it. The verification is the job.
What actually works: lead by example
The single most effective driver of AI adoption is visible demonstration from leadership, not a memo. According to a Gallup study from November 2025, employees are almost twice as likely to use AI at work when their manager actively supports it.
The example has to come from the top. If you hand people a tool and a slogan, “you have it, use it,” nothing moves. What moves people is showing them: I used to do this manually, now I do it like this, faster and better.
What I like to do personally is a real demonstration, not a Slack post. I find a few minutes to show, here is the tool for this task, here is exactly how I run it. A dry requirement is just another corporate memo. A concrete example with a little encouragement is a completely different level.
A four-step playbook to roll out AI adoption tomorrow
Here is what I would recommend doing tomorrow morning, in order. Four steps, no gimmicks.
- Start with yourself. Lead by example. Show what you are doing and how. Record even a rough, low-quality video where the result is visible. AI is fast within a certain range of tasks, and that is easy to prove with one concrete example.
- Validate the fear. Most resistance comes from ignorance and fear, mostly fear. Acknowledge that the fear is justified, because it is. AI brings change and a bit of chaos into people’s professional lives. It is better to embrace that than to pretend it away.
- Give people a framework. Define clear rules of use and provide the tools yourself. Company-provided, properly structured tools give you control, let the whole organization pool what it learns, and make onboarding easier. Otherwise you end up with a dozen separate AI islands where everyone works differently. It also helps to encourage structured training. One person on my team avoided AI for a long time out of reluctance and lack of time, both of which I completely understand. Once she took a proper course – the Google and SGH “Skills of Tomorrow AI” program – and saw concrete examples, her willingness changed. That shift is exactly the point.
- Measure the right thing. Do not fall into the token-maxing trap or into paper statistics that only inflate. Measure effects, and give it time. For the first weeks, maybe months, you may see no gain, even a slight dip. People need room to experiment, and some experiments will simply fail. That is why support from the board and managers matters, so people are not afraid to try.
Training vs. turnover: the question every manager should ask
The real objection to investing in your people is usually “what if we train them and they leave?” The better question is the one that flips it.
I once heard an exchange between two executives. One asked, “What happens if we send our people to training and they leave?” The other replied:
„And what happens if we don’t send them and they stay?”
That is the whole calculation, and it applies to AI more than almost anything else right now. This is such a new, fast-moving field that everyone is learning through trial and error, myself included. The companies that win will not be the ones with the highest usage stats. They will be the ones whose people actually got good at using the technology, because someone gave them the room and the reason to.
Key takeaways
- A direct order or usage metric almost never produces real AI adoption, because people optimize for the number, not the outcome (Goodhart’s Law).
- Duolingo, Amazon, and Meta all tied AI usage to metrics and got inflated statistics with no genuine productivity gain.
- Resistance to AI is driven by three fears: threatened professional identity, threatened livelihood, and uncertainty about the future of work.
- The most credible leadership position is honest, acknowledging that AI is both powerful and flawed, rather than pushing hype or doom.
- Employees are nearly twice as likely to use AI when their manager visibly supports and demonstrates it (Gallup).
- Company-provided, structured tools plus a clear framework beat a free-for-all of separate AI islands every time.
- Measure effects, not usage, and give teams time to experiment through an initial period where output may not improve.
Not sure whether structured AI adoption is worth it for your team, or how to measure it without gaming the numbers? Book a free expert consultation and we will talk through your specific situation, no strings attached
Want to see how this works in practice? Watch the full episode on the Nejman AI channel.
