Last week I published a completely serious guide to destroying your AI rollout in five easy steps. Step 3 was this: Measure logins. Someone opened the platform. Someone clicked a button. The button was clicked in the building. This is adoption.
I meant it as a joke until I read a Fortune article about Amazon this week.
What actually happened.
According to Fortune, Amazon employees have been running the company’s internal AI tool on trivial tasks — not because the tasks need doing, but because it inflates their token counts on an internal leaderboard measuring AI usage. The behavior even has a name now: “tokenmaxxing.”
Microsoft and Meta have reportedly seen similar patterns.
These are not small organizations with unsophisticated people operations. These are three of the most heavily resourced technology companies on the planet. And they have collectively discovered what happens when you tell people their success will be measured by how much they use a tool:
They use the tool. For nothing. Very enthusiastically.
Gil Luria, head of technology research at D.A. Davidson, put it plainly when asked about the dynamic: it doesn’t sound very healthy. You get the behavior you create the incentive for.
He’s right. And the implications go well beyond leaderboards.
This is not just an Amazon problem.
Every organization deploying AI right now is making a version of this bet. The metrics just look different.
Instead of token counts, it’s training completion rates. Login frequency. Utilization dashboards. Adoption percentages reported to the board. All of them share the same foundational flaw: they measure exposure to the tool, not what’s actually happening because of it.
An employee can complete every module, log in daily, and run the tool on low-stakes tasks that require no real judgment — and every single metric will look exactly like success. Meanwhile the questions that actually determine ROI go unasked and unanswered:
- Are people applying AI in ways that change outcomes?
- Are managers creating the conditions for their teams to use it well?
- Do employees trust the direction enough to actually change how they work — or are they performing compliance while privately waiting for this to blow over?
“Tokenmaxxing” is just the most visible version of a dynamic that is quietly operating inside organizations everywhere. The leaderboard made it legible. Most organizations don’t have a leaderboard. They just have a dashboard that says things are going well.
Why this keeps happening.
Luria said something worth sitting with: humans are rigid in how they do things. If you don’t create an incentive to change behavior and try something new, most people won’t.
He’s describing a real problem. But the solution most organizations reach for — measure usage, reward usage, report usage — creates a different problem. It optimizes for the appearance of change without requiring the conditions that make change real.
The conditions that make change real are not technical. They are not solved by a better metric or a sharper incentive structure. They are solved by leaders who can answer the actual question employees are asking — not “how do I use this tool” but “why should I trust this direction, what does it mean for my future, and does my manager actually believe in this or are they just passing along the memo.” (Please bring this statement your next AI workgroup meeting)
Those questions require a manager who has been prepared to answer them. Not with talking points. With genuine capability — the kind that comes from understanding how to hear what people are actually asking, navigate uncertainty without pretending to have answers they don’t have, and create enough psychological safety that resistance surfaces as a conversation instead of a leaderboard full of fake token counts.
Amazon’s employees didn’t tokenmax because they’re cynical. They tokenmax because the incentive told them the number mattered more than the work. When organizations skip the leadership layer, they teach the same lesson — just more slowly, and without a name for it.
What this should tell every CHRO reading this.
If your AI adoption metrics look healthy right now, that’s worth celebrating. It’s also worth pressure-testing.
Ask a different set of questions this quarter. Not how many people completed the training — but whether your managers can lead a real conversation about what AI means for their team’s future. Not what the utilization dashboard shows — but whether employees feel the change is happening with them or to them. Not whether the rollout is on schedule — but whether the human infrastructure surrounding it is strong enough to carry the investment to actual return.
Tokenmaxxing made headlines because it was measurable and absurd. The quieter version — compliance without commitment, usage without trust, adoption metrics that mask resistance — doesn’t make headlines. It just shows up eighteen months from now when the ROI review asks what happened.
The metric was fine. The foundation wasn’t there.
See More Posts
5 Ways Managers Can Connect AI to Career Growth
The data is clear: organizations where leaders explicitly connect AI to career growth are more than twice as likely to see higher adoption. Not twice as likely to see better outputs. Twice as likely to see people actually use it. And yet most organizations are leaving...
Trust Is the Real ROI of AI Adoption
Everyone Is Investing in AI. But Is Your Organization Ready? AI continues to dominate conversations across the enterprise. Budgets are increasing, pilot programs are expanding, and boards are asking tougher questions about how AI investments will translate into...
“AI Feels Like I’m Training My Replacement.” The Hidden Belief That Could Undermine Your AI Strategy
As organizations accelerate AI adoption, many leaders are focused on technology implementation, productivity gains, and return on investment. But there is another conversation happening inside organizations—one that is often quieter, less visible, and far more...