For most of 2026, the corporate AI playbook was simple: hand everyone a seat, track usage, reward the heaviest users. That era just ended. Amazon scrapped its internal AI-usage tracking last month after employees started running bots to rack up fake activity. Meta killed its employee token leaderboard in April. And a growing body of research is explaining why the “just use it more” strategy was never going to produce the productivity boom executives promised.
The Paradox, By the Numbers
Apollo chief economist Torsten Slok coined the phrase that’s now stuck: “AI is everywhere except in the incoming macroeconomic data.” He’s echoing a version of the Solow paradox — the observation Nobel laureate Robert Solow made in 1987 that computers were showing up everywhere except in the productivity statistics. Reuters reported in June that U.S. first-quarter worker productivity was revised lower, not higher, even as AI adoption climbed.
At the same time, the London School of Economics found AI is boosting productivity by the equivalent of a full workday per week. Boston Consulting Group’s 2026 Global AI at Work report, which surveyed nearly 12,000 frontline employees, backs that up: 42% said they’re regularly saving around eight hours a week using AI. Both things are true, which is exactly the puzzle. Individual workers are saving real time. It isn’t showing up in company performance.
This point in the BCG’s report is a key takeaway: 66% of employees who reported time savings said they got little to no guidance on what to do with the hours they freed up. Half aren’t redirecting that time toward anything more strategic. The time is being saved, or rather wasted. It’s just evaporating. There is no increase in productivity.
Tokenmaxxing: The Behavior You Get When You Reward the Wrong Metric
The most visible symptom of this gap was “tokenmaxxing” — companies pushing employees to maximize AI usage as a proxy for AI value. The Financial Times reported in May that Amazon staff were gaming internal AI metrics, in some cases deploying bots to generate usage on tasks nobody needed done. Amazon senior vice president Dave Treadwell had to tell staff directly: “Please don’t use AI just for the sake of using AI.” The company pulled its tracking dashboard days later.
Gil Luria, head of technology research at D.A. Davidson, put the mechanism plainly: “You get the behavior that you create the incentive for. So if you tell people they’ll succeed if they use a resource more, of course they’ll use it more.” More usage, in other words, not more value.
The bill for that behavior is now landing. Microsoft reportedly canceled a large share of its direct Claude code licenses over cost, and Uber burned through its entire 2026 AI coding tools budget in the first four months of the year. Nvidia’s Bryan Catanzaro told Axios: “For my team, the cost of compute is far beyond the costs of the employees.” Token-based pricing made usage cheap to generate and expensive to sustain, and companies optimized for the wrong side of that equation.
The Real Bottleneck Is Leadership, Not the Model
People & Organization leader at BCG David Martin analyses the reason to this paradox brutally honest as: bad management. “Senior leaders are really struggling to articulate what the vision and strategy is on AI,” he said. “It increases employee fear. It makes it harder for them to even understand what objectives they’re pushing for, and it trickles through to adoption, usage, and the like.”
That fear has a second-order effect worth noting: when companies treat AI agents as digital coworkers rather than tools, employees get more anxious about being replaced — and anxious employees hide what they’re doing instead of sharing it. Martin calls open sharing “incredibly important” for organizational speed, and says it’s “not natural for fearful employees.” A company that never explains its AI strategy beyond “use it more” is quietly training its own people to work in silos.
What Operators Should Actually Do
The fix BCG and Fortune’s reporting point to isn’t more AI access — most companies have already blanketed their workforce with tools. It’s specificity: which roles should use AI for which tasks, what the freed-up time is supposed to produce, and how success gets measured beyond token counts. Okta COO Eric Kelleher put his finger on the deeper habit companies need to break: leaders are still asking “what’s the org chart look like?” instead of what work actually needs doing differently now that AI has changed the cost of producing a first draft, an analysis, or a customer response. That is what makes a real difference and that is what companies who are turning AI-native actually do differently already.
The Takeaway
AI is generating real time savings at the individual level — that part of the hype is holding up. What’s not holding up is the assumption that savings convert to performance automatically. They don’t. They convert when leaders tell people specifically what to do with the reclaimed hours, and they evaporate — or turn into an expensive vanity metric — when the only instruction is “use it more.” Tokenmaxxing died this year because companies finally looked at the bill. The harder work — deciding what AI adoption is actually for — is just getting started.
Become AI-native now – hire workforce which builds its workflows around AI now and win big with the help of CASURO.AI.
Sources
Picture credits to Ron Lach
- AI productivity gains are real but so is bad management — Fortune
- The AI productivity paradox: More work, not less — Fortune
- BCG 2026 Global AI at Work report
- US first-quarter worker productivity, labor costs revised lower — Reuters
- AI boosts productivity by the equivalent of one workday per week — LSE
- Amazon employees were “tokenmaxxing” — Fortune
- Meta killed its employee AI token dashboard — Fortune
- Microsoft’s AI cost problem — Fortune
- Uber COO on AI spending and tokens — Fortune
- The cost of compute vs. human workers — Axios
- “Tokenmaxxing is dead” — Fortune
