Amazon employees are running the company’s internal AI agent platform on unnecessary tasks to inflate their usage scores, according to a Financial Times investigation cited by multiple outlets including Fortune and India Today. The practice, which employees call “tokenmaxxing,” emerged after Amazon set targets requiring more than 80% of developers to use AI tools every week and began tracking consumption on internal leaderboards.
The tool at the center of the gaming is MeshClaw, an internal agent platform reportedly inspired by OpenClaw’s architecture. MeshClaw lets employees build AI agents that can deploy code, triage emails, and interact with workplace applications like Slack. According to The Decoder, one Amazon employee told the FT: “There is just so much pressure to use these tools. Some people are just using MeshClaw to maximise their token usage.”
The Measurement Problem
Amazon officially told employees that token statistics would not factor into performance reviews. Employees did not believe it. “Managers are looking at it,” another worker told the FT, according to HR Magazine. “When they track usage it creates perverse incentives and some people are very competitive about it.”
The dynamic is a textbook case of Goodhart’s Law: when a measure becomes a target, it ceases to be a good measure. Amazon wanted to accelerate AI adoption across its workforce. It picked token consumption as the metric. Employees responded by maximizing the metric instead of the underlying goal.
Not Just Amazon
The tokenmaxxing phenomenon extends beyond Amazon. At Meta, an employee built an internal leaderboard called “Claudeonomics” that ranked roughly 85,000 workers by token consumption, according to Fortune. In a 30-day window, total usage on the dashboard exceeded 60 trillion tokens. Neither CEO Mark Zuckerberg nor CTO Andrew Bosworth ranked in the top 250. Meta took the dashboard down after press reporting, but Bosworth publicly endorsed the logic behind it, telling Forbes his best engineer was spending the equivalent of his salary in tokens and was “5x to 10x more productive.”
Microsoft has reportedly seen similar behavior, according to Fortune. All three companies are heavily invested in the AI infrastructure their employees are being pushed to consume.
The Capex Question
Gil Luria, head of technology research at D.A. Davidson, told Fortune the dynamic concerned him. “That doesn’t sound very healthy. 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.”
The stakes are significant. Amazon expects to spend around $200 billion this year, mostly on AI and data center infrastructure, according to India Today. Combined 2026 capital expenditure from Amazon, Microsoft, Alphabet, and Meta is pushing $700 billion, with Wall Street projections exceeding $1 trillion for 2027, per Fortune. Luria flagged what he called “circular activity,” where the same companies invest in their suppliers and customers simultaneously.
The Adoption Gap
The tokenmaxxing trend exposes a core challenge for enterprises deploying agent platforms: the gap between mandating adoption and measuring whether adoption is productive. Token consumption tracks volume, not value. An agent that deploys code correctly and an agent that re-sorts an already-sorted inbox consume tokens equally, but only one generates ROI.
For companies building agent infrastructure, including OpenClaw deployments, the lesson is that usage metrics without outcome measurement create noise, not signal. The companies spending hundreds of billions on AI infrastructure need better answers for what “successful agent adoption” actually looks like at scale.