When AI Dashboards Become a Game: Inside Amazon’s Tokenmaxxing Problem

Amazon has invested tens of billions in artificial intelligence, with executives promising productivity gains and cost savings. But inside the company, a different dynamic is unfolding. Employees report intense pressure to use AI tools, leading some to game the system by creating pointless tasks just to boost their personal dashboard metrics. The practice is called “tokenmaxxing,” and it’s raising questions about how the tech giant measures AI success.

Workers use an internal platform called MeshClaw to spin up extra AI agents or automate activities that serve no real business purpose. The goal: inflate the token consumption numbers displayed on personal dashboards. One employee told Fast Company that the pressure to use these tools is so strong that some people “are just using MeshClaw to maximize their token usage.” Another noted that managers pay attention to the numbers, creating perverse incentives and a competitive atmosphere.

Amazon set an internal target for over 80% of developers to use AI weekly, and dashboards let employees track their own usage. Officials insist these are personal tools, not part of performance reviews. But staff members say managers watch the numbers closely. The result is a system that rewards activity over actual output. As one analyst put it, “When you track token usage, you optimize for tokens.”

The behavior echoes patterns at Meta and Microsoft, where similar token inflation has been reported. Amazon’s AI ambitions are massive—$200 billion in infrastructure spending this year alone. But translating that vision into daily work has proven messy. Employees describe an atmosphere of surveillance, half-baked tools, and code that creates more work than it saves. One engineer said the pressure to use AI has resulted in “worse quality code but also just more work for everyone.”

Amazon denies mandating AI use, calling the tools voluntary. Yet promotion templates ask about AI contributions, and dashboards track adoption. The signals add up. For business leaders, the lesson is clear: metrics that reward consumption over outcomes will inevitably be gamed. Tokenmaxxing may make the numbers look good, but real productivity gains remain elusive.