A May 12, 2026 report says some Amazon employees have been automating trivial work to boost token use on internal AI

A May 12, 2026 report says some Amazon employees have been automating trivial work to boost token use on internal AI leaderboards, a practice called “tokenmaxxing,” raising concerns that usage metrics misstate real AI adoption.

Amazon engineers have been deliberately automating low‑value tasks to increase token consumption on the company’s internal AI leaderboards, a behavior coverage on May 12, 2026 labeled “tokenmaxxing.” The practice matters because the leaderboards were introduced as part of targets meant to accelerate AI use, and inflated token counts can distort measures meant to show genuine adoption and impact.

Employees are using an internal tool called MeshClaw to build AI agents that trigger code deployments, triage emails and interact with apps such as Slack. Staff told reporters they have used MeshClaw specifically to drive up token use after the company set goals requiring more than 80% of developers to use AI each week and began tracking token consumption on public internal rankings earlier this year.

Although the tracked token numbers are not formally part of performance reviews, employees say managers notice usage and that the visibility creates perverse incentives to generate activity rather than business value. The coverage also notes similar “tokenmaxxing” at another major tech company, and observers quoted in the report argue raw token counts are a poor proxy for productivity gains.

The practical consequence is that leaderboard‑driven metrics may misrepresent where AI is actually delivering value, encouraging wasteful automation and complicating internal assessments. That gap between measured activity and measurable benefit could prompt companies to rethink broad usage quotas and the public ranking of token consumption if they want clearer signals about AI’s business impact.