Amazon workers tokenmaxxing inflate internal AI metrics

Amazon employees are routing nonessential jobs through MeshClaw, the company’s internal platform for creating AI agents, to increase token counts and climb visible team leaderboards. Inside the company the practice is called tokenmaxxing.

MeshClaw was developed by a team of 36 engineers to let developers build agents that automate routine tasks. Amazon displays usage statistics on team leaderboards and set a target that more than 80% of developers use AI tools weekly.

Employees report they sometimes run automated tasks that do not affect their actual work solely to boost usage numbers and leaderboard position. The activity increases token counts that the platform tracks.

Amazon informed staff that token metrics will not be used in formal performance reviews. Despite that, workers report anxiety about low leaderboard rankings and feel pressure to meet visible adoption targets.

Similar practices have been reported at other major technology companies where internal adoption metrics and leaderboards are used to track tool use.

Company presentations sometimes cite internal adoption rates to show returns on AI investments. Internal sources and employees report that inflated usage metrics may not reflect real productivity gains or deeper integration of AI into workflows.

Frequency-of-use measures show how often tools are opened or tokens are consumed. That metric does not directly measure whether agents save time, improve work quality, or reduce costs.

MeshClaw remains part of Amazon’s internal AI tooling. The company continues to display team leaderboards and encourage broader adoption of AI agents across developer teams.