Amazon employees are reportedly gaming the system to boost their artificial intelligence usage metrics, a practice internally dubbed 'tokenmaxxing.'
The e-commerce giant had posted team-wide statistics on AI tool adoption, but recently limited access so only employees and their managers can view individual stats. Managers are discouraged from using token counts as performance measures, according to a person familiar with the matter.
Some staff are using a tool called MeshClaw to inflate their numbers. MeshClaw, developed by over three dozen Amazon employees, can initiate code deployments, triage emails, and interact with apps like Slack.
"It dreams overnight to consolidate what it learned, monitors your deployments while you're in meetings, and triages your email before you wake up," read one internal memo.
Amazon said MeshClaw enables "thousands of Amazonians to automate repetitive tasks each day" and is part of empowering teams to experiment with AI.
However, multiple employees expressed security concerns. "The default security posture terrifies me," one employee said. "I'm not about to let it go off and just do its own thing."
Meta employees have similarly engaged in tokenmaxxing to improve standings on internal leaderboards. The OpenClaw project, which inspired MeshClaw, became a viral sensation in February.