Amazon built MeshClaw, an internal AI platform staffed by 36 engineers, to help developers automate tasks. But some employees are now gaming the system in a practice known inside the company as 'tokenmaxxing.'
They run unnecessary automated jobs through MeshClaw not because they need the automation, but to inflate their token counts on AI leaderboards. Amazon mandates that over 80% of developers use AI tools weekly-a target that is tracked and displayed publicly within the organization.
While Amazon says these metrics won't factor into performance reviews, workers admit to feeling anxious about low rankings. The behavior mirrors similar patterns at Meta and Microsoft, where employees optimize for adoption metrics instead of actual outcomes.
For investors tracking the AI boom, tokenmaxxing raises a yellow flag: if internal usage numbers are inflated, the adoption figures tech companies cite publicly may be less reliable than they appear.