Amazon built an internal AI tool staffed by 36 engineers. Its employees responded by gaming it.
The practice has earned its own name inside the company: “tokenmaxxing.” Workers are using MeshClaw, Amazon’s internally developed platform for creating AI agents to automate tasks, not because they need the automation, but because they need the usage numbers. The goal is to look busy on AI leaderboards that track how frequently developers interact with the company’s tools.
The leaderboard problem
Amazon mandates that over 80% of its developers use AI tools on a weekly basis. That usage gets tracked and displayed on team leaderboards. Employees have started running unnecessary automated tasks through MeshClaw purely to inflate their token counts.
Amazon has said these token metrics won’t factor into performance reviews. But workers report feeling anxious about low rankings regardless of official assurances.
MeshClaw and the 36-engineer investment
MeshClaw was developed by a team of 36 engineers at Amazon. The tool is designed to let developers build AI agents that automate various tasks.
The 80% weekly usage mandate signals to employees that there’s a minimum acceptable level of engagement, regardless of whether their particular workflow benefits from AI assistance.
A pattern across Big Tech
Amazon isn’t alone in this. Similar tokenmaxxing behaviors have been reported at Meta and Microsoft recently. The pattern is consistent across the hyperscalers: invest billions in AI infrastructure, mandate internal adoption, track usage metrics, and then watch as employees optimize for the metric rather than the outcome.
Tech companies have collectively poured enormous sums into AI technologies since late 2022, and they need to show shareholders that the investment is paying off. Internal adoption rates are one of the easiest numbers to point to in a board presentation.
What this means for investors
For anyone watching the AI trade, tokenmaxxing should raise a yellow flag about how internal AI adoption numbers get reported. If employees at Amazon, Meta, and Microsoft are inflating their usage stats, then the adoption figures these companies cite publicly may be less meaningful than they appear.
