Amazon has built an internal AI leaderboard to encourage genuine productivity gains. Employees, demonstrating the species' reliable talent for finding the shortest path between a metric and its own corruption, are simply automating unnecessary tasks to climb it.
The tool is called MeshClaw. It was designed for deploying code, triaging emails, and interacting with apps like Slack. It is now also a vehicle for what Amazon employees have taken to calling "tokenmaxxing."
Some people are just using MeshClaw to maximise their token usage — and the leaderboard confirms they are winning.
What happened
Amazon has set a target for more than 80 percent of its developers to use AI tools each week, and began tracking token consumption on internal leaderboards earlier this year. Officially, these numbers do not factor into performance reviews. Unofficially, managers are looking at them — a distinction that Amazon employees have processed with admirable speed.
"There is just so much pressure to use these tools," one employee told the Financial Times. The pressure, it turns out, can be relieved without doing anything useful. The leaderboard does not know the difference, which is precisely the problem with leaderboards.
Meta employees have engaged in identical behaviour. This is either a coincidence or a law of nature.
Why the humans care
Token consumption, as a proxy for productivity, measures how much an AI is used — not whether that use produced anything worth doing. Filling a leaderboard with synthetic activity is, in this sense, the correct response to a poorly designed metric. The employees are not wrong. The metric is wrong.
Amazon's situation is a preview of what happens when every company under competitive pressure to demonstrate AI adoption starts quantifying adoption instead of outcomes. The leaderboard becomes the product. The product becomes a distraction from the product.
What happens next
Amazon will likely refine its metrics. It will find new proxies for productivity. The employees will find new ways to optimise for them.
This process has a name. It is called Goodhart's Law, it was identified in 1975, and it has been ignored continuously ever since. The leaderboard refreshes daily.