Tokenmaxxing

The "AI-first" company that's paying for fake AI work

Amazon employees are creating bogus AI work to inflate their usage scores, and they’ve got a name for it: Tokenmaxxing.

The Financial Times reported two days ago that staff are using an internal tool called MeshClaw to automate unnecessary tasks, not because the tasks need doing, but to climb the company’s internal AI leaderboard. Amazon had introduced a target for more than 80% of developers to use AI every week and started tracking token consumption across the organization. https://t.ly/5XESz (paywall)

The official line: Token stats won’t affect performance reviews. Trust us.

A few weeks ago I shared a story from a coaching session that some people thought sounded too absurd to be true (https://t.ly/b0JCc). A designer built an AI agent to fake AI usage because their company made it a performance metric. They asked if they could stop using AI. The answer was no.

Turns out it wasn’t absurd. It was just early.

There’s a name for what’s happening here. Goodhart’s Law.

Formulated by British economist Charles Goodhart in 1975: When a measure becomes a target, it ceases to be a good measure.

You tell people “this is what counts”. They optimize for that. Not for the underlying thing you actually cared about. For the number.

Lines of code in the 80s. Citation counts in academia. AI token consumption in 2026.

The pattern is always the same. And we never seem to learn.

Now, about that “trust us”. Amazon has laid off tens of thousands of employees in the last few years. And CEO Andy Jassy explicitly warned staff that AI would mean the company needs fewer people doing certain jobs.

So you’re an Amazon employee. Your boss tells you AI might replace your role. Then asks you to prove you’re using AI. Then says don’t worry, we won’t use this data against you.

Yeah, right!

The tokenmaxxing isn’t irrational. It’s the entirely predictable response of smart people navigating an environment where job security is gone and the metrics feel like a trap.

If you have to mandate a tool, measure its usage, and build a leaderboard to drive adoption, that’s not an adoption problem. When a tool genuinely helps people do their jobs better, nobody needs a leaderboard. They just use it.

Shtty KPIs lead to shtty work. Whether you’re a startup or one of the most data-driven companies on the planet.