Amazon Employees Are Faking Their AI Usage

### TL;DR Amazon employees, under pressure from management using AI token consumption as a performance metric, are forced to use AI tools unnecessarily to boost their numbers, even writing bots that automatically burn tokens, leading to resource waste and perverse incentives. ### Amazon’s AI Usage Pressure and “Token Maximization” On May 12, the *Financial Times* reported that Amazon employees are using internal AI tools to automate unnecessary tasks—just to show managers they’re using AI more frequently. This practice is called “token maximization.” The pressure stems from a goal Amazon set this year: requiring over 80% of developers to use AI tools weekly, and creating an internal leaderboard tracking each employee’s AI token consumption. Employees told the *Financial Times* they feel “immense pressure to use these tools,” and that the leaderboard creates “perverse incentives.” Amazon officially says these statistics don’t affect performance reviews, but several employees said they believe managers still check them. Amazon later restricted visibility of usage data to only the employee and their manager, and is reportedly discouraging managers from using token consumption as a performance metric altogether. ### The Reciprocal Cycle: Industry-Wide “Token Arms Race” “Token maximization” isn’t unique to Amazon. Similar phenomena have appeared at Meta and Microsoft—Meta’s internal AI leaderboard reportedly lasted only a few days after being made public. Nvidia CEO Jensen Huang has said he would be “deeply shocked” if an engineer earning $500,000 a year consumed less than 4 million tokens annually. This means the entire ecosystem is encouraging consumption and driving it ever higher, even if many of those tokens end up being for show. This adds a new layer to the financial reciprocal cycle around AI infrastructure buildout and AI company valuations. If all employees are using their own company’s AI tokens as much as possible, driving up revenue, which in turn creates higher valuations... A developer friend in Silicon Valley revealed that not only are people using AI unnecessarily to keep their numbers up, some have even built their own tools that automatically use AI—like a token virus—just to burn tokens without any interaction, in an attempt to inflate their own metrics. ### From “Mouse Jigglers” to AI Token Bots This is reminiscent of the old trick of using a “mouse jiggler” to pretend you’re at your computer. Now it’s been upgraded to automated AI token scripts: burning your tokens purely for appearances. Think about the environmental impact—massive amounts of water, energy, various forms of pollution, and huge expenditures—all just to avoid losing your job. One employee’s story is particularly amusing: he built a small tool that consumes tokens without producing anything of value. He explained that if you’re measured solely by how much AI you use, you get the same perverse incentives as the old “lines of code” metric. Some AI enthusiasts boast, “My agent output 37,000 lines of code today—what did you do?” But the key question is: were those 37 lines of good code? Probably not. ### Perverse Incentives: Back to the “Lines of Code” Trap It’s like measuring truck driver performance by how much fuel they burn. It can be a proxy for work, but it can just as easily be completely misleading—for those who just floor the accelerator or even pour gas on the gravel in the parking lot, knowing it will earn them a good performance review. That’s a perverse incentive. Put a brick on the accelerator, leave it idling overnight—obviously terrible. The arms race will only continue. Managers now get reports and try to ask the system what the user actually did with the AI, only to find it’s all garbage data. The other side will build more sophisticated context-aware waste tools: you tell the waste tool about the various tasks you’ve been working on, and it will research that and generate garbage output. Eventually, the arms race might reach a point where these tools actually do useful work—just kidding. They’ll become so good at hiding the fact that they aren’t doing anything useful that they’ll start doing something useful? That sounds a lot like corporate life, and a great short story. --- Source: YouTube video: *Amazon Employees Are Faking Their AI Usage* (https://www.youtube.com/watch?v=Mhm1C76GhoU)