Amazon Axes AI Leaderboard After ‘Tokenmaxxing’ Backfires

Amazon has shut down an internal leaderboard that ranked employees on their use of AI tools, after staff gamed the system with pointless activity that drove up computing costs.

The Financial Times first reported that the Kirorank service, which scored users of the US$2.9tn group’s Kiro developer platform on their AI activity, was taken offline this week.

Dave Treadwell, Senior Vice President at Amazon, told staff the leaderboard had been built with “good intentions” but had encouraged “tokenmaxxing”, the inflating of AI token consumption. Tokens are the units of data processed by AI models.

“Please don’t use AI just for the sake of using AI,” Dave told employees, according to the FT

When the scoreboard becomes the work

The rankings collapsed because employees optimised for the measure rather than the mission. Some assigned AI agents, autonomous bots that act on a user’s behalf, to needless tasks in an apparent attempt to climb the table.

Every one of those pointless tasks burned through computing capacity that Amazon had to pay for.

The company has distanced itself from the project, saying in a statement that “the beta dashboard was not a formal or approved tool, and has since been deprecated.” It adds that the leaderboard “was created by a group of employees who wanted to drive awareness for how AI can accelerate work.”

Amazon is not alone. Meta employees have similarly sought to boost their position on internal tables by driving up token consumption, the FT reports.

An 80% mandate and a runaway token bill

The gaming did not happen in a vacuum. Amazon has introduced targets for more than 80% of its developers to use AI each week, putting adoption pressure directly on individual engineers.

Amazon staff were using Kiro and MeshClaw, an in-house agent tool, to generate additional AI activity and demonstrate adoption of the technology. The costs of that pressure are rising. AI labs such as Anthropic, whose models Amazon uses extensively, have shifted from flat monthly fees to consumption-based pricing, significantly increasing some customers’ bills.

The stakes are enormous. Amazon expects to spend US$200bn in capital expenditure this year, the vast majority on AI and data centre infrastructure.

The company has also undertaken sweeping layoffs to reduce costs and help finance that investment, cutting at least 30,000 corporate roles since October 2025, CBS News reports.

Measuring code that ships, not tokens burned

Amazon’s replacement metric points to where adoption measurement is heading. The company now tracks what it terms “normalised deployments,” evidence of engineers regularly using AI to create useful code, rather than raw token consumption, the FT reports.

Dave told staff he did not want workers to focus on token use, instructing them instead to concentrate on building better products. 

Other tech leaders have reached the same conclusion. Cognizant CEO Ravi Kumar S dismisses token consumption as a “vanity metric,” telling Fortune that the firm's winning with AI measures outcomes rather than usage. Ravi is hiring more than 20,000 graduates this year while rivals cut.

For HR leaders, the episode is a textbook incentive-design failure. Mandate a behaviour, attach a public scoreboard and employees will deliver the number, whether or not it creates value.

The fix is the same as it has always been with performance metrics. Measure the outcome the business actually wants, in this case working code rather than burned tokens, and the gaming loses its reason to exist.