Staff allegedly inflated AI token use to climb the internal leaderboard on AI use, according to reports
Amazon has shut down an internal AI leaderboard after employees were found "tokenmaxxing" at a cost to the company, according to reports.
The ranking, known internally as "Kirorank," scored staff according to their activity on Kiro, the company's AI developer platform.
Workers were notified this week that the service had been switched off, two people familiar with the matter told The Financial Times.
Dave Treadwell, an Amazon senior vice-president, told employees this week that the tool had been created with "good intentions," according to people who heard his comments.
Even so, the leaderboard had pushed up costs as workers padded their token consumption, the data units that AI models process, the senior vice-president said.
"Please don't use AI just for the sake of using AI," Treadwell told staff, as quoted by The Financial Times.
'Tokenmaxxing' at Amazon
The trouble, according to the report, was that the leaderboard rewarded volume over value.
As a result, some staff started assigning autonomous AI agents to carry out needless actions on a user's behalf to climb Kirorank in a workplace trend known as "tokenmaxxing."
The trend refers to the practice of maximising AI usage, specifically by consuming as many tokens as possible with autonomous agents, according to job-listing platform Built In.
On its website, Built In noted that tokenmaxxing has emerged in some corners of the tech industry, where high AI use regardless of output is treated as a signal of productivity.
But for Amazon, those extra tasks carried out by "tokenmaxxing" employees burned through computing capacity that the company had to fund.
In a statement, Amazon distanced itself from the project, saying "the beta dashboard was not a formal or approved tool" and that it had since been deprecated.
The leaderboard, it added, had been built by a group of employees keen to raise awareness of how AI could speed up their work, and the company remained focused on operating efficiently.
Amazon has set a target for more than 80% of its developers to use AI each week.
But to gauge success more meaningfully, Amazon has adopted a measure it calls "normalised deployments," which tracks engineers regularly using AI to write useful code rather than counting raw token totals.
Treadwell has also told staff he wanted them to stop concentrating on usage and instead focus on building better products.
The move comes as Amazon reduces its workforce as it carries out a sweeping AI expansion. The company is expected to lay out roughly $200 billion in capital spending this year, the bulk of it earmarked for AI and data-centre infrastructure.
