Enterprises everywhere have been urging employees to adopt AI, with internal leaderboards springing up to show who has used the most AI tokens. Such games can backfire, though, as Amazon recently discovered.
Kirorank, an unofficial leaderboard tracking usage of Amazon’s Kiro AI tool, ranked workers according to their AI activity, but senior managers at Amazon found that employees were creating AI agents to carry out unnecessary tasks in an attempt to boost their scores, a practise known as tokenmaxxing, according to a report in the Financial Times. The tracker has now been taken offline.
Kirorank
The FT quoted an Amazon senior VP as saying that the leaderboard had been built “with good intentions” but the compute costs it generated were too high.
Amazon is not the only company to find that attempts to boost AI use can have negative implications. In April, Meta killed an unofficial ranking system called Claudeoconomics, which had also led to a frenzy of tokenmaxxing.
Token usage is easy to measure; the business benefits of that usage less so. AI vendors are looking for better ways to measure the value of their services, so far with little success. Salesforce’s attempt to develop a new metric was not well received.
Enterprises now find themselves facing a delicate balancing act, as they look to encourage AI use without racking up considerable computing costs. In March, PwC’s US CEO Paul Griggs told executives that those who did not use AI would not be at the company for very long. Use AI — but not too much, appears to be the message.