Amazon has shut down an internal AI leaderboard after employees reportedly inflated their use of artificial intelligence tools to climb company rankings, turning a workplace productivity experiment into a costly warning about how easily corporate AI targets can be gamed.
The system, known internally as Kirorank, ranked developers by activity on Kiro, Amazon's AI coding platform. According to reports, some workers began using autonomous AI agents to generate unnecessary activity, pushing up token consumption and adding costs at a time when Amazon is spending heavily on AI infrastructure while reducing parts of its workforce.
The incident may have started inside a single company, but it highlights a growing challenge spreading across the technology sector. Businesses are investing billions of dollars in artificial intelligence and demanding evidence that employees are using it. Measuring adoption is relatively straightforward. Determining whether it is genuinely improving results is proving much harder.
According to reports, Amazon senior vice-president Dave Treadwell told employees that the leaderboard had been created with good intentions but was encouraging behaviour that failed to improve products or outcomes. Instead, some workers became focused on maximising token consumption, the units processed by AI systems during tasks and interactions.
The practice has become known as "tokenmaxxing," a trend in parts of the technology industry where users attempt to maximise AI activity regardless of whether the output creates meaningful value.
For Amazon, every unnecessary token carries a real cost. AI systems require enormous computing resources, specialised hardware and vast data-centre capacity. When AI agents are running tasks that serve little practical purpose, companies still absorb the infrastructure and energy costs behind that activity.
The episode exposes a wider problem facing employers as artificial intelligence moves deeper into daily work. Executives want proof that expensive AI investments are paying off. Investors want evidence that efficiency is improving. Managers are being asked to show that teams are embracing new technology.
That pressure can create incentives that are difficult to control.
Employees naturally respond to the metrics organisations reward. If workers believe promotions, recognition or career prospects are linked to AI usage, some may focus on maximising visible activity rather than producing better outcomes. The result can be a growing gap between what internal metrics report and what employees are actually delivering.
Amazon appears to have recognised that risk. The company has reportedly shifted toward a measurement known as "normalised deployments," which focuses on whether engineers are regularly using AI to produce useful code rather than simply generating large volumes of activity.
Across the technology industry, executives are now under growing pressure to prove that soaring AI budgets are delivering results rather than simply generating more usage. Boardrooms want evidence that spending on artificial intelligence is leading to faster software development, stronger output and lower costs instead of becoming another expensive corporate target.
The issue is becoming more important as firms spend extraordinary sums on artificial intelligence while simultaneously looking for ways to control costs elsewhere. Many have reduced hiring, cut jobs or reorganised teams while presenting AI as a major driver of future efficiency.
Amazon alone is expected to spend roughly $200 billion this year, much of it directed toward AI infrastructure and data centres. Similar spending programmes are underway throughout the industry as businesses compete to secure the computing power needed for increasingly sophisticated AI systems.
For workers, the transition is creating a more complicated workplace environment. Employees are being encouraged to adopt AI tools while also facing rising expectations around performance, efficiency and output. In some cases, simply being productive may no longer be enough. Workers may also feel pressure to demonstrate that they are using the approved tools in ways management can easily measure.
That dynamic is beginning to reshape how performance is judged in parts of the economy. Employers now face a difficult balancing act. They want staff using AI more often, but they also need to know whether those tools are producing better work rather than simply generating more activity.
Amazon's abandoned leaderboard may ultimately be remembered as a small warning about a much larger shift. As more employers tie performance to AI adoption, workers and managers alike could find themselves spending increasing amounts of time proving they are using the technology rather than proving the technology is helping.
