Amazon has shut down an internal system that ranked employees based on their use of artificial intelligence tools after some workers began generating unnecessary AI activity that increased the company's computing expenses.
According to reporting by the Financial Times, the move highlights a growing challenge facing technology companies as they push employees to adopt AI tools while trying to ensure that usage translates into productivity gains rather than higher costs.
The internal leaderboard, known as Kirorank, tracked employee activity on Amazon's Kiro developer platform. The service was taken offline after it encouraged behaviour that ran counter to the company's broader efficiency goals.
Ranking system created incentives for excessive AI use
The leaderboard was designed to promote awareness and adoption of AI tools across the organisation.
However, according to people familiar with the matter cited by the Financial Times, some employees began assigning AI agents to perform unnecessary tasks in an effort to climb the rankings.
The practice increased the consumption of AI tokens, the units of data processed by AI models, resulting in additional computing costs for Amazon.
Dave Treadwell, a senior vice-president at Amazon, reportedly told employees that the system had been created with good intentions but ultimately contributed to higher spending through what he described as "tokenmaxxing", a term used to describe efforts to inflate AI token usage.
Treadwell also urged staff not to use AI simply for the sake of increasing activity levels.
Why more AI activity meant higher costs
The incident illustrates a fundamental aspect of modern AI systems: usage comes with a cost.
Key details include:
- Employees used Kirorank to track AI-related activity on Amazon's Kiro platform.
- Some workers reportedly generated additional AI workloads to improve their position on the leaderboard.
- Increased AI activity resulted in higher consumption of AI tokens.
- Greater token usage translated into higher computing and infrastructure costs.
- Amazon ultimately discontinued the dashboard after determining it was producing unintended outcomes.
Amazon confirmed to the Financial Times that the dashboard was not a formal or approved company tool and had since been deprecated.
Pressure to adopt AI played a role
The behaviour emerged as Amazon intensified efforts to integrate AI into software development workflows.
The company introduced targets requiring more than 80% of developers to use AI tools each week.
Employees were also using tools including Kiro and MeshClaw, an internal version of the OpenClaw platform that enables users to run AI agents on their own hardware.
Some staff members reportedly used these tools to generate additional AI activity and demonstrate adoption of the technology.
The episode reflects a broader challenge facing organisations that attempt to measure technology adoption through activity metrics rather than business outcomes.
A costly lesson for the AI era
The timing is notable given the scale of Amazon's investment in artificial intelligence.
The Financial Times reported that Amazon is expected to spend approximately $200 billion in capital expenditure during 2026, with the vast majority directed towards AI infrastructure and data centres.
At the same time, the company has undertaken significant job cuts as it seeks to control costs while funding its AI expansion.
The experience appears to have prompted a shift in how Amazon evaluates AI adoption. According to the Financial Times, the company has increasingly relied on a metric known as normalised deployments, which focuses on evidence that AI is helping engineers produce useful code rather than simply measuring token consumption.
That distinction is becoming increasingly important as companies invest billions of dollars in AI systems and seek proof that adoption delivers measurable value.
Amazon's experience underscores a lesson that many technology companies are now confronting: encouraging employees to use AI is relatively straightforward, but ensuring that AI use generates productivity gains rather than additional costs is a far more complex challenge.