Seattle: Amazon has reportedly taken corrective steps after an internal experiment to encourage artificial intelligence adoption led to unintended consequences, including rising operational costs and misuse of AI tools by employees.
According to reports, the company has shut down an internal leaderboard system called “KiroRank”, which tracked and ranked employees based on their usage of AI tools on its Kiro developer platform. The move follows concerns that some workers were engaging in “tokenmaxxing” — artificially inflating their AI usage to climb rankings rather than focusing on productive outcomes.
The development highlights a growing challenge for global technology firms: balancing rapid AI adoption with cost efficiency and meaningful application.
From innovation push to unintended behaviour
The KiroRank system was originally introduced with the aim of encouraging employees to integrate AI into their daily workflows. By measuring usage, Amazon hoped to accelerate adoption of AI tools among engineers and developers.
However, the initiative appears to have backfired. Instead of promoting efficiency, it reportedly incentivised excessive and unnecessary use of AI systems.
Employees began using AI tools for tasks that did not require automation, simply to boost their scores on the leaderboard. This behaviour, described as “tokenmaxxing”, refers to maximising the consumption of AI tokens — the units of data processed by AI models — regardless of actual utility.
Dave Treadwell, a senior vice-president at Amazon, acknowledged the issue in an internal communication. While noting that the leaderboard had been created with “good intentions”, he reportedly urged employees to avoid using AI without purpose. “Please don’t use AI just for the sake of using AI,” he told staff.
Rising AI costs force rethink
The decision to remove the leaderboard comes amid broader concerns about the cost of deploying advanced AI systems. Unlike traditional software tools, AI platforms often rely on usage-based pricing, where expenses increase with the volume of data processed.
For large organisations like Amazon, even small inefficiencies in usage can quickly scale into significant costs.
Industry-wide, companies are beginning to reassess how they measure AI adoption. Earlier, metrics such as total usage or number of interactions were seen as indicators of progress. However, the Amazon episode suggests that such metrics can be misleading if not aligned with productivity or business outcomes.
Amazon has now reportedly shifted to a new performance metric called “normalised deployments”, which focuses on whether engineers are using AI tools to deliver meaningful results, such as writing effective code or completing projects efficiently.
Industry-wide cost pressures
The issue is not limited to Amazon. Across the technology sector, companies are grappling with the financial implications of large-scale AI deployment.
Anthropic, which develops advanced AI models, has moved towards consumption-based pricing structures. While this approach reflects actual usage, it has also led to higher costs for some enterprise customers.
Similarly, reports suggest that Microsoft has been reviewing its AI-related expenses and scaling back certain tool licences to manage operational budgets.
Executives across the industry have also begun raising concerns about whether heavy reliance on AI tools is always cost-effective. In some cases, the expense of running advanced AI systems may rival or exceed the cost of hiring skilled human professionals.
Amazon’s massive AI investment
Despite efforts to control costs, Amazon continues to invest heavily in artificial intelligence infrastructure. The company is expected to spend nearly $200 billion (approximately ₹16.6 lakh crore) in capital expenditure this year, with a significant portion allocated to data centres and AI systems.
Amazon also relies on external AI partnerships, including its use of models developed by Anthropic, even as it builds its own capabilities.
At the same time, the company has undertaken workforce reductions in recent months as part of broader cost-cutting measures. This reflects the delicate balance between investing in future technologies and maintaining financial discipline.
A shift towards responsible AI use
The KiroRank episode underscores a key lesson for organisations adopting AI at scale: usage alone is not a meaningful metric unless it translates into tangible value.
Experts note that companies must design incentive systems carefully to avoid unintended behaviour. Metrics that reward quantity over quality can lead to inefficiencies, increased costs and even reduced productivity.
By shifting its focus to outcome-based evaluation, Amazon appears to be aligning its AI strategy with practical business goals rather than raw adoption figures.
Conclusion
Amazon’s decision to dismantle its AI usage leaderboard highlights the growing pains of integrating artificial intelligence into large organisations. While AI remains a critical driver of innovation, its effective use requires careful management of both human behaviour and financial resources.
As more companies invest billions into AI technologies, the emphasis is likely to move from how often tools are used to how well they contribute to real-world outcomes. For Amazon and its peers, the challenge will be ensuring that AI delivers value without becoming an unchecked expense.
