What is the story about?
Amazon will no longer track employee AI activity through an internal rankings system after concerns emerged that workers were exploiting the tool to artificially increase their usage of artificial intelligence services, driving up the company’s computing costs.
According to the Financial Times, employees were informed this week that Amazon had taken offline “Kirorank”, a dashboard linked to its Kiro developer platform that ranked workers according to their AI-related activity. The system had been introduced to encourage wider use of AI tools among developers, but reportedly led to unintended behaviour as some employees sought to boost their scores through unnecessary activity.
The issue became tied to what staff internally referred to as “tokenmaxxing”, a practice in which workers deliberately inflated their consumption of AI tokens, the units of data processed by AI models, in order to appear more engaged with the company’s AI initiatives.
Dave Treadwell, Amazon senior vice-president, acknowledged the problem in comments to employees earlier this week, according to the Financial Times. He reportedly said the leaderboard had been created with “good intentions” but had ultimately contributed to higher operational costs because some workers were using AI tools excessively or without practical need.
“Please don’t use AI just for the sake of using AI,” Treadwell reportedly told staff.
Amazon later confirmed that the dashboard had been removed, saying in a statement that the beta tool “was not a formal or approved tool, and has since been deprecated”.
The incident highlights the challenges large technology companies are increasingly facing as they attempt to accelerate AI adoption internally while also controlling the soaring costs associated with running advanced AI systems.
Some Amazon employees reportedly used Kiro alongside MeshClaw, an internal version of the open-source OpenClaw framework that allows AI agents to run tasks autonomously on personal hardware. According to the Financial Times, some workers allegedly used these systems to generate additional AI activity simply to increase their rankings and demonstrate higher levels of engagement with the technology.
The behaviour comes as Amazon intensifies pressure on staff to integrate AI tools into their daily workflows. Reports suggest that more than 80 per cent of the company’s developers are expected to use AI each week as part of broader internal targets tied to adoption.
The issue also reflects a wider trend across the technology industry, where companies are investing heavily in AI infrastructure while trying to ensure that usage remains efficient and productive.
Similar concerns have reportedly surfaced at Meta, where employees are said to have increased token consumption to improve their positions on internal AI adoption tables.
At Amazon, the financial stakes are particularly significant. The company is expected to spend around $200bn in capital expenditure this year, with the bulk of that investment directed towards AI technologies and data centre infrastructure.
Amazon also relies heavily on AI models developed by Anthropic, which has recently moved towards a consumption-based pricing structure instead of flat monthly fees, increasing the financial impact of heavy AI usage.
In response to the issues surrounding Kirorank, Amazon has reportedly started using alternative performance measures. Rather than focusing on raw token usage, the company is said to be tracking “normalised deployments”, a metric intended to assess whether developers are using AI tools to create useful and functional code rather than simply generating activity.
The decision to shut down the rankings system underlines the balancing act facing major tech companies as they encourage rapid AI adoption without incentivising wasteful behaviour that adds little practical value while significantly increasing computing costs.
According to the Financial Times, employees were informed this week that Amazon had taken offline “Kirorank”, a dashboard linked to its Kiro developer platform that ranked workers according to their AI-related activity. The system had been introduced to encourage wider use of AI tools among developers, but reportedly led to unintended behaviour as some employees sought to boost their scores through unnecessary activity.
The issue became tied to what staff internally referred to as “tokenmaxxing”, a practice in which workers deliberately inflated their consumption of AI tokens, the units of data processed by AI models, in order to appear more engaged with the company’s AI initiatives.
Amazon’s AI push sparks unintended behaviour
Dave Treadwell, Amazon senior vice-president, acknowledged the problem in comments to employees earlier this week, according to the Financial Times. He reportedly said the leaderboard had been created with “good intentions” but had ultimately contributed to higher operational costs because some workers were using AI tools excessively or without practical need.
“Please don’t use AI just for the sake of using AI,” Treadwell reportedly told staff.
Amazon later confirmed that the dashboard had been removed, saying in a statement that the beta tool “was not a formal or approved tool, and has since been deprecated”.
The incident highlights the challenges large technology companies are increasingly facing as they attempt to accelerate AI adoption internally while also controlling the soaring costs associated with running advanced AI systems.
Some Amazon employees reportedly used Kiro alongside MeshClaw, an internal version of the open-source OpenClaw framework that allows AI agents to run tasks autonomously on personal hardware. According to the Financial Times, some workers allegedly used these systems to generate additional AI activity simply to increase their rankings and demonstrate higher levels of engagement with the technology.
The behaviour comes as Amazon intensifies pressure on staff to integrate AI tools into their daily workflows. Reports suggest that more than 80 per cent of the company’s developers are expected to use AI each week as part of broader internal targets tied to adoption.
Growing AI costs reshape internal metrics
The issue also reflects a wider trend across the technology industry, where companies are investing heavily in AI infrastructure while trying to ensure that usage remains efficient and productive.
Similar concerns have reportedly surfaced at Meta, where employees are said to have increased token consumption to improve their positions on internal AI adoption tables.
At Amazon, the financial stakes are particularly significant. The company is expected to spend around $200bn in capital expenditure this year, with the bulk of that investment directed towards AI technologies and data centre infrastructure.
Amazon also relies heavily on AI models developed by Anthropic, which has recently moved towards a consumption-based pricing structure instead of flat monthly fees, increasing the financial impact of heavy AI usage.
In response to the issues surrounding Kirorank, Amazon has reportedly started using alternative performance measures. Rather than focusing on raw token usage, the company is said to be tracking “normalised deployments”, a metric intended to assess whether developers are using AI tools to create useful and functional code rather than simply generating activity.
The decision to shut down the rankings system underlines the balancing act facing major tech companies as they encourage rapid AI adoption without incentivising wasteful behaviour that adds little practical value while significantly increasing computing costs.
