Amazon has shut down an internal leaderboard that ranked employees by how much they used its AI coding tool, Kiro. The system, called KiroRank, tied into Amazon’s internal company directory, PhoneTool. It showed badges next to employee names based on AI usage.
The company says the tool met its goal. Employees tell a different story.
Amazon said KiroRank helped drive early AI adoption. A company spokesperson described it as a beta dashboard that was never a formal or approved system. The tool has now been deprecated.
But workers reported deeper problems. They said the leaderboard was easy to game. Some employees inflated their AI activity to climb the rankings. Others felt pressure to use AI more, even when it did not improve their work.
The result was a new workplace behavior employees called “tokenmaxxing.”
The term comes from tokens, the units large language models use to process text. In simple terms, more prompts and longer outputs mean more token use. Under tokenmaxxing, employees chased higher AI numbers instead of better outcomes.
Some workers admitted they used AI agents to generate needless tasks just to increase usage. Rather than using Kiro to solve real engineering problems, they optimized for leaderboard placement.
One employee said they started cheating after receiving feedback in a performance review that they were not using AI enough.
The episode shows a growing tension inside tech companies. Leaders want workers to adopt AI tools. But once companies start measuring usage, employees often respond to the metric instead of the mission.
This problem is not new.
How AI Usage Incentives Can Drive Waste Over Value?
In business, people have long shaped behavior around targets. When companies reward calls made, workers make more calls. When they reward hours worked, people stay online longer. AI metrics create a similar risk. If usage becomes the signal for performance, employees may focus on maximizing prompts rather than producing better code.
That can create real costs.
AI systems are expensive to run. Every prompt consumes computing power. Heavy usage increases infrastructure spending. Employees familiar with KiroRank said the leaderboard encouraged wasteful AI use, which pushed costs higher without clear gains in productivity.
Amazon has acknowledged that it tracks token utilization, but says it does so to understand efficiency and cost patterns, not to monitor individual employee behavior. The company also said it does not require teams to use AI tools.
Still, the existence of KiroRank sent a message many employees understood clearly: AI usage mattered.
Amazon executives have tried to draw a line between useful adoption and empty activity.
Dave Treadwell, Amazon’s Senior Vice President, told staff not to use AI for its own sake.
“Please don’t use AI just for the sake of using AI,” he said. “Use AI to help you solve customer problems, to help you solve business problems, to innovate.”
That advice points to a broader shift happening across the tech industry.
Amazon and the AI Trap: Why Measuring Usage Over Value Kills Productivity
Over the past few years, companies rushed to roll out AI tools across engineering, support, research, and operations. The focus was speed and adoption. Getting people to use AI often mattered more than measuring whether it improved results.
Now the mood is changing.
As AI budgets grow, companies face pressure to show returns. Leaders want proof that AI saves time, improves products, or cuts costs. Simple usage numbers no longer look like enough.
The KiroRank story highlights the danger of treating AI activity as a performance metric. Measuring prompts, tokens, or tool usage can create distorted incentives. Employees may learn to feed the system rather than improve the work.
The lesson is simple: AI use is not the same as AI value.
A useful AI tool should help engineers ship better code, solve harder problems, or reduce repetitive work. If workers start creating extra tasks to raise usage numbers, the metric has failed.
Amazon’s decision to shut down KiroRank may reflect that reality.
The company wanted to encourage experimentation with AI coding tools. Instead, it discovered what many organizations learn when they measure the wrong thing: people optimize for the scoreboard.
And when the scoreboard rewards consumption over results, efficiency often loses.
