Key Notes
- Amazon’s engineering guidance asks teams to connect AI usage with useful outcomes instead of treating token counts as targets.
- Earlier reporting described staff assigning unnecessary tasks to agents, and Amazon confirmed in May that the employee-created KiroRank dashboard had been retired.
- The memo does not disclose a total cost or number of employees involved in tokenmaxxing, so those details remain unquantified.
Amazon is telling its software engineers to judge AI by what it helps them deliver. The warning follows reports that some employees assigned unnecessary work to agents to climb internal usage rankings, turning a push to adopt AI into a competition to consume more of it.
The company’s engineering guidance, published on October 27, warns that treating token counts and code activity as targets can increase costs and complexity. Usage data remains useful, but managers are expected to connect it to customer and business outcomes.
The behavior has a name: tokenmaxxing. Tokens are the units of data models process and generate, so counting them measures activity. It does not establish whether an engineer fixed a bug, improved a product or created another task for someone to review. A leaderboard can make that distinction surprisingly easy to overlook.
The Leaderboard Was Retired in May
The earlier accounts concerned MeshClaw, an internal agent platform that employees could use to automate workplace tasks. In May, the Financial Times reported that some staff gave it unnecessary work to inflate their usage scores. TechSpot’s account of that reporting described employees who believed managers were watching the numbers despite assurances that token statistics would not determine individual performance evaluations.
A separate employee-created dashboard, KiroRank, was subsequently retired. An Amazon spokesperson told Business Insider on May 29 that the informal tracker had been deprecated and was never meant to encourage AI use for its own sake. That confirmation establishes the dashboard’s removal; the accounts of workers deliberately gaming usage came from earlier reporting.
The chronology matters. KiroRank’s shutdown happened months ago. The development this week is Amazon’s public explanation of how it wants engineers to assess AI-assisted work. The published memo does not put a dollar figure on unnecessary token spending or quantify how many employees were involved.
Useful Work Is Harder to Count
The guidance also leaves room for engineers to explain when current AI tools do not help a task. Amazon calls that “a finding, not a failing.” Teams using AI heavily should be able to explain whether delivery is improving along with consumption.
That creates a more demanding management question than who used the most tokens this week. Imagine two agents tackling the same defect: one makes a small, verified change, while another generates several large revisions that still need substantial correction. Their activity totals can favor the second approach even when the first leaves the team closer to shipping a reliable fix.
Heavy usage can still be justified. A difficult investigation may require repeated experiments, long documents or several approaches before one succeeds. The useful comparison is between the resources spent and the result, including the work people must do afterward. A low token count by itself offers no guarantee of quality, either.
The incentive problem connects with FOMAT, the pressure to keep agents working while their human supervisors are away. That feeling turns idle machine time into a perceived missed opportunity. A usage ranking can add an organizational reason to keep the queue full, even when the next assignment has little value.
For an engineering team, a better discussion starts with what changed for the user: a problem resolved, a service made more reliable or a useful feature delivered. Tests, review time and the cost of correcting mistakes help explain that outcome. Token counts can inform the conversation, but a high score cannot finish it.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
