Amazon tokenmaxxing is now discouraged for the company’s software engineers. In a memo circulated internally in September and posted to its website on October 27, Amazon said token counts, AI adoption rates, and code-change volume are deeply flawed productivity measures that can raise cost and complexity when treated as targets. The guidance revises how Amazon defines the roles of software development engineers and managers, and says teams should be judged on long-term outcomes.
Key Takeaways
- Amazon called token counts, AI adoption, and code volume deeply flawed measures.
- It shut an internal usage leaderboard in May after employees gamed rankings.
- Amazon has not capped tokens and wants long-term outcomes over input targets.
- Skipping AI where it does not help is a finding, not a failure.
Amazon Tokenmaxxing and the Metrics Problem
Tokenmaxxing means using AI as much as possible with little regard for output. CNBC reported that Amazon shut an internal leaderboard in May after employees gamed token consumption to climb the ranks. An Amazon spokesperson told CNBC the company has not capped token use.
Terk Terkowitz, vice president and distinguished engineer for Worldwide Amazon Stores, signed the memo, The Next Web reported. It said the memo states that “tokenmaxxing is at odds with frugality.” The Indian Express quoted Amazon: “Using these input metrics, with care and judgement, as you think about your career and your organization’s performance is appropriate.” Alongside that caution, the company wrote: “We believe that it is best to measure, and be measured by, long-term outcomes.”
Judgment for Engineers, Not a Usage Quota
This is not a call to drop the tools. The Next Web reported that the memo says AI will be “at the center of software engineering practice.” It says frontier teams, those that have embraced AI the most, show higher productivity and delivery, while teams and individuals not adopting AI are falling behind.
The memo asks managers to:
- Study the bottom 10% of teams by AI adoption and find out why.
- Check whether the top 10% deliver more to match.
- Use the tools on production work, and describe one task where AI helped and one where it did not.
“If your work does not benefit from current AI tooling, that is a finding, not a failing,” Amazon wrote. “An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for.”
Feature Velocity Over Token Targets
Marc Brooker, a vice president and distinguished engineer at Amazon Web Services, told CNBC Amazon has looked to feature velocity: “how many launches are we doing, how many successful changes are we making in production.” “Broadly, we think about this as a return on investment,” he said of the overall approach, not of that metric alone. “If there is an important problem to solve for our customers, important problem to solve for our business, we’re going to invest tools, time, tokens, human effort, etc. into those.”
“What we don’t want to see is wasted effort where we are going and spending those tokens solving problems that don’t matter and sometimes it’s hard to know, so we do need to have some appetite for experimentation,” Brooker said. He described investing tools, time, tokens, and human effort in important customer or business problems, while leaving room to test ideas when the payoff is unclear.
How Enterprises Are Curbing AI Spend
Mint cited an IBM report that tokenmaxxing peaked in spring 2026 and then waned, as high bills met a lack of visible improvement. In March 2026, Nvidia CEO Jensen Huang said he would be “deeply alarmed” if a software developer with a salary of $500,000 did not spend at least $250,000 on tokens each year. In May, Uber said it had spent its entire 2026 AI budget by the end of April. Its chief operating officer said that spending was getting “harder to justify” without a proven link to better customer features or developer productivity.
The Next Web reported that Uber’s tech chief said the tokenmaxxing era is ending, that Microsoft set division-level AI budgets in August, and that Accenture capped staff Claude spending at $29 a month. Token prices are falling, The Indian Express noted, but the cost of completing a task is rising as AI firms shift from flat subscriptions to usage-based pricing.
| Company | Reported step | Source detail |
|---|---|---|
| Amazon | Leaderboard closed; no caps | Board shut in May; caps not instituted |
| Uber | 2026 AI budget exhausted | Entire budget spent by end of April |
| Microsoft | Division-level AI budgets | Set in August |
| Accenture | $29 monthly Claude cap | Staff cap, 404 Media reported |
CEO Andy Jassy has said AI will “change the scope and speed” of day-to-day work. For enterprise leaders, the reading is to keep token and adoption figures as diagnostics and to judge teams on launches, successful production changes, and problems solved. Organizations already tracking an AI cost squeeze, or inference against sold-out cloud capacity, have little reason to pay for usage that does not change an outcome.
