Amazon published a memo to its software engineers on Tuesday, October 27, 2026, telling them that using more AI is not the same as doing better work, and warning against a habit it calls tokenmaxxing, or driving up AI token usage with little regard for the result. The memo, which circulated internally in September before Amazon posted it on its own website, revises how the company defines the roles of software development engineers and their managers in the age of AI. It calls token usage, AI tool adoption rates and code changes or reviews deeply flawed measures when used on their own, because they can raise costs and add complexity, and it says engineers should instead be measured by long-term outcomes.
What Tokenmaxxing Means
Tokens are the small chunks of text that AI models read and generate, and every prompt and response consumes them. Tokenmaxxing is the practice of maximising that consumption to look productive, for example by pointing an AI tool at work that does not need it, or by running it far more than the task requires. Amazon’s memo puts the problem in its own terms: tokenmaxxing is at odds with frugality, and measuring lines of code is at odds with inventing and simplifying. In other words, a metric that rewards volume will eventually produce volume, whether or not it produces value.
CNBC reported that Amazon shut down an internal leaderboard in May 2026 after discovering employees were tokenmaxxing to climb the ranks, and that an Amazon spokesperson said the company has not instituted caps on individual employees’ token usage. The memo itself does not ban the metrics outright. It says input measures such as token usage correlate well with the outputs that matter, and that teams and individuals who are not adopting AI or increasing output metrics are falling behind. The guidance is to use those metrics with care and judgement, paying special attention to the teams at the 10th percentile of adoption, who may need help, and those at the 90th percentile, who may be using AI heavily without matching delivery gains.
The Renaissance Developer
At the centre of the memo is a phrase Amazon wants to define the next era of its engineering: the Renaissance Developer. Written by Terk Terkowitz, vice president and distinguished engineer for Worldwide Amazon Stores, the memo describes this developer as someone who blends AI tools with human judgment and domain expertise to build entirely new things, rather than someone whose role or output is defined by how much AI they use. The memo says the most prolific engineers at Amazon have shown that neither the developer role nor their output is defined by the use of AI.
That framing comes with a practical test. Amazon says a leader’s understanding of AI should be hands-on: they should have used the tools on production work, and be able to describe a task where AI helped and one where it did not. The memo’s most useful line may be this: if your work does not benefit from current AI tooling, that is a finding, not a failing.
Outcomes Over Inputs
The memo argues that what is not changing matters more than what is: impact on customers, businesses and teams. Long-term outcomes can be delayed and hard to isolate from other trends, which is why companies reached for input metrics in the first place. Amazon’s answer is not to abandon measurement, but to connect inputs and outputs properly, for example by looking at feature velocity, meaning how many launches reach production successfully.
Companies that set ambitious AI adoption targets, then built dashboards to track them, created incentives that drifted toward consumption rather than contribution. Amazon is now saying, in public, that it would rather have engineers who use AI selectively and can defend that choice than engineers who use it everywhere and cannot explain why.
Why It Matters
This matters because Amazon is one of the world’s largest employers of software engineers, and its internal guidance often becomes a template for other companies. If the company that built much of the cloud now says, publicly, that token counts are a deeply flawed target, then other firms using similar dashboards will have to defend theirs. The memo also reframes a debate that has been framed too simply as humans versus AI: the real question is not whether engineers use AI, but whether they can judge when it helps, when it does not, and what it costs. That judgement, not the token count, is what Amazon says it wants to reward.
Background
Amazon’s memo says it is still early for AI in software engineering, quoting its own view that we are just seeing the beginning of AI’s effect on the discipline, and that whatever future best practices emerge, AI will be at the centre of software engineering practice. The company says it will continue to evolve the software development engineer role over the coming years. Enterprise AI vendors are making a parallel push into everyday work, and the controls around AI agents are becoming a competitive issue in their own right, while developer tools built for agents are trying to make AI-assisted work easier to govern. Amazon’s contribution to that debate is a cultural one: measure the outcome, not the meter.
Source: Amazon engineering memo
0 Comments