Amazon Memo Warns Engineers Against 'Tokenmaxxing' and Urges Selective Use of AI

Amazon has published guidelines on how its software engineers should use artificial intelligence, and the document pushes back against "tokenmaxxing," the practice of using AI as heavily as possible with little attention to the quality of the results.

The memo was circulated to staff internally in September and posted to the company's website on Tuesday. It updates how Amazon describes the roles of software development engineers and their managers as AI becomes part of daily work.

"We're just seeing the beginning of AI's effect on software engineering," the company wrote. "While we have a vision for what the future looks like, there's a lot that's still emerging. We are certain that whatever future best practices are, AI will be at the center of software engineering practice."

The guidance arrives as Amazon, under CEO Andy Jassy, has moved quickly to build AI into its wide-ranging operations. Jassy has previously said AI will "change the scope and speed" of everyday work and could reduce the company's corporate workforce over the next few years through automation, while also creating "other jobs."

Across the industry, businesses are wrestling with rising AI costs and with the difficulty of showing whether the tools deliver higher productivity or other worthwhile returns. Some firms have responded to tokenmaxxing by capping how many tokens employees can use. Tokens are the units of data AI models process and produce. Others have moved certain tasks to cheaper, open-source models.

An Amazon spokesperson said the company has not placed caps on employee token usage. In May, Amazon shut down an internal leaderboard after finding that employees were tokenmaxxing to move up the rankings.

The memo calls measures such as token usage, the pace of AI tool adoption, and the number of code changes or reviews "deeply flawed" ways to gauge how teams or organizations use AI, saying they can raise costs and add complexity. Even so, it does not tell staff to disregard them. Token usage and code generation should be weighed "with care and judgment" next to other data. "We believe that it is best to measure, and be measured by, long-term outcomes," the company wrote.

Marc Brooker, a vice president and distinguished engineer at Amazon Web Services, told CNBC the company has looked at "feature velocity," which tracks things like "how many launches are we doing, how many successful changes are we making in production." He described the approach as a return on investment question. "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," he said.

Brooker added that Amazon wants to avoid wasted effort on problems that do not matter, though that can be hard to judge in advance, so the company needs "some appetite for experimentation."

The memo also tells employees not to force AI into tasks where it does not fit. "If your work does not benefit from current AI tooling, that is a finding, not a failing," Amazon wrote, adding that an engineer who can explain where AI does not yet help "is demonstrating the judgement we are asking for."