Amazon is (once again) pleading with its workforce to stop using AI indiscriminately.
In an internal memo published to Amazon’s website on Tuesday, the company described “tokenmaxxing”—the once widespread business practice of encouraging employees to use AI as much as possible—as “deeply flawed” and “disconnected from what matters.” Too great an emphasis on employee inputs (i.e., token usage) rather than outputs, the memo said, can wind up driving up company costs and make some tasks more complex than they need to be.
That message flies in the face of the sales pitch coming from big tech companies (Amazon included), who have promoted AI products as an easy way for individuals and businesses to save time, and in some cases, even save money. (Meta, for example, has said its Muse AI agent can flag and cancel old subscriptions that users have forgotten about and are still paying for.) There’s been a wholehearted push over the past four years, backed by many billions of investor dollars, to integrate AI into virtually every consumer product. Businesses of all sizes, meanwhile, have faced pressure to experiment with AI and weave it into their daily workflows, or risk being left in the dust of other, more tech-savvy competitors.
Tokenmaxxing was the extreme manifestation of that business trend. For a brief period earlier this year, some tech companies introduced internal leaderboards tracking employees’ AI usage and incentivizing them to drive it up. (Dictionary.com cited the tokenmaxxing wave as one of the factors behind its decision to select “maxxing” as its 2026 word of the year.) But the AI spending bonanza soon ran into sobering reality checks. By April of this year, Uber software engineers had reportedly depleted the company’s entire annual AI budget on Claude Code. The following month, Amazon shut down an internal tokenmaxxing leaderboard, saying it encouraged AI usage among employees without necessarily improving business outcomes.
Echoing that earlier missive, Amazon’s new internal memo asks its software developers to tread a fine line: Experiment with AI to your heart’s content, but make sure you’re doing so in a genuinely helpful and productive way, and not just using the technology for its own sake. The company even flips the tokenmaxxing philosophy on its head, saying that moving forward it will reward employees for figuring out new and creative ways of removing AI from their workflows. “If your work does not benefit from current AI tooling, that is a finding, not a failing,” the memo read. “An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for.”
The memo—which was reportedly circulated internally last month—admits that much remains in flux, and that given the pace at which AI is advancing, Amazon will almost certainly continue to update its vision for what the day-to-day work of the modern software developer ought to look like. But it’s clear that it doesn’t consider tokenmaxxing to be the way of the future—and by extension, that its employees’ futures with the company will depend on their ability to be a little more discerning about their AI usage.
