Amazon has a message for its software engineers: using more AI is not the same as doing better work.
The company published a memo laying out how its engineers and engineering managers should use AI tools. It explicitly discourages a habit that picked up a name this year: tokenmaxxing.
The memo first circulated internally in September. Amazon posted it to its website on Tuesday, turning an internal guidance note into a public statement on how one of the world’s largest employers of software engineers thinks about AI and productivity.
What the memo actually says
The core instruction is simple. Engineers should use AI selectively, and they should measure its value by long-term outcomes rather than by token usage.
A quick primer helps here. AI models process text in small chunks called tokens, and every prompt and response consumes them. Think of tokens as the electricity meter on an AI tool: the more you run it, the higher the reading climbs.
Tokenmaxxing is the practice of driving that meter as high as possible to look productive.
The memo also revises how Amazon defines the role of software development engineers and their managers now that AI tools are part of daily work. It introduces the idea of a “Renaissance Developer,” an engineer who combines AI tools with human judgment instead of handing everything to the machine.
Amazon framed the moment as early days for the whole discipline.
“We’re just seeing the beginning of AI’s effect on software engineering,” the company wrote.
The leaderboard that went away
The memo did not come out of nowhere. Amazon previously had an informal internal leaderboard called KiroRank, which ranked employees on their token consumption.
The problem is easy to spot. KiroRank rewarded engineers for how much AI they used, not for how well they solved problems. The leaderboard was deprecated within weeks of its introduction in 2026.
Amazon has tracked token usage internally for cost management. The company does not use those figures to evaluate employee performance, a distinction the new memo makes formal.
Senior Vice President Dave Treadwell had already flagged the issue. In May 2026, he advised staff to use AI to solve real customer and business problems, not to use it simply for the sake of using it. The memo effectively turns that warning into written policy.
How tokenmaxxing became a thing
Tokenmaxxing emerged earlier in 2026 at several Silicon Valley firms. Companies set ambitious AI adoption goals, then built internal dashboards to track progress toward them.
Those dashboards created a competitive atmosphere among developers. The incentives drifted toward consuming as many tokens as possible rather than making substantive contributions.
What this means for Amazon and the industry
The most immediate stakeholders are Amazon’s own engineers. The memo tells them their performance will be judged on what they deliver, not on how loudly they use AI to deliver it.
The cost angle matters too. AI infrastructure is expensive, and rising costs have pushed tech companies to rethink how they measure productivity. A workforce racing to burn tokens is a workforce quietly running up a compute bill. Amazon’s guidance signals that unchecked usage is not something it wants to subsidize in the name of looking modern.
The “Renaissance Developer” framing is worth watching as well. It positions human judgment as the scarce resource, with AI as an amplifier rather than a replacement. That is a notable stance from a company whose cloud unit sells AI tools to other businesses.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.