Amazon Tells Engineers to Stop 'Tokenmaxxing' for Productivity Scores

Amazon has told its software engineers that using more AI does not mean doing better work, warning in a memo published Tuesday that chasing token counts and lines of code is a "deeply flawed" way to measure productivity.

The memo, posted on Amazon's corporate news site, revises the company's guidelines for software engineers. It declares that Amazon has entered "the era of the Renaissance Developer" — an engineer who blends AI tools with their own expertise and judgment "to build entirely new things." Under the new guidelines, that includes telling management when AI is not useful to their work, and explaining why.

Terk Terkowitz, vice president and distinguished engineer at WorldWide Amazon Stores, wrote the memo. His central message: pushing AI use for its own sake can work against the company, not for it.

What does the memo actually say?

Terkowitz drew a hard line between useful reporting and quiet failure. His instructions to engineers are specific.

"If your work does not benefit from current AI tooling, that is a finding, not a failing," Terkowitz wrote. "Document why, share it with your community of peers, bring it to your manager, and work to find solutions. An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for."

The reasoning is economic as much as cultural. Terkowitz wrote that counting tokens and measuring lines of code do not necessarily reflect the quality of work, and that turning those metrics into "targets" could drive up costs. He put it bluntly: "tokenmaxxing is at odds with frugality."

Frugality is one of Amazon's long-standing leadership principles. Terkowitz also argued that gaming the metrics adds unnecessary complexity, which undermines another company principle, "invent and simplify." Of the raw adoption numbers, he wrote: "They are disconnected from what matters."

Why is Amazon worried about token counts?

The memo addresses a problem Amazon has already lived through. Employees were reportedly running AI agents on meaningless tasks to inflate their token counts and climb an unofficial internal AI leaderboard called Kirorank, according to the Financial Times.

Amazon shut the leaderboard down in May after the practice raised concerns about unnecessary computing costs — costs that add up quickly at enterprise scale. An unnamed company accidentally spent $500 million on Claude in a single month after failing to put limits on employees' usage, Axios reported.

Amazon is not alone in refining its AI message. Shopify CEO Toby Lütke, who earlier told employees in a memo that embracing AI is "a baseline expectation," said last month that some workers are producing "slop grenades" — AI-generated emails and code that create more work for the colleagues who have to review them.

Is Amazon backing away from AI metrics?

No. The company is recalibrating, not retreating. Terkowitz wrote that the adoption metrics still "correlate well with the outputs that matter."

He told managers to compare the 10th percentile and 90th percentile of AI users on their teams. For those using AI the least, managers should understand why. For those using it the most, they should determine whether output is actually rising in proportion.

"Using these input metrics, with care and judgement, as you think about your career and your organization's performance is appropriate," he wrote. "Working together, over time, we will find better metrics and set smarter goals."

The move signals a broader shift across large employers: two years after mandating AI adoption, companies are now writing the fine print on how to use it without burning money on inflated usage. Expect more memos like Terkowitz's as the cost of unmanaged AI access shows up on corporate invoices.