Amazon warns engineers against 'tokenmaxxing' as AI metrics face scrutiny

What's the deal? Amazon published a memo on Tuesday revising its software engineering guidelines, warning staff against "tokenmaxxing" — inflating AI usage to look productive. Terk Terkowitz, vice president and distinguished engineer at Worldwide Amazon Stores, called token counts and lines of code "deeply flawed" measures of productivity. "They are disconnected from what matters," he wrote.

Why now? The warning follows Amazon's May shutdown of an unofficial internal AI leaderboard, Kirorank, after employees ran AI agents on meaningless tasks to inflate their scores. The practice raised concerns about computing costs, which can climb fast.

What could go wrong? Terkowitz wrote that turning usage metrics into "targets" drives up costs — "tokenmaxxing is at odds with frugality" — and adds complexity that undermines Amazon's "invent and simplify" principle. Axios reported that one unnamed company accidentally spent $500 million on ClaudeDealroom has a profile for this one. Try Dealroom → in a single month after failing to cap employee usage.

What's the endgame? Amazon says it has entered "the era of the Renaissance Developer," blending AI with human judgment. The memo does not impose token caps, but pushes engineers to flag when AI falls short. "If your work does not benefit from current AI tooling, that is a finding, not a failing," Terkowitz wrote.

The signal: Amazon is not retreating from AI — it still wants the technology central to engineering — but correcting the incentives that treat easy-to-count activity as a proxy for productivity. It joins peers like Shopify, whose chief executive officer Toby Lütke recently warned of "slop grenades" of AI-generated work that burden colleagues. The shift marks a broader enterprise move from tracking AI adoption to asking whether it delivers results.

Image credit: Generated with Gemini

Read more: CNBC, tomshardware.com, fortune.com, thenextweb.com