Amazon warns engineers against ‘tokenmaxxing’, says metrics on AI tool usage and adoption ‘deeply flawed’

Amazon cautioned employees against tokenmaxxing, saying token usage and AI adoption rates are flawed measures of productivity. Amazon urged engineers to use AI where it helps and assess its value alongside other data, emphasizing long-term outcomes.

As companies around the world continue to embrace artificial intelligence tools and integrate them into their workflow, Amazon has warned its employees against tokenmaxxing, claiming that looking at AI adoption through these metrics is ‘deeply flawed’. Tokenmaxxing refers to the aggressive usage of AI tokens by coding agents or employees to complete tasks.

Tokenmaxxing - from badge of honor to liability

Until recently, as companies pushed employees to use more AI tools in their daily tasks, tokenmaxxing was seen as a badge of honor, with companies like Meta, Amazon and Uber having internal leaderboards showing token consumption. In March 2026, Nvidia CEO Jensen Huang made headlines after stating on a podcast appearance that he’d be “deeply alarmed” if a software developer with a salary of $500,000 did not spend at least $250,000 on tokens each year.

However, after they were hit with massive AI bills, several of these companies have walked back on the aggressive AI push. According to CNBC, Amazon laid out its AI usage guidelines for software engineers in an internal memo to staffers in September and posted it to the company’s website on Tuesday.

What Amazon said

“We’re just seeing the beginning of AI’s effect on software engineering,” the memo stated, according to CNBC. “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.”

Amazon said metrics like token usage, the rate of adoption of AI tools, and code changes or reviews remain “deeply flawed” ways of measuring teams' or organizations’ AI usage, adding that they can lead to higher costs and too much complexity.

Bobins Vayalil Abraham