Amazon warns against tokenmaxxing, telling engineers that using more AI doesn’t necessarily mean they’re doing better work

Amazon has cautioned its engineers against an unchecked reliance on artificial intelligence, warning that such an approach may not always lead to improved work. In a memo, the company announced a shift in its guidelines for software developers, emphasizing the need for a balanced use of AI alongside human expertise. Terk Terkowitz, a vice president at Amazon, stated that engineers should document when AI is not beneficial for their tasks and share these findings with their peers and managers before seeking solutions.

The memo highlighted the potential pitfalls of indiscriminate AI use, such as the "tokenmaxxing" phenomenon where engineers focus on using AI tokens without consideration for the quality of work. Terkowitz criticized the practice of counting tokens and measuring lines of code as unreliable indicators of productivity, arguing that such metrics could lead to increased costs and unnecessary complexity, conflicting with Amazon's principle of "invent and simplify."

The company previously faced issues with employees using AI agents for trivial tasks to artificially boost their token counts and climb an internal leaderboard called Kirorank. This practice was discontinued after concerns were raised about the excessive computational costs. Amazon is not the only organization addressing AI best practices; Shopify's CEO recently cautioned that some workers are creating "slop grenades" - AI-generated emails and code that complicate matters for their colleagues.

However, Amazon remains committed to AI adoption, urging managers to carefully assess AI usage among their teams and focus on developing more effective metrics and goals.