Amazon halts Kirorank, its AI manipulated by employees

Amazon has recently decided to discontinue its internal ranking system based on artificial intelligence, known as "Kirorank." This decision comes after employees manipulated the tool to artificially boost their scores by using AI to perform tasks with no real added value. This behavior not only skewed the rankings but also led to increased cloud usage costs, according to a report from the Financial Times.

The Kirorank dashboard was intended to assess employee activity on Amazon's Kiro development platform. However, some workers began exploiting the system by directing AI agents towards trivial tasks. In response, Dave Treadwell, Amazon's senior vice president, reminded staff not to use AI unnecessarily. Although the tool was designed with good intentions, it ultimately resulted in additional expenses for the company.

This situation arises as Amazon has set an ambitious goal: to ensure that over 80% of its developers use AI on a weekly basis. The company plans to invest around $200 billion by 2026, primarily to strengthen its AI-related infrastructure. A similar phenomenon has been observed at Meta, where employees were also encouraged to maximize their use of AI.

To address these issues, Amazon has decided to shift its focus away from raw token consumption. Instead, the company is now tracking "normalized deployments," which refers to AI-generated code that has concrete utility.