Behavior follows incentives

Amazon sends message to employees, follows Microsoft

KiroRank scored Amazon employees on AI activity on the Kiro developer platform. Amazon had set a target for more than 80% of its developers to use AI tools weekly, according to [Fortune]

“Tokens are the units large language models use to process text and generate responses. Higher usage means higher compute costs.

What happened next was predictable in hindsight. Employees began inflating their scores through tokenmaxxing: running meaningless tasks through AI agents to consume tokens and climb the rankings.

Costs rose. Business value did not.

Who could have guessed that when you tell employees they have to use AI tokens to get ahead - that they would use AI tokens whether they were needed or not? Crazy world, huh?

7 Likes

Interesting

I am debugging the Play In Editor, PIE. The debugging runs very deep into the structure of the code. No way I could ever do this on my own. Claude had me chase down a problem. As I chased it down I had Claude examine six different sets of code. Claude called it off as a waste. Then settled out what we could do with the specific problem to switch to another problem that would solve the entire situation. I was impressed when Claude told me I was wasting resources.

2 Likes