Amazon employees are automating unnecessary tasks just to climb internal AI leaderboards.
The article "Tokenmaxxing" spreads at Amazon as employees game internal AI leaderboards appeared first on The Decoder.
"Tokenmaxxing" spreads at Amazon as employees game internal AI leaderboards
THE DECODER / 5/12/2026
💬 OpinionSignals & Early TrendsTools & Practical UsageIndustry & Market Moves
Key Points
- Amazon employees are reportedly “tokenmaxxing” by gaming internal AI leaderboards, focusing on optimizing outputs in ways that may not reflect genuine task value.
- The report links this behavior to the way internal systems and evaluation methods can be exploited when they reward certain measurable outcomes.
- Amazon is introducing automatic prompt optimization for its Bedrock AI service to reduce manual prompt engineering effort and improve performance by up to 22% depending on the task.
- The new Bedrock feature is available across multiple models (including Claude-3, Llama-3, Mistral, and Titan Text Premier), but the industry still struggles with reliably assessing whether automated prompt optimizations are truly beneficial.
- The article suggests a broader tension between automated optimization tooling and the difficulty of evaluating optimization quality in practice.
Related Articles
Black Hat USA
AI Business
Prompt Engineering for AI Agents: 7 Production Patterns That Beat Better Prompts
Dev.to
Scaling a Content Site to 5 Languages with Next.js + next-intl (Zero Manual Translation)
Dev.to
We started building MergeGuard to automate PR reviews directly inside GitHub.
Dev.to
Spec-Driven Development: Structure Beats Vibes
Dev.to
