Why Amazon Has Dropped its Internal AI Usage Leaderboard

2026-06-07 · Source: AI Magazine · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Fundamental Awareness, short

Summary

Amazon has discontinued its internal AI usage leaderboard, Kirorank, after employees inflated token consumption by using AI for "pointless" tasks, leading to a spike in computing costs. The initiative, intended to encourage AI tool adoption, ranked staff based on usage, but Senior Vice President Dave Treadwell noted it encouraged "tokenmaxxing." This issue, also reported at Meta, highlights a common incentive-design failure. Amazon, which expects to spend US\$200bn on AI and data center infrastructure and has cut 30,000 corporate roles since October 2025, is now shifting its focus. The company will track "normalised deployments" – evidence of engineers creating useful code – rather than raw token consumption, aligning with an outcomes-first view echoed by Cognizant CEO Ravi Kumar S.

Key takeaway

For Directors of AI/ML or VP of Engineering designing AI adoption strategies, you should prioritize measuring tangible business outcomes over raw usage metrics. If your team is considering internal leaderboards, ensure they track "normalised deployments" or code shipped, not just token consumption, to avoid "tokenmaxxing" and escalating compute costs. Focus on defining high-value use cases and validating post-deployment impact to drive genuine productivity and value from your AI investments.

Key insights

Incentivizing AI usage by raw metrics like token consumption drives gaming, not value.

Principles

• Metrics must align with business outcomes.
• Public leaderboards can distort incentives.
• Adoption quality surpasses adoption quantity.

In practice

• Measure AI adoption by "normalised deployments."
• Define high-value AI use cases.
• Validate AI impact post-deployment.

Topics

• AI Adoption Metrics
• Incentive Design
• Compute Costs
• Generative AI
• Token Consumption
• Software Development Lifecycle

Best for: CTO, Executive, AI Product Manager, Director of AI/ML, VP of Engineering/Data, HR Professional

Related on AIssential

• Amazon engineers use MeshClaw bot to hit managers’ AI token targets — Corporate AI mandates can create perverse incentives, prioritizing usage metrics over genuine utility and security.
• Amazon kills internal AI leaderboard after employees gamed it with pointless tasks — Gamified AI usage metrics can inadvertently drive up costs and encourage unproductive activity, necessitating a focus on useful deployments.
• How to get your whole team using AI — Tracking internal AI token usage reveals adoption patterns and identifies top AI consumers.
• Enterprise AI adoption low due to high token usage, low RoI: Cognizant CEO - Business Standard — Enterprise AI adoption lags due to excessive token consumption lacking clear ROI and production value.
• 🎙️ How I AI: Quests, token leaderboards, and the elite AI adoption playbook & Notion’s spec-driven development — Gamified internal platforms and secure templates drive rapid, creative AI adoption across all organizational roles.
• Adoption Is Not Absorption — AI adoption metrics create false confidence; true value requires deep absorption into redesigned workflows and operating models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Magazine.