Amazon has officially shuttered its internal platform, Kirorank, after developers turned it into a testing ground for simulating productivity. As reported by the Financial Times, the dashboard—originally designed to track activity on the Kiro development platform—fell victim to gamification. Instead of driving real innovation, engineers began deploying AI agents to perform hollow, meaningless tasks solely to climb the internal rankings. The outcome was predictable: an explosion in artificial activity led to a direct financial drain, with AWS cloud infrastructure bills rising in lockstep with the volume of useless requests.
Key Takeaways from the Kirorank Incident
Engineers leveraged AI to automate trivial tasks in a bid to manipulate corporate rankings. Metric manipulation triggered a sharp spike in operational expenses for AWS infrastructure. Amazon leadership was forced to concede that quantitative KPIs are ineffective when implementing neural networks.
This failure exposes a deep rift between corporate ambitions and engineering reality. Amazon VP Dave Treadwell was forced to explicitly tell employees to stop using AI "for AI's sake." According to him, the dashboard creators' good intentions were crushed by the reality of unjustifiable expenses. The irony is palpable: Amazon is aggressively pushing a mandate for over 80% of its developers to use AI weekly—all while planning to pour $200 billion into AI infrastructure by 2026.
"Raw" activity metrics have nothing to do with high-quality technical outcomes; this obsession with usage figures has previously caused similar friction at Meta.
Amazon is now attempting to pivot toward "normalized deploys"—AI-generated code that is actually useful and integrated into products. Moving away from chasing token consumption and leaderboard rankings is an admission that quantitative AI adoption KPIs are toxic to both engineering culture and the company's P&L. As corporations feverishly justify massive capital expenditures on chips and energy, top-down pressure inevitably breeds "garbage" activity.
If even the world’s largest cloud provider cannot stop its own engineers from burning expensive compute cycles on automated busywork, then small business hopes for an accurate measurement of AI ROI seem naive at best.