Uber Burned Its 2026 AI Budget by April — The Agentic Cost Crisis

Short summary

Uber exhausted its 2026 AI budget by April because agentic workflows inflate API call volumes 8-15x beyond typical estimates, with later calls in a sequence costing 5-10x more due to context accumulation. Forrester reports 22% of agent deployments show negative ROI as infrastructure costs exceed productivity gains. A routing architecture that classifies tasks and directs them to the cheapest capable model can reduce costs by 70%.

• •Agent workflows create 8-12+ LLM calls per user request; context accumulation makes later calls 5-10x more expensive per token
• •22% of enterprise agent deployments now report negative ROI—infrastructure costs exceed anticipated productivity gains
• •Solution: routing layer that classifies task complexity and routes to cheapest models capable of correct outputs, reducing costs 70%+

Generated with AI, which can make mistakes.