Uber cut customer service jobs and blamed AI. Then it revealed its own token bill.

Uber just eliminated 10% of its customer service workforce.

First time the company ever linked layoffs directly to artificial intelligence.

VP Megha Yethadka told her team: "Our organization has become too complex and siloed."

But here's the number that matters.

Uber burned through its entire 2026 Claude Code budget in four months.

Per-engineer bills ran $150 to $250 a month. Heavy users hit $2,000. CTO Praveen Neppalli Naga spent $1,200 worth of tokens during a single two-hour demo.

They've now capped spending at $1,500 per employee per month for each agentic coding tool.

COO Andrew Macdonald is openly skeptical. He says it's hard to draw a line between rising token consumption and useful features shipped to riders and drivers.

He warned that token costs would eventually be weighed against headcount.

That tradeoff just became real.

The "agentic pods" program embedded 30 AI-proficient engineers across finance, legal, and HR. It cut financial pacing reports from two days to ten minutes.

But the broader math doesn't balance yet. You're cutting human jobs while running up token bills you can't fully justify. The COO is telling you the ROI is fuzzy. And you're still hiring 500 people including robotaxi engineers.

This is the pattern at every enterprise running agentic AI right now.

Audit your token spend per department. If your COO isn't asking hard questions about cost-per-decision, start asking them yourself.

SOURCE: https://www.bloomberg.com/news/articles/2026-07-22/uber-cuts-10-of-customer-service-jobs-citing-embrace-of-ai
VERIFIED: Bloomberg, The Verge, Engadget, Economic Times
SIGNAL: Uber's internal token burn rate exposed the gap between AI efficiency promises and actual ROI. The COO's skepticism about cost-per-decision is the conversation happening inside every enterprise running agentic tools.