Table of Contents
Uber just burned through its entire 2026 AI budget—four months into the year.
The culprit? Rapid, unplanned adoption of agentic coding tools across its 5,000-engineer workforce. When Uber rolled out access to Claude Code in December 2025, usage nearly doubled by February. By March, 84% of developers were classified as agentic coding users, and today, roughly 70% of committed code comes from AI. Even with a $3.4 billion R&D budget, the company is now “back to the drawing board.”
This isn’t an Uber problem. It’s a warning for every enterprise scaling AI.
Why Traditional Budgets Break Down
Most enterprise software follows predictable pricing: per-seat licenses, annual contracts, usage tiers with clear ceilings. AI cost optimization wasn’t a discipline because costs were manageable by design.
Agentic AI tools don’t work that way.
Token-based pricing scales with activity, not headcount. When engineers run parallel agents, execute full-codebase refactors, or automate repetitive workflows, consumption compounds quickly. The productivity gains are real—but so is the spend. And unlike traditional software, there’s no natural ceiling.
Uber’s situation reveals what happens when adoption outpaces financial planning. The engineers weren’t doing anything wrong. They were doing exactly what the tools enabled. The gap wasn’t in usage—it was in visibility and control.
The Visibility Problem
For most enterprises, AI spend is a black box. Finance teams see invoices after the fact. IT leaders know tools are in use but can’t track consumption at the model, agent, or team level. Budget owners set annual forecasts based on assumptions that become obsolete the moment adoption accelerates.
This isn’t sustainable. As AI moves from experimentation to production, and from copilots to autonomous agents, cost unpredictability becomes a strategic liability.
What CIOs Should Do Now
Enterprises don’t need to slow down AI adoption. They need to govern it. That means building financial discipline into the infrastructure layer, not bolting it on after budgets blow up.
Effective AI cost optimization requires:
- Real-time cost visibility across every model, agent, and team so you can see spend as it happens, not months later
- Intelligent request routing that balances cost, accuracy, and latency, directing workloads to the right model for the job
- Automated financial guardrails enforced at runtime, setting limits that prevent overruns before they occur
- Forecasting tools that model spend scenarios before scaling, giving finance and IT a shared view of what’s coming
AI delivers transformational value. But that value erodes quickly when costs are unpredictable and uncontrolled. The enterprises that scale AI successfully will be the ones that treat cost optimization as a core capability, not an afterthought.
Uber’s experience is a signal, not an outlier. The question isn’t whether your AI costs will scale. It’s whether you’ll see it coming.
Explore how Airia brings financial discipline to AI at scale, or book a demo with Airia’s team to learn how we can help you build a secure, governed AI strategy.
