Uber’s AI Budget Blowout: How Claude Code is Redefining Software Engineering Costs
Uber is discovering that the road to hyper-efficient software development comes with a steep price tag. Although the company has seen significant gains in development speed, the rapid adoption of AI coding tools has pushed spending far beyond internal projections. Uber’s Chief Technology Officer, Praveen Neppalli Naga, recently admitted to The Information that the company’s AI spending assumptions have already been exceeded, forcing a complete rethink of their financial planning.
The Surge in AI Tool Adoption
The budget crisis is the result of an aggressive internal push to integrate AI into the developer workflow. In December, Uber granted 5,000 engineers access to Claude Code, an advanced tool developed by Anthropic. The uptake was immediate and explosive; by February, usage had nearly doubled.
Uber didn’t just provide the tools—they incentivized their employ. The company encouraged engineers to use Claude Code and Cursor, even implementing internal leaderboards to rank engineers based on their usage levels. While Cursor’s growth eventually plateaued, Claude Code became the dominant tool in the ecosystem. This surge in activity led Naga to state that he is “back to the drawing board” because the projected budget was “blown away already.”
The Shift Toward Agentic Software Engineering
The financial pressure stems from a fundamental shift in how Uber builds software. The company is moving away from simple AI assistants—which offer autocomplete suggestions—toward what Naga calls “agentic software engineering.”
In this new paradigm, AI systems aren’t just helping humans write code; they’re independently generating it with minimal human oversight. The impact is already visible in Uber’s infrastructure: approximately 11% of live backend code updates are now written by AI agents. These agents handle critical functions, including:
- Ride-matching algorithms
- Pricing adjustments
- Bug fixes
Naga’s long-term vision involves “agent engineers”—AI systems capable of fully handling the entire lifecycle of coding, testing, and deployment, with other AI tools acting as supervisors.
The Financial Toll of Innovation
The cost of this transition is reflected in Uber’s research and development spending. In 2025, Uber’s R&D expenses rose by 9%, reaching $3.4 billion. As the company expands its AI stack—including plans to test OpenAI’s Codex—these costs are expected to keep climbing.
This creates a complex tension for the company: AI is acting as a massive productivity lever, but it’s also becoming a primary cost driver. The scale of adoption has simply outpaced the company’s ability to forecast the expense of high-token-usage agentic tools.
Key Takeaways: Uber’s AI Transition
- Budget Exhaustion: Uber maxed out its full-year AI budget just a few months into the year due to high Claude Code usage.
- Rapid Scaling: Access started with 5,000 engineers in December, with usage doubling by February.
- Agentic Shift: 11% of backend code is now AI-generated, moving toward a model of “agent engineers” who handle deployment, and testing.
- R&D Growth: AI adoption contributed to a 9% increase in 2025 R&D spending, totaling $3.4 billion.
Frequently Asked Questions
What is “agentic software engineering”?
Unlike traditional AI assistants that suggest lines of code, agentic software engineering involves AI systems that can independently plan, write, and execute code updates with minimal human intervention.
Why did Uber’s AI budget blow out so quickly?
The combination of providing access to thousands of engineers, incentivizing usage through leaderboards, and the high computational cost of agentic tools like Claude Code led to spending that far exceeded initial forecasts.
Is Uber replacing engineers with AI?
While the company is moving toward “agent engineers” to handle coding and deployment, the current focus is on increasing speed and efficiency. However, the shift toward AI-driven backend updates raises questions about the future role of human engineers in the development pipeline.
Looking Ahead
Uber’s experience serves as a cautionary tale for enterprise AI adoption. While the productivity gains of agentic AI are undeniable, the cost models for these tools are volatile. As Uber integrates more tools like OpenAI’s Codex and refines its “agent engineer” workflow, the industry will be watching to see if the efficiency gains can eventually offset the massive surge in R&D spending.