Uber’s AI hiring shift: fewer incremental hires, more code from agents

That changes the hiring equation. Instead of adding engineers every time Uber needs more software capacity, the company can try to add some of that capacity through AI coding agents, developer tools, automation, and compute. Khosrowshahi has also discussed the longer-term possibility of replacing some incremental engineering headcount with AI agents and GPUs, but the available reporting still points to a human-in-the-loop model today .

What AI is doing inside Uber engineering

The biggest shift is from AI as autocomplete to AI as an active participant in software delivery. Uber CTO Praveen Neppalli Naga said the company has “leaned in hard” to AI coding, that 95% of Uber engineers use AI tools every month, and that an internal AI agent is making about 1,800 code changes per week .

The company’s developer-productivity work appears to reach beyond code drafting. A Developer Productivity Engineering session described Uber’s AI effort as spanning the software development lifecycle, with work on coding-assistant customization for large monorepos, agentic systems for large-scale code migrations, and AI-powered testing and code-review workflows .

The control point remains review. Khosrowshahi said AI-generated code is checked by employees before it is added to a repository . In other words, Uber is using agents to produce and prepare more work, but it is not presenting that work as unsupervised production engineering.

Why “10% of code changes” is not the same as “10% of code”

Uber’s AI numbers describe different parts of the development process, and they should not be collapsed into one metric.

Khosrowshahi’s 10% figure refers to code changes produced by autonomous agents . Separately, The Pragmatic Engineer reported that 84% of Uber developers were agentic coding users, meaning they used command-line agents or made more agentic requests than simple tab-completion requests in an IDE . The same report said 65% to 72% of code was AI-generated inside IDE-based tools .

Those figures can all be true at the same time because they measure different things: autonomous-agent code changes, developer adoption of agentic workflows, and AI-generated code inside IDE tools. The practical takeaway is that AI may help draft a much larger share of code than the share of merged changes attributed specifically to autonomous agents .

How this reduces hiring needs

If engineers can ship more work with the same staffing level, Uber can grow engineering output without growing headcount at the same pace. That is the economic logic behind the company spending more on AI while hiring less .

The cost does not vanish; it shifts. Instead of only paying for additional employees, Uber is also paying for AI tools, agents, and compute. Reporting on Uber’s AI coding rollout said a surge in Claude Code usage exhausted the company’s 2026 AI coding budget earlier than expected, and that Uber has used tools including Claude Code and Cursor . That report should be treated as a snapshot of tool demand rather than a full accounting of Uber’s AI economics, but it illustrates the tradeoff: software capacity is increasingly being planned as a mix of people, agents, tooling, and infrastructure.

AI is moving into operations too

Uber’s AI strategy is not limited to engineering. Khosrowshahi said Uber has used AI for years to price ridesharing trips and match drivers with passengers . More recent reporting says generative AI and agentic AI are also being applied to customer support, driver onboarding, and parts of the engineering development lifecycle, reducing manual intervention in some workflows .

That matters because productivity gains outside coding can also reduce hiring pressure. If AI can speed support, simplify onboarding, or help diagnose internal-service issues, Uber can remove bottlenecks without necessarily adding the same number of people it would have needed before .

What this means for engineers

Uber’s current model points to supervised AI engineering, not a no-engineer model. Agents can draft code, prepare changes, support migrations, and assist with testing or review workflows, but human employees still review AI-written code before it is merged .

The likely impact is strongest on incremental headcount growth. Uber can keep expanding engineering capacity while hiring fewer additional employees than it otherwise might have needed, as long as AI tools produce reliable gains in real workflows . The open question is measurement: adoption rates and code-generation percentages show that AI is widely used, but they do not by themselves prove a precise productivity increase across quality, reliability, maintenance, and long-term engineering cost.