Summary
Uber engineers Uday Kiran Medisetty and Adam Huda detail how the company has built what they call a managed software factory powered by agentic AI, reaching a point where more than 70% of pull requests are now generated by local or cloud agents — doubling lines of code per engineer year-over-year. The talk, delivered at AI Engineer, covers six production building blocks: a model gateway handling over 100 million model requests per day across 800-plus internal projects with sub-100ms PII redaction and safety guardrails; an MCP gateway that unifies access to internal APIs and SaaS tools like Google, Slack, and Jira through a single Omni MCP endpoint; a plugin and skills marketplace with 2,500 skills and 20,000-plus daily executions; and a context graph with 150 node and edge types and 40 million entries spanning mobile, backend, data lake, design docs, and incident history.
The system has also handled 250 automated migrations covering 9 million lines of code. Adam Huda closes by walking through an end-to-end demo of feature development — from linear ticket to merged PR — using all six layers together. Each component is designed around enterprise constraints: per-user cost attribution, access-controlled tool discovery, and continuous evaluation loops feeding skill improvement back to authors.
For engineering leaders and AI platform teams, this is among the most detailed public disclosures of how a large technology company has operationalized agentic software development at production scale.
📺 Source: AI Engineer · Published August 21, 2026
🏷️ Format: Keynote Launch
