Agentic SDLC at Uber — Uday Kiran Medisetty & Adam Huda, Uber

By ai.engineer

Categories: AI, Tools

Summary

Uber automated 70% of PRs through agentic AI, doubling lines of code per engineer yearly—achieved through a six-layer infrastructure (model gateway, MCP gateway, agentified DevPods, code execution, observability, and orchestration) that handles 100M+ daily model requests while reducing token costs by 40% fleetwide.

Key Takeaways

  1. Build a model gateway as your first critical infrastructure layer: route all AI requests through a single endpoint with PII redaction (20+ types), specialized safety models (5 variants), and sub-100ms latency for compliance and attribution per user/project/team.
  2. Create an MCP (Model Context Protocol) gateway to standardize tool access: consolidate thousands of internal APIs and SaaS tools (Slack, Jira, Google) into one installation method, saving 40% of tokens fleetwide through Omni MCP pattern and CLI-mode optimization.
  3. Pre-provision agentified environments (Kubernetes pods) with snapshotted repos and pre-built search indexes to enable agents to run longer workloads with sub-second startup times—essential for monorepo-scale engineering (millions of lines of code).
  4. Implement automated migrations at scale: Uber handled 250+ automated migrations touching 9 million lines of code automatically, enabled by six years of monorepo and Bazel investment creating the foundation for agentic workflows.
  5. Track and attribute all AI model requests in real-time through a data lake—enables spend guardrails, benchmarking loops, and continuous self-improvement across 800+ internal projects handling 100M+ requests daily.

Related topics

Transcript Excerpt

[music] Hey uh let's get started. Uh good morning everyone. I'm Udai. I'm here with my colleague Adam. We'll talk about our journey towards manage software factory. And in the beginning in the first part of the talk I'll talk about the key building blocks that we are investing in. And later Adam's going to talk about how we take all of these blocks to build an end to-end cohesive solution for our engineers. To set some context, we have few thousand engineers across 12 global tech sites. Over the last year, all of the investments we made in agentic AI have led to more than 70% of our PRs now either by local or cloud agents. And all of this led to twice the number of lines of code per engineer year-over-year. And this extends way beyond coding and we see it in every aspect of the engineering…

More from ai.engineer