Model Gateway

Headlines · 1

1. 2d
   Inside Uber's AI infrastructure: Model Gateway, AIFX and agentic developer tooling

   In a Pragmatic Engineer podcast episode, Uber distinguished engineer Uday Kiran Medisetty describes the company's internal AI developer infrastructure. Uber built a Model Gateway in 2023 so AI applications stay model- and harness-agnostic, switching between Anthropic, OpenAI and other vendor models as well as open models Uber runs itself, with a security middleware called AI Guard, per-user spend tracking and audit logs layered on top. AIFX wraps Claude Code, Codex and OpenCode so settings, skills and plugins follow engineers when they change harnesses. AI spend uses allocated tiers with manager sign-off plus a CostCheck plugin for savings tips, and the source says token costs fell about 50% in recent months. Code review remains a bottleneck: Uber's Code Review tool runs AI review, attaches screenshots or videos and assigns risk scores before a human looks. After CI jobs more than doubled in 12 months, Uber declared an internal "code red" and moved simple jobs such as linting and builds into the inner loop, run by agents before a PR is opened.

   The Pragmatic Engineer · 🔥 0

Experience and discussion from the community

Share my Model Gateway experienceAsk about Model Gateway