Inside Uber's AI infra: Model Gateway, AIFX, and 50% token cost cuts

The Pragmatic Engineer · rss · 2026-10-15

The Pragmatic Engineer interviews Uber Distinguished Engineer Uday Kiran Medisetty on the company's dev tooling and AI infrastructure: a model-agnostic Model Gateway with AI Guard middleware, the AIFX wrapper that lets engineers switch between Claude Code/Codex/OpenCode without losing settings, hybrid AI spend tiers that cut token costs 50%, CI under 'code red' after jobs doubled in a year, PRs tagged by type as PR velocity loses meaning, and how the engineer's job has shifted to context engineering and verifying agent output.

Original post →

More from coding & agent

coding & agent channel →