Uber’s AI infra, from the inside

Uber’s engineering culture centers on building robust, custom infrastructure that enables seamless transitions into the AI era. Long-tenured engineer Uday Kiran Medisetty details how foundational systems like Michelangelo, Cadence, and Jaeger provided the necessary architecture to integrate agentic AI effectively. By leveraging these battle-tested tools, Uber developed the Model Gateway and MCP Gateway to standardize AI interactions and tool access. The company now utilizes "Minion" for automated coding tasks and the "Context Graph" to ground agents in organizational knowledge, significantly reducing the cognitive load for engineers. This infrastructure-first approach allows for sophisticated automated workflows, such as self-healing CI pipelines and capture-replay testing via Capri, ensuring that high-velocity feature shipping remains reliable. Ultimately, software engineering at Uber is shifting toward "context engineering," where developers focus on designing systems that empower agents to execute and verify complex tasks autonomously.