Uday Kiran Medisetty

An earlier Uber engineering article, co-authored by Medisetty, traces the evolution of its API layer. His author biography describes moving mobile APIs from polling toward bidirectional streaming and leading a rewrite of the fulfillment architecture. The gateway separates concerns that had accumulated in larger services and provides a common way to manage APIs. That history helps explain why agent integration at a large company depends on discoverable, operable internal services as well as the language model.[3]

Infrastructure that agents can use

In his Pragmatic Engineer interview, Medisetty explains how Uber’s existing developer tools support coding agents. A model gateway allows teams to switch providers while retaining shared guardrails, cost tracking and logs. Tool access runs through a gateway, while prepared development environments let agents build and test code. The interview distinguishes generating changes from accepting them: engineers still review the results, aided by automated reviews, risk information and evidence such as screenshots.[2]

Resilience with differentiated capacity

Medisetty is a co-author of Uber’s Failover Architecture paper. The work replaces a uniform approach, in which every service keeps enough capacity in two regions for the full global load, with one based on service criticality. Less-critical workloads use spare failover capacity during ordinary operation and can be preempted when a region fails. The paper describes dependency checks and safeguards to keep critical services working when those other services are absent. Its measurements apply to the Uber deployment studied.[4]