Day 2 · Recorded 7 May 2026
Inside Uber's agent-ready platform behind 11% of merged PRs
If your AI program is still chasing novelty, Uber's story is a useful correction: start with the boring work that steals engineering focus. Nikhil Ramakrishnan walks through the developer-platform investments behind Minion, from remote environments and CI to migration tooling and review routing, and why those older bets became the substrate for useful background agents.
Nikhil Ramakrishnan, AI Foundations & Developer Experience, Uber
What's in this session
Uber's background agent work did not start from a blank slate. Years of investment in remote dev environments, cloud compute, CI infrastructure, and internal developer tooling created the conditions for agents to run real engineering work without forcing humans to babysit every step.
In this interview, Nikhil Ramakrishnan traces the line from an early container prototype to DevPod, then to Minion: a platform aimed at backgrounding toil tasks such as validation-heavy fixes, cleanup work, and other interruptions that take engineers away from product work. He also connects Minion to Shepard for large-scale migrations and Code Inbox for smarter review routing.
The lesson for platform leaders is that agent readiness is a compounding infrastructure problem. The teams that can give agents reproducible environments, trusted checks, migration primitives, and review workflows will turn background agents into a way to protect engineering focus, not just generate more diffs.
Inside the recording
- 00:00 Uber's compounding developer platform bets
Remote environments, CI, review tooling, and migrations become the base for background agents. - 03:00 Minion starts with toil, not novelty
The platform gives engineers back time by backgrounding tedious, high-rigor work. - 06:00 Rolling agents out where the pain lives
Use case selection starts with boring work engineers already know is slowing them down. - 09:00 Shepard and evergreen engineering
Large-scale migrations become a managed primitive instead of a recurring fire drill. - 12:00 Adoption by control, not mandate
Why builders need tools they can configure, trust, and mold to their own workflows.
More sessions on agent infrastructure
- Spectre: Harvey's collaborative cloud agent platform — Joey Wang, Harvey
- Enabling AI Tools Without Losing Control: Lessons from a Regulated Bank — Suhail Patel, Monzo
- What I learned building a software factory in public — Zacharias Malguitou, software-factory.dev