Adam Huda leads AI Developer Tools at Uber, where he designs infrastructure that lets autonomous agents help build, validate, and maintain software without overwhelming engineers or shared systems. His career spans Apple’s early iPhone build infrastructure, mobile startups, Twitter engineering, and the emerging agentic software factory.
After earning a master’s degree in computer science from the University of California, Santa Barbara, Huda joined Apple, working on OS X Leopard build infrastructure and the build system for iOS 2.0. He founded Think Top Down, co-founded the mobile-application company uApp, and led mobile engineering at Posterous before its acquisition by Twitter in 2012.
At Twitter, he worked on iOS product features, accessibility, mobile developer operations, feature switching, and internal engineering tools. He later led distributed teams at Supermassive. His engineering-management guide frames effective organizations as interconnected systems of hiring, team structure, decentralized leadership, execution, and feedback.
At Uber, Huda led Mobile Foundations and Developer Experience before moving into AI developer tools. He co-authored Uber’s account of its generative-AI hackathon, examining applications including automated testing, code review, refactoring, and documentation alongside privacy and security safeguards.
Making autonomous development operational
In Uber’s agentic software-development demonstration, Huda traces a proposed stadium-pickup feature from collaborative product research and design alternatives through implementation, review, and long-term upkeep. His most distinctive priorities include:
- End-to-end feature development. Uber’s Cortana assistant helps evaluate a product idea, sketch alternatives, and identify relevant existing code before handing implementation to the Minion cloud coding agent, which can make coordinated frontend and backend changes across repositories.
- Validation before continuous integration. Agents check their work before consuming shared CI capacity: static analysis, simulator screenshots compared with design specifications, frontend-backend integration checks, and preliminary code review. Pull requests then document completed checks so human reviewers can assess autonomous changes.
- Layered verification. Huda advocates combining complementary safeguards: unit and integration tests, automated code and security review, execution in realistic environments, and additional agents assigned to hunt for edge cases.
- Managed maintenance loops. Recurring agents can remove obsolete feature flags and handle similar cleanup, but their schedules and output must reflect CI availability and reviewers’ capacity. Accepted changes, rejected changes, and review comments provide feedback for improving future runs.
- Engineering judgment as a bottleneck. As agents accelerate implementation, constrained build infrastructure, experiment capacity, reviewer attention, and product decisions become more consequential. The difficult question shifts from whether a feature can be built to whether it should be built.
