Agents are coding like crazy, shifting the burden from creating code to trusting and validating code. Developers suddenly have more work in flight, more decisions to make, and more code to understand than ever. Unclear requirements, huge code changes, and slow reviews keep stalling as agents keep feeding the queue. The promised freedom from writing every line has become a new job managing everything around it. How do developers gain leverage from agents without allowing chaos - or becoming a bottleneck?

This keynote explores a GitHub-native approach to agentic engineering that puts explicit intent, bounded delegation, and trustworthy feedback at the center of delivery. Using GitHub and Copilot, we'll connect parallel agent work with small, reviewable changes, automated verification, and risk classification. We'll examine when developers should stay hands-on, when to delegate, and where product, design, architecture, and QA expertise belongs in the process. You'll leave with a practical framework for choosing what agents do, deciding which judgments remain human, and improving the engineering practices that turn generated code into software worth shipping.