Gisela Torres Unveils AI Agents for Hyper-Productive Development Workflows on GitHub

Microsoft Global Black Belt Gisela Torres presented advanced AI agent integration within GitHub and Visual Studio Code, emphasizing developer happiness and productivity. Her presentation centered on a retro-themed event agenda application, illustrating how AI agents—including GitHub Copilot, Claude, and Codex—can streamline complex development tasks. Torres demonstrated the critical practice of breaking down large projects into specific, actionable issues for AI agents, ensuring focused and efficient execution. A key highlight was the ability to assign the same task to multiple agents concurrently and monitor their progress via a dedicated “Agents” section in GitHub. Furthermore, an innovative in-IDE browser integration allowed agents to directly interact with and test the application, performing actions like activating UI elements or verifying responsive design, thereby automating tedious manual checks.

The session also delved into advanced capabilities like custom agents tailored for specialized tasks, such as mobile testing or generating documentation. Torres introduced “Agentic Workflows,” a new research-driven feature that leverages AI to enhance GitHub Actions. These workflows, defined as Markdown with YAML configurations, enable agents to automatically improve issue descriptions with relevant context, steps, and references, significantly boosting clarity and efficiency for development teams. The concept extended to AI-driven code reviews and a preview memory feature for agents to learn from past interactions. Finally, Torres underscored the importance of GitHub’s Advanced Security features—Dependency Graph, CodeQL Analysis, and Secret Protection—as integral components for maintaining secure and robust codebases in an AI-assisted development environment.