The Rise of Agentic Workflows: AI That Understands, Decides & Codes
Bernhard Merkle
Software development is shifting toward a model where natural language becomes the primary interface for automation. This session offers a deep dive into GitHub Next’s Agentic Workflows, a research project that enables developers to describe software tasks in everyday language and let intelligent agents execute them autonomously. These workflows are context-aware, capable of decision-making, and able to take meaningful actions across repositories—far beyond traditional CI/CD or scripting.
We explore how Agentic Workflows ground themselves in repository context, plan actions, iterate through decision loops, and perform tasks such as refactoring, dependency updates, documentation improvements, and code generation. Participants will learn how Agentic Engineering enables more autonomous developer workflows and prepares the path toward emerging A2A (Agent-to-Agent) architectures. Practical use cases, architectural insights, and safety considerations round out the session.
As an active contributor to the gh-aw project, I’ve experienced firsthand how contribution models are changing: in an agentic era, you increasingly submit issues while agents fix bugs, implement features, and even improve their own workflows.
Bernhard Merkle
Bernhard Merkle works as a Software Specialist and Technology Scout for Gen AI, AI, ML, Cloud, XOps related technologies in the central Research & Development Department at SICK AG, one of the world’s leading producers of sensors and sensor solutions. He serves as internal consultant new Software Development Technologies and also designed and built up the DevOps infrastructure internally. In his spare time he gives a lecture about development with Models and AI. He writes technical articles and also gives sessions at various conferences (e.g. ACCU, Codegeneration, Conquest, QCon, OOPSLA, EclipseCon and OOP).