Agentic Workflows

Building on the familiarity many already have with GitHub Copilot in VS Code, we will see how modern agents such as Claude Code, OpenAI Codex, and open-source alternatives like OpenCode and Pi have grown far more capable, using contextual awareness of your files, shell, and project to debug system issues, write clear research documentation, generate scientific diagrams in TikZ, and scaffold and fix LaTeX documents. Through live demonstrations, I will show how these tools close the diagnose–edit–test loop on real tasks, and how you can shape their behaviour with simple instruction files like AGENTS.md and CLAUDE.md and with Skills, the emerging open standard that teaches agents how to use tools (an evolution of the MCP concept). We will close with a brief, cautionary look at the enthusiast frontier — autonomous tools like OpenClaw and Hermes — highlighting their exciting ideas alongside their sharp edges, namely API cost and the real dangers of unsupervised operation. By the end of this session, you'll understand why agentic workflows are a genuine step-change and be ready to start building your own — safely — that same day.