GitHub Agentic Workflows: Automation That Actually Reads the Room | DEM350
Ari LiVigni and Alejandro Menocal demonstrate how GitHub Agentic Workflows use GitHub Actions to run an AI agent that can improve a repository end-to-end (issues, CI failures, docs, tests) and deliver a ready-to-review PR with minimal workflow setup.
Overview
GitHub Agentic Workflows are presented as a way to let a repo “improve itself” using:
- A simple markdown file to define the workflow intent
- One command to kick off the workflow
- GitHub Actions as the execution environment
- An AI agent that performs tasks like:
- Triage issues
- Fix CI failures
- Update documentation
- Improve tests
- Produce a pull request for human review
The session emphasizes reducing the need for complex YAML while still keeping the process safe and reviewable.
What the demo covers (session chapters)
- Overview and demo goal: what Agentic Workflows are and what the presenters aim to show.
- Benefits for automation and CI/CD: how agentic workflows fit into CI/CD and repo maintenance.
- Live demo: starting from a minimal workflow file and generating an automated PR for a fictitious “Mona” website.
- Setup and scaffolding: how the workflow scaffolding process works for creating workflows.
- Selecting an agentic workflow agent: choosing an agent to create a website update process.
- Modifiers: an example of customizing an agent so it always performs specific actions during workflow creation.
- Security and sandboxing: addressing restrictions and using a sandboxed pipeline during compilation/execution.
- Next steps: encouragement to experiment and a pointer to GitHub Skills.