In the first part of this blog series, I shared what agentic workflows are and how they integrate with GitHub. Now, it’s time to step away from theory and see the magic in action.

To demonstrate its potential in a real-world environment, I built a concrete demo: a Jira ticket that converts itself, 100% autonomously, into a pull request ready for review.

The idea is simple: you create a ticket, assign it, and that action triggers the execution of the Agentic Workflow. Minutes later, a draft PR appears in your repository without a single developer having to touch a line of code along the way.

‍

How Does an Agentic Workflow Work Inside?

When you create and assign the ticket, an automation rule in Jira fires a repository_dispatch event over to GitHub. This event wakes up the orchestrator workflow, which doesn't try to solve the task all at once. Instead, it breaks the work down into five phases, each delegated to a specialized sub-agent:

1. Context: Reads the Jira ticket using the Atlassian MCP and analyzes the repository files.
2. Design: Defines the main objective, user stories, and technical requirements.
3. Architecture: Makes the technical decision and generates the exact list of files to be modified.
4. Development: Creates a new branch in Git and edits the code following the established plan.
5. Quality: Verifies that requirements are met and allows up to 3 retries if something fails.
   ‍

If the quality sub-agent validates that everything aligns, the orchestrator opens a Pull Request in draft mode and automatically assigns Copilot as a reviewer.

‍

If you prefer to test it yourself in a controlled environment, the workflow can also be executed manually from your terminal by pointing to a specific ticket:

‍

# Trigger the orchestrator for a specific ticket on your board
gh aw run jira-orchestrator -f ticket_id=SCRUM-1

# SFollow the sub-agent logs in real time
gh aw logs jira-orchestrator

‍

The Plumbing Behind the Automation

Setting up this environment requires connecting a few cables: the Atlassian MCP (Model Context Protocol) so the AI can read Jira, the automation rule on your board to send the repository_dispatch, and configuring secrets in your repository (the Copilot token and Atlassian credentials). The detailed step-by-step for this configuration is fully available in the repository's README file.

The truly important takeaway from this demo: the agent handled the heavy lifting, but the final decision is always human. You inspect the run, review the changes in the generated PR, and if everything looks good, you hit merge. The human always closes the development lifecycle.

‍

Clone the Demo and Build Your Own Flow

All the code for this workflow—the orchestrator workflow, the five sub-agents, and the sample page—is available as open-source for you to fork, experiment with, and adapt to your team's needs. Explore it here (available in Spanish).

‍

Resources to Keep Learning

If you want to dive deeper into the AI-guided development ecosystem, I highly recommend checking out the following official links:

• Official Documentation: github.github.com/gh-aw
• Video - GitHub Agentic Workflows in Action: youtube.com/watch?v=TBVtoHrhFG4
• Video - Practical Use Cases: youtube.com/watch?v=kaIz0X_YByE

‍

‍