Git is a great memory system. But GitHub? That’s where the work actually happens.
I wrote recently about git as memory for AI—how version control captures the cognitive trace of human-AI collaboration. But git alone is just storage. GitHub adds the primitives that turn storage into workflow: Issues, PRs, Actions, and now Copilot.
In my experience, the delta between git and GitHub is where agentic work becomes possible.
What GitHub Adds Beyond Git
Issues: Structured Work Tracking
Git commits tell you what changed. Issues tell you what needs to happen next—and more importantly, why. A commit says “updated styles.” An issue says “we need to update styles because the brand guidelines changed, and here’s what needs to change and who owns it.”
- Assignment and ownership: Who’s working on this? (Including AI agents)
- Status tracking: Open, closed, in progress—visible state management
- Labels and milestones: Prioritization without a separate tool
- Templates: Standardize how work gets captured
- Relationships: Link issues to commits, PRs, and each other
I file issues for everything now—content ideas, research tasks, things to explore. Each issue is a work item that Copilot can see, reference, and help execute.
More and more, I use GitHub’s new agent tasks feature to drive my workflow—all from my phone. Sometimes I don’t even file an issue; I just create an agent task directly, and the work happens while I’m away from my desk.
PRs: Review and Quality Gates
Git branches are just pointers. PRs are review workflows with:
- Code review comments: Context that persists with the code
- Required approvals: Human oversight before changes land
- CI integration: Automated validation before merge
- Draft state: Work in progress that’s visible but not ready
- Most importantly: Access to GitHub Copilot and any custom agents I’ve written
For content work, every post goes through a PR. Copilot drafts, I review, the agent revises based on feedback. The PR captures the entire editorial process.
When I review, that’s my “1:1 time” with the agent. Sometimes, a surprise—the agent wrote something I disagree with—becomes not just an opportunity to fix the current PR, but also a chance to update the agent definition with the lesson. “Remember for next time: we don’t use this phrase.” This is the flywheel of agentic work. Each review makes the agent better, which makes the next review faster, which creates more bandwidth for higher-level thinking.
Actions: Automated Workflows
This is where GitHub becomes an orchestration platform. Actions let you:
- Trigger on events: New issue? Run a workflow. PR merged? Deploy content.
- Chain operations: Lint → build → test → deploy, all automated
- Schedule work: Daily builds, weekly reports, timed publishing
- Integrate agents: Copilot can trigger and respond to Actions
I use Actions to publish content, run quality checks, and automate repetitive tasks. The agent doesn’t just draft—it can kick off entire workflows.
Copilot: Native AI Integration
The agent isn’t a separate tool—it’s part of the platform:
- Context-aware: Copilot sees issues, PRs, commits, and code together
- Actionable: It can create issues, draft PRs, suggest changes
- Persistent: The work it does becomes part of the repo’s history
- Conversational: Brainstorm in context, then execute
“Summarize my work over the last 3 months” actually works when your work lives in commits, issues, and PRs. What if you could start your next performance review draft by asking an agent to summarize the activity that’s already in a GitHub repo?
The Workflow
Here’s what my agentic workflow looks like:
- Work dispatched → I describe what I want via an agent task or GitHub Issue
- Agent executes → Copilot produces a PR based on the task description
- Review cycle → PR with comments, revisions, approval
- Publish → Actions trigger on merge to deploy content
- History preserved → Everything queryable for future work
This isn’t git with a web UI. It’s a complete work orchestration system with an AI collaborator built in.
With agent tasks, I just describe what I want, select which custom agent to use, and Copilot produces a PR. I can inspect the agent session to peek into the agent’s internal dialog, and use that to adjust agent definition files if the internal reasoning was faulty. It’s important to review agent sessions always, especially when starting out with your agentic system: you want to make sure that your agent not only arrives at the right conclusions, but also arrives at them because of the right reasons and following the right reasoning.
The Deeper Insight
There’s something more profound happening here than just better organization.
Organizing work has always been information work itself. As developers, we’re information workers—but so is the act of managing tasks, tracking context, and coordinating efforts. That meta-work used to be pure overhead.
Now, with agents like Copilot integrated into the workflow, something shifts: the agent becomes the information worker, and the human becomes the manager.
I file an issue. Copilot helps draft the solution. I review and approve. The agent executes; I direct.
This is the real definition of frontier work. Not humans replaced by AI, but humans promoted to managers of digital workers. The information work doesn’t disappear—it gets delegated to agents who can actually handle it at scale.
GitHub makes this transition natural because it was already designed for collaborative work tracking. Adding an AI collaborator feels like adding a team member, not bolting on a tool.
The Realization
The best organization system is the one that’s already part of your development workflow. I was looking for productivity tools when I already had one—I just wasn’t using it for non-code work.
If you’re already using GitHub for code, consider using it for everything else too. The structure is there. The AI integration is there. And the portability means you’re never locked in.
What tools do you use to organize work that also integrate well with AI assistants? Connect with me on LinkedIn to share your approach.