Yeda AI Tips · #33

Assign Your Backlog To Copilot, Not Yourself

Your backlog has forty issues and you have time for two. Most of the other thirty-eight aren't hard — they're just tedious: raise test coverage on a module nobody wants to touch, fix a small reported bug, do a mechanical refactor. GitHub Copilot's coding agent is built for exactly that gap: you assign it an issue the way you'd assign it to a teammate, and it works the task in the background while you do something more interesting.

The problem: synchronous work doesn't scale

Every issue in a backlog competes for the same resource — your attention, one task at a time. Even a genuinely simple fix has fixed overhead: switch context, read the issue, open the repo, make the change, open a PR, wait for review. When there are more small tasks than there is attention to spend on them, most of them just sit. Copilot's coding agent removes the "your attention" bottleneck for the subset of issues that are well-scoped enough for an agent to run with.

The fix: assign the issue to Copilot

Open a GitHub issue, click assign, and pick Copilot the way you'd pick a person. From that point, the work happens asynchronously: Copilot starts working the task in the background and, when it's done, opens a pull request for you to review. You can assign more than one issue at once — it works them independently, in parallel, rather than one at a time.

What actually happens after you assign it

Copilot doesn't disappear and reappear with a finished PR — the process is visible the whole way:

  1. A draft pull request appears almost immediately, before the work is finished. This isn't a placeholder — it's where Copilot posts its plan.
  2. The PR body updates live with a running checklist: Copilot states its plan for tackling the issue, then checks off each step as it completes it, so you can watch progress without pinging anyone.
  3. A session view shows the agent's actual tool calls — reading package.json to understand the project, running the test suite, running the linter, and in some setups, pulling additional context through an MCP server (for example, a linked internal doc). This is the same kind of transparency you'd want from a human contributor narrating their work in a PR description.

Triggering it from your editor

You don't have to go through the GitHub issues UI. From VS Code, highlight the relevant code and ask Copilot Chat to open a pull request for the change — Copilot hands you back a link to the resulting PR, without you leaving the editor. This is the faster path when you're already looking at the code and know exactly what needs to change, versus filing an issue first.

Iterating on the result

Treat the draft PR like you would one from a human contributor: leave review comments. Copilot picks up comments left on its own PR and keeps iterating — you don't need to reassign the issue or start a new session, just comment where you'd want changes and let it respond.

Where this earns its keep

The strongest fit is the backlog nobody has time for: boring tech debt, raising test coverage on a project that's been neglected, small well-defined bug fixes, mechanical refactors. It's a weaker fit for ambiguous, judgment-heavy work — issues where half the effort is deciding what to build, not just building it. Reserve the coding agent for tasks you could describe unambiguously in a couple of sentences; save your own time for the ones you can't.

Pitfalls to avoid

Takeaway

Assigning an issue to Copilot isn't "let the AI write my code" in the vague sense — it's a specific, observable workflow: assign, watch a draft PR appear with a live plan, review the diff, and comment if it needs adjusting. Use it to clear the backlog you'd never get around to yourself, and spend your own attention on the issues that actually need it.

Resources

Building an AI feature? Yeda AI designs, audits, and ships production LLM systems.

Talk to us · Read the blog