How the GitHub Copilot coding agent works | GitHub Checkout

1,595 words · about 6 min to read

0:00 Andrea: I am so excited for the future where all these agents are doing my boring work for me. I’m Andrea Griffiths, Senior Developer Advocate, and today, we’re exploring something that’s going to change how you manage your development backlog. We’re diving into the world of GitHub Copilot coding agent. Welcome Tim. Tim: Thanks, Andrea. Andrea: So, what in the world is GitHub Copilot coding agent?

Delegating backlog issues to background agents

1 Highlight · Assigning multiple backlog issues to Copilot in the cloud

0:23 Tim: Well, really the goal with GitHub Copilot coding agent is to take all the great stuff we’ve already done with agent mode in the IDE that lets you have Copilot help you write and then test code, and turning that in something that isn’t just running on your machine, but something that you can run in the background. Andrea: Could we take a look at how it works? Tim: Yep, definitely. So here I’ve got my GitHub issues on github.com. And like any project, I’ve got a bunch of stuff here. There’s new features I want to add, bugs to fix, teach debt that I want to work on as well.

0:50 And that can get kind of overwhelming when you’ve got this big backlog. And the great thing with the coding agent is that I can pick the thing that’s maybe most interesting, most exciting, and most challenging, and I can have Copilot work in the background on some of the other stuff, and I can actually assign multiple issues at the same time. So, I’m just going to pick a bunch of issues here, click into the “Assign” menu, and then pick Copilot. Now, Copilot is going to pick all those issues up at once, work on them in the background, create a pull request, and then once it’s done, it’ll ping me for review.

1:18 Andrea: Wow. So you can move on and go work on that exciting new feature you want to add to your project, and Copilot is taking care of all the boring stuff?

1:24 Tim: Yeah. That’s right. And then when Copilot’s done, it’s going to kind of bring you in and ask you for feedback. Here’s an issue that I assigned to Copilot a bit earlier on today. And this issue is a bit interesting because it doesn’t actually have a lot of the details in the issue itself, and rather, it links out to Notion, which has the document that the product manager wrote with all the requirements. So, not everything we need

Tracking progress and leveraging external context

2 Highlight · Monitoring the agent via draft PRs, session views, and MCP

1:44 is in the issue. Earlier on, one of my team assigned that to Copilot and our pull request was created here. And we can click into that and we can see this pull request that Copilot has created for us. So, right at the beginning, as soon as you assign the task to Copilot, it’s going to go ahead and create that pull request as a draft and then it’s from that pull request that you can monitor the agent, understand what it’s doing and keep track of its work. So when it first gets started, we’re going to have a very basic pull request body and a very basic work-in-progress title that just says— basically has a copy of the issue.

2:17 Then as Copilot works, it’s going to gradually update that pull request body and say, “Here’s my plan. This is the stuff that I’m going to do.” And then gradually tick off things from that list as it works. So, I can really keep an eye on Copilot and know what it’s doing. Of course, you can also go into more detail and just like you have the chat logs in VS Code or in JetBrains, we also have a session view in GitHub where you can really see in detail what the agent is doing.

2:42 So I’ve just clicked into that here. We can see the agent talking to itself, and then we can see it using tools to basically understand the problem and explore the repo to figure out what code it needs to change, run tests to make sure that the code it’s writing actually works, and then when it’s done, it pushes and asks me for a review. So here we can see it’s looking at the package.json so it understands how the repo works. It’s looking at files. It’s running bash to find things.

3:09 Another interesting bit here is where it actually goes to Notion to find that document that I linked in the issue. So that’s powered by MCP. I can include model context protocol service with this agent that it can use to get other information and get other context from outside of GitHub. And actually from within GitHub with a GitHub MCP server. If we scroll a bit further down, then we’ll see the agent is even running my linter. So, at the end, when Copilot pushes and requests review from me, I’m not getting a PR that’s half-baked and broken, but most of the time I’m going to get a PR that’s good and ready to go, and hopefully I can merge it right away.

3:42 So, going back to the pull request here, we can see Copilot requested a review from me and then I can review that just like a PR from a member of my team.

3:49 I can decide to work on that myself, or I can leave comments. And Copilot is going to pick up those comments and do the stuff that I tell it to do. So it’s really easy to iterate. Andrea: Very cool. So you can keep control of what Copilot is working on. And then if it’s completely off-base, then guide it back to where you want it. Tim: Yeah, that’s right. Hopefully, what we could do with this is let developers do more of the stuff they really like,

Triggering background tasks directly from the editor

3 Highlight · Triggering background agents directly from VS Code chat

4:10 and let Copilot take over the boring stuff. Andrea: Fantastic. Okay. So what happens next? Tim: One of the other really nice things we can do with Copilot coding agent is actually trigger it straight from VS Code. So, here’s just an example of— this is a React component. And I’ve got a loading indicator here. And that’s quite a big block of HTML. So it’d be really nice to move that into its own component so it’s separate and abstracted out. I could do that now, but that’s going to distract me from my work.

4:36 So what I’m going to do instead is I’m going to highlight this, going to head over to Copilot Chat. And I can say, “Open a PR to refactor this into a separate component.” I send that off to Copilot and Copilot Chat is actually able to trigger this background workflow as well. So, obviously I’m used to working synchronously, directly with Copilot in Copilot Chat, but now we have the ability to actually say, “Can you go and do this thing for me in the background?” And it’s going to, in exactly the same way, open up a pull request and then tag me when it’s done.

5:07 And you can see here in chat, we’ve literally got the URL of that pull request. I can click onto it. And then from here I can see Copilot start to work. Andrea: That’s awesome. So while are you working and you don’t have to leave the editor to kick off another— Tim: Yeah, that’s right. You don’t have to get distracted. You can stay in the flow, stay on the stuff you’re working on, but also start to be in this world where you’re a developer working on your own thing, but you’ve got almost a team of AI interns who can be doing stuff for you

5:32 in the background and helping you to get more done, and hopefully, leaving the fun stuff for you. Andrea: This is amazing. How do you see teams using GitHub Copilot coding agent? Tim: Just an example, our billing team used it to increase their test coverage for their main projects. They didn’t have time to do that because of other priorities, but they could have AI do that for them in the background, and they just came in, gave it a review, maybe made some tweaks, and then they’ve got this project that’s a lot more maintainable and a lot more reliable for the future. Andrea: That’s amazing. So tech debt does not stand a chance.

5:59 Tim: I hope so. Hopefully this is going to be the end of tech debt. So at the moment this is available for people with Copilot Pro Plus or Copilot Enterprise. So that’s our higher premium plans, but we are looking at opening this out wider in the near future. So if you don’t have one of those, go ahead and upgrade, it’s a great opportunity to try this. But if you don’t, then watch this space and we’ll have something for you soon. Andrea: Thank you, Tim. I appreciate you coming by and showing us. I am so excited for the future where all these agents are doing my boring work for me.

6:27 Tim: No worries. I’m excited for that too. And for my open source projects, I hope they’re going to get a lot less tech debt in the future. Andrea: And that’s your first look at GitHub Copilot coding agent. Drop a comment below. Let us know how you’re using this in your repositories. And if this was helpful, please hit that “like” button and subscribe to GitHub’s YouTube channel for more feature updates and dev tips. Push these changes to main and we’ll catch you in the next release.