Transcript
You can imagine how much value you can get from something like this. And I also love this scenario because it's GitHub data. So it's showing the power of using GitHub for these types of things. Megan, welcome to GitHub Checkout. Thanks for being here. Thanks for having me. I'm very excited. I'm super excited too, because we are truly meeting developers where they are. And this feature that you bring us today, I think is going to be, like LLMs love to say, a game changer for the way that teams work together and how they use Copilot. So please, could you give us a little bit of an intro as to what it is that you're showing us today? Today, I'm going to be working on a fitness booking app, just as an example of what you can do with this. The first thing I'm doing is I'm inviting GitHub to the channel and you'll see, it says it's an agent. So it's very clear and GitHub introduces itself. Hey, I'm Copilot, mention me in the thread and I'll start a session to help with your code. So I did a quick little setup before we joined, which is just choosing the repository. So you can see, this is my fitness class booking app. And then from there, I say, GitHub, can you create five issues for new features on the app from looking at the repo? So from there, it gives me an implementation plan, which has the steps it's going through so I can track its progress as it's going along and returns five issues that it's come up with to extend that further. I read through the list and I thought that this create class detail and instructor profiles was the most interesting based on what we've heard from users. And so I said, okay, GitHub, let's start on the number 13 issue. And from there, it created a Slack code channel, which allows me to just iterate back and forth with the agent. There's some interesting things that it'll show here. It'll show a diff view, as well as render the HTML of the app. And so you can see now at the bottom, I have the instructor's name and a little bit more clarity on what studio it is. Well, well, well. So not only were you able to get Copilot to IDA, create the issues, assign the issue, but then you can actually do a preview. And this is all happening within Slack. We've not let Slack. Exactly. And so that's great for people who are already in Slack, right? Obviously a lot of us internally are using the GitHub Copilot app or the Copilot CLI for a lot of this, but the game changer is going to be the interaction with other teams. And so now I could tag legal and say, hey, can you review this? I could tag our design team and say, how does this fit our design standards? And we are all working on it together in like a shared form of work instead of like locally hidden away in a silo. And then I have to like copy and paste back and forth from the agent. Can you tell me a little bit about the code channel? Because that's a really interesting way. Cause you started a conversation, but then this is sort of like a sub conversation that's only relevant to that issue or how do you explain code channel? So this is a new Slack capability and it essentially is a channel that an agent can mutate. And so there's a couple of features that's different about it than a normal channel. One, we give the agent the ability to open a right side panel for various things. It can render the HTML like I showed, render the diff. It can even show files there and you can share files with your team. But also it allows for more moving like the noisiness of working with an agent in Slack into a more concentrated place. And so you could imagine if I had done this without the Slack code channel, it would have posted all of that iteration in this thread. And there might be a lot of other people on the thread that I'm kind of a little bit spamming. And so the advantage is it's like putting it in its own area where you can go back and forth and only invite those people that need to know, as well as all these benefits of being able to see what the agent's doing. And if I show the code channel here, there's some meta information that's different than a normal channel. It shows me the repo. It shows me the branch. If we had a pull request open, it would also link to the pull request and then the session itself on Copilot as well as the model, which in this case is auto. So it has like a different treatment that's very agent native and makes working with agents a little bit more clear versus the normal Slack context. There's also a status. So right now it says idle, but if the agent needed me to like confirm something, like if it said, hey, is it okay if I read this file or do this operation? It would say needs attention. If it was working, it would say working. And so it kind of allows you to like manage all of your agents inside Slack. Megan, this looks amazing. And I'm so excited because I get to use it, but what if I'm a developer that does not work in Slack? What if I work on Microsoft Teams, for example? Yeah, we're really excited to support the same functionality inside Microsoft Teams, specifically the GitHub agent that has the full power of Copilot. I've come in here and asked GitHub a analytics question, which is show me the merged PRs over time by language. And then the agent picks it up and has this to do plan as it works through all of the steps in order to generate that chart and kind of walks me through its logic. It specifically looked at the language that the PR touched and then broke that down for me. It also does these insights so that there's some summary of what, you know, like May was the biggest month, et cetera. And then built me this chart that's just right inside Teams for the Copilot SDK repo, which is an open source repo, showing both the by language per month bar graph as well as a line graph. So you can see it cumulatively. This is so cool. You can imagine how much value you can get from something like this. And I also love this scenario because it's GitHub data. So it's showing the power of using GitHub for these types of things. So it works in Teams basically the same way. You're asking the question, Copilot is kicking off a process, and then you can also create issues. The same thing that you show us earlier, right? Like when you're working more specifically with a particular repo, like within a project. It functionally is the same as Slack on the GitHub side. It's just a different UI that you're interacting with it in. Yeah, so you can see that GitHub has created a test issue just because I wanted to show that functionality and it's gone ahead. It's hyperlinks to the repo. And I can also open my session in GitHub. This is amazing. So then if, let's see, you have a conversation with Mario. If Mario wanted to give some input to the agent to either steer the issue a different direction, would it listen to him? Like, would he be able to just comment or how does that work? Yes, exactly. And so it's completely multiplayer and we can both steer the same agent, which I think is the beauty of this integration. We're all working together now with the same shared context. And so I can show an example. I can jump onto Mario's thread here and say, okay, show me in graph the top experts. And it'll use the same agent session and just steer it. So this little googly eyes means Copilot's got it. Yeah, which I actually, I love that UX of, okay, I know that it's seeing what I'm writing and that it's thinking instead of it just kind of being a little bit of a delay. It updates me that it's now working on it. It gives me this great to-do plan where I can watch it go through each of the steps. Are we planning on bringing this anywhere else or Slack and Teams like the only two channels that we're thinking of supporting? The GitHub product that this is using is called GitHub Copilot Cloud Agent. And we have integrations with Linear, Jira, the Slack and Teams ones, but then also in github.com you can use Copilot Cloud Agent. So one of the things that these integrations have kind of forced us to think about is how to do multiplayer. And we're working on bringing that functionality to github.com as well. So you could imagine that you and I just want to work on a session in GitHub already. And you can do that today on a PR. You and I can go back and forth and be like, Copilot do this, do that, which is great. But we also want to light up scenarios where we want to do things like this data analysis. It doesn't need to be tied to a PR. That's beautiful. We have this graph of our top contributors by commit to the Copilot SDK repo. And I actually love this gradient. The agent's really getting creative with it. Can you say a little bit about like, for the folks who have already set up like Copilot instructions in their accounts, like does it respect those instructions or how does it work with like our organizational settings per se? Today, it mainly works off of repo settings. So if you have Copilot instructions or even agents.md, those instructions will be pulled into this because it is effectively running the Copilot Cloud Agent, which is already specified to just use that those repo instructions. And similarly, like if you manage your firewall rules on a repo basis for the Copilot Cloud Agent, like if you have the firewall locked down or if you have it open, those settings are also respected. And we also have enterprise managed settings, which is a layer higher than that, which is largely about governance, like choosing what models are allowed, all of that flows through this. So you get quite a bit of controls around how it's used. Hi, I'm looking forward to using this both in Teams and Slack. I know that you just launched this and it was a great collab with both the Slack and Microsoft Teams team. What's gonna be next? Like what's in the roadmap for this amazing new capability of Copilot Cloud Agent? There's a lot we wanna do. This is very much like V1 of many versions to come. One of the things that will be really exciting for our users is being able to add MCPs. That's something we haven't opened up yet but it's obviously available in the rest of the Copilot clients. So we're hoping to add that, which will allow for scenarios where you can link up your data warehouse and ask questions about your internal data or Datadog, other services that you might use in your day-to-day work. So that will be exciting. We're also hoping to expand to more multi-repo scenarios. There's some security challenges with doing that. Like you wanna be really careful about what access your agents has, but it will be great to be able to have in a channel multiple repos specified. We support that today, but you have to let GitHub know you're changing repos versus being able to do work across three repos and open a stack PR or something like that. So that's a use case that we wanna focus on. And then we also wanna focus on how we bring memory into this and automations to light up certain scenarios where you might say, hey, every day in the Slack channel, post the three most important issues we have or the three PRs that really need a review. And so we're excited about being able to control CCA automations, which exists inside github.com, but throughout Slack and Teams. That's gonna be so great. I'm thinking of Teams that can already use Slack and Teams to like do their triage or like their standups. And this is just going to bring the power of the cloud agent right there where they need it. In our launch last week, we had the Teams integration has meeting support, which is really exciting. So inside of a meeting, you can be doing a standup and say, at github update this issue. And we're hoping on the team side to be able to expand to the meetings integration so that inside of a meeting, you can access the live transcript and things like that. What do I need to do to get my organization to enable this? Yeah, so for both the Slack and Teams integrations, you'll need to install an updated app. And so you can find that in the Teams marketplace or in the Slack marketplace. And once your Slack or Teams admin installs the updated app, everyone should get access to it. Users accessing it for the first time have to authorize GitHub just via their user account. And then they need to just set the repo and you're away and everything will work great. Admins will also need to make sure that the Copilot coding agent enterprise policy is enabled. And then you should be able to use it. And hopefully you do use it and give us a lot of feedback. Megan, thank you so much. I am so excited for us to help developers bring the context where it matters, have these conversations in an order that makes sense and the possibilities of the future, being able to have NCP support and multiplayer. I'm excited, I'm excited. Thank you so much for showing us this today. Thank you for having me. And that was your first look at Copilot Cloud Agent, your human and your AI agent teams working together in Teams and Slack. Try it out, let Megan and the team know how it's working out for you. And don't forget to like and subscribe to this channel. Push those changes to main and I'll catch you on the next release.
