The biggest announcements from GitHub Universe

[Music] Woo! [Music] Woo! [Music] Heat. Heat. [Music] Woo! [Music] Woo! [Music] [Music] Woo! [Music] Hey! Heat! Hey, what is up chat? What's up everybody? Welcome to the VS Code live stream and GitHub live stream. VS Code plus GitHub live stream coming at you not live from Universe but also live. Well, we were live from Universe. Now we're live from, it looks like to me, the West Coast. Yes, >> West Coast, Oregon area, Cal, uh, California area, and Tennessee for myself. Uh, this is a big, we got a big show for y'all today. We have uh the one and only James Montagno here. >> Yeah. Three straight days of vibe coding at GitHub Universe, right? Without sleep. >> Straight off a plane. Straight off a plane. Just landed. Just landed. >> Man, >> just landed. 27, 27 degrees. I'm ready. Let's do this. >> Is it really? That's freezing. Live in a desert. Live in a desert. >> That's Fahrenheit, right? >> Yeah, we should specify. And we have a special guest, April from GitHub is here. Hey, April. >> And April and I were chatting because April's been up since like 2 am or something. She, as, as I understand it, she got up at 2 am, went swimming, swam to Alcatraz and back. Is that right? Do I have that correct? >> Pretty much. Yeah. >> How'd you do? >> Sounds right. Uh, this is a huge show because as if you don't know this chat, uh, GitHub Universe is probably one of the biggest times of the year, not just for GitHub, certainly for GitHub, but also for VS Code. It's where we make our really big announcements, and there was a lot of big announcements. I was not at Universe this year, chat. I was actually on vacation, and I came back and I learned about all this stuff at the same time you did, and I was, I was completely blown away. I had no idea that a lot of this stuff was coming. That's how we roll around here. And so we want to go back and recap today for you the biggest announcements from Universe, both on the VS Code and the GitHub side. So I thought we'd kick this off by just kind of going around and maybe let's see here. Let's start with, let's start with James, and James, maybe just talk about, because there's so many, and the stream is titled, it's the biggest announcements. I thought it was the one big ann, there's not one big announcement, but to you, if you had to pick one big announcement from GitHub Universe, what are you going with?

>> Wow. Uh, one biggest announcement. I think for me, the biggest announcement is like this kind of agent control plane. And I'll pick this because I think it actually isn't just about, you know, uh, Copilot and github.com, but it's actually sort of like anywhere developers are using agents and using GitHub together or using our editors together as well. So the control plane really enables not only Copilot, uh, but also enables, uh, other, uh, agent vendors such as, you know, OpenAI and Anthropic to basically enable this like entire dashboard panel to have a, a similar experience to see where your agents are working, what they're working on, how the diffs show up, not only on github.com and this like beautiful, you know, um, uh, orchestrator, but also additionally inside of VS Code. So like when I think of this control plane, uh, the words were there, we actually call it a control plane internally, funnily enough, but it said control plane. So, I'm going to say control plane. Uh, but I think of it as like the, the ability to have this openness, uh, to the, the world's most open source, open sourcess, you know, obviously, uh, of of VS Code, but also of GitHub, uh, where, yeah, you know, Anthropic and Claude and OpenAI and C-pilot can live harmoniously together with other folks as well. I think Kyle was inviting everyone else in. Now, when you go into VS Code, right? Like you not only have Copilot as an option, but you have a lot of options, you know, that are extensions, but now extensions can live side by side into the ecosystem that the VS Code team is building on. That's the same thing on github.com. So we'll kind of show it later, but now, for example, your CLI work and your agents working in CodeX and the ones that are working in just local chat, or ones that are working on the, the cloud agent with Copilot remotely, are all kind of living harmoniously together. So as a developer, I love that, right? I love that I have one place to look for things. I don't have to go to 8,000 different areas, and I can just like log in with my GitHub account and my C-pilot subscription, and I'm like, good to go. So, I think that was like amazing. Oh, and also like VS Code and TypeScript are like the number one like languages in open.

>> Yeah, because we did, um, Report of the Octiverse, right? So, TypeScript's leading the charge again. Um, and actually, James, you talked about it, like, I think in my, my long, long time in tech, that ability to just work in one place has been absolutely huge, and it's been so hard to like change context all the time. And we talk about models, we talk about, well, we want to try the different models, and having that new control plane and Agent HQ is is huge, and having Claude on stage was really cool, like that's, you know, shown the partnership and where we're going in the direction with that. So it's been, it was pretty cool to see that.

>> Totally agree. Now, James, when you say the agent control plane, is that Agent HQ? Are we talking about the same thing here?

>> I, I mean, I think when I talk about the control plane, I'm talking about Agent HQ, and I'm sort of talking about like the, the technical aspects of like, there is the Agent HQ like on, like github.com. April will probably talk a little bit to it, a little bit more, like, like when you actually click in to see recent sessions, you're actually going through and then seeing like everything happen. Uh, but there's a lot, there's actually agents that are like being plugged in that you can say, okay, hey, I want to like assign this issue and use this agent as well. Maybe April could talk about that a little bit more. But I'm also talking about like the kind of flexibility that we're seeing, which is, hey, you know, inside of, you know, not only just github.com and kind of that experience of Agent HQ, but also like inside of VS Code, right? I am there, I click on my agents sessions panel, and, uh, my agent sessions will show me my local chat, shows me my Copilot cloud agent, it shows me my Copilot CLI, it shows me my OpenAI CodeX, and it'll soon show me Claude and other things as well, right? So just like April was saying, it's like anything I'm working at, I can see everything in one place, and then additionally, I'm just logging in with one account, right? And I'm good to go if I want, if I want to roll that, uh, option.

>> So maybe >> I was going to say the big, the big part of the control plane is saying I need to orchestrate. So if we look at where we are today with agent mode, we are going one to one, and now we're going, now one to many. So having that control plane means we can go us as almost like the manager or the orchestrator of all of our agents off doing tasks. So we can work asynchronously doing a lot of things with our code, and that's exactly where the vision's going, and that's what the, that control plan is that first step. So not only is it just Copilot anymore, we're bringing in the other, uh, agents in there, and then we can orchestrate a whole slew of things, and giving developers the choice, giving them that single pane of glass. And a lot of our customers are like, well, I want this in my IDE. I want this in outside of VS Code. I'm like, well, VS Code is our flagship, number one. Number two, it's only GitHub.com experience. Do you have that control plan, that view? Um, are you guys going to put out some extensions around that? Do you know that?

>> Uh, April, I think your mic got switched over. We lost fidelity there for a bit at the end. Um, do we have, now, do we have anything we can show people like right now on Agent HQ? I know we have some stuff in VS Code, James, that we can show VS Code, but if April, if you have anything Agent HQ related with some custom agents.

>> Um, I don't have anything I can show at the moment. So, um

>> Let me, let me show you what I got.

>> Yeah. So, let's Hashim is asking a question. I want to understand the conceptually the control plane. What does it mean to have other agents? I think if you see this in VS Code, it'll make more sense. So let's do, can James, can we do that on your screen here?

>> Boom. Look at that. Yeah, just brought this up. Okay, so this is the, the thing that I was actually vibe coding out, um, the other day. Uh, I, I built this with Copilot. This is my little like, it's now called the readme badge creator. Or it was called the MCP badge creator, but now it's something else because I want to do a whole bunch of different things. So, for example, here, um, here I can see I have my MCP like badges. I can go in here and I could, for example, come in. I have this little like, you know, monkey MCP, and then I could give it a URL, and then I have like a full readme markdown. I have like all these different editors and different integrations that I can kind of do. I have VS Code extensions. So if I go to like the, the marketplace, for example, and let's say I do like, you know, the docs or whatever, I can like go create little badges. So now I have these little VS Code badges that I can go and like, you know, put on my readme right away. So, for example, let's say I wanted to come in into here. I have my my agents, but I also have this agent sessions. Now, I actually pinned it. You'll see it on the left-hand side. This little agents view right here. I think it actually makes the most sense. I didn't know you could do this, Burke, but you know, you can, you can take a panel. Watch this. Boom. So, I got my chat and I got my agents. Boom. Like right there, side by side. So, what's cool about that is I now have all of my GitHub CLI going on here. I also have my C-pilot cloud agent going on here. And then I can start new local chats as well. So, I could just open up a new chat. I could open up, uh, a new CLI chat. So, actually, this is like connected to the, the C-pilot CLI. So, local chat, CLI, I could start an OpenAI CodeX session here as well. So, if I click on that and I open a new CodeX editor, let's see, it's actually going to, I just, I just reset up this machine for the first time. So, let's see, it's going to load here, and I can like load and actually log in with CodeX. There we go. It's loading up for the first time. If you haven't installed CodeX, it'll actually go ahead and, um, bring that up. So, okay, perfect. That's good. Okay. So, here, that's this, this new, let me open up two of them here, right? So, this is pretty cool. So, here I can like sign in with ChatGBT, or I could just sign in with my GitHub Copilot if I have Pro Plus, specifically to do that. So, I could come in and I could say, sign in, and I'm going to go, continue. And now I'm just logged in. I don't have a chat, I don't have a ChatGBT account. Spoiler, spoiler alert, I'm the one person on the internet that doesn't, that doesn't have a, so I don't have like, I just, I now I'm just coding with CodeX, right? So, you're, it's kind of, you have the control of like, what you

>> Something. Yeah. Something I want to call out, Vance, here that you skipped over real quick. When you when you first open that CodeX window, chat, it it tells you in there that when you sign in with GitHub Copilot, it's using GitHub, your GitHub Copilot subscription. So, it's using the GitHub Copilot service to access the models that that OpenAI Codex actually needs. And Codex is confusing because it's both a model >> and it's an agent. So when we say Codex, >> there's GBT5 CodeX, which is the model, and then there's CodeX, which is OpenAI's agent, which is what you see running here, which is actually running through GitHub's service layer and using your C-pilot subscription to do that. So it handles all of that, and as James said, you don't need an OpenAI account, which >> you were clarifying >> exactly. So here, for example, like, you know, I can just work locally. I can decide if I want chat mode, agent mode, agent full access mode, and then the models. And again, these are models that OpenAI is providing through Codex, which are GPT5 and GPT5 Codex, right? And this is nice. So people ask like, especially I was there at Universal, why would I, why would I decide to use like, here's like, you know, the, the CLI, and I have access to a bunch of different things. I could say, how does this app work, right? And this is actually going to spin off a CLI agent, right? So I have like a CodeX agent happening here. I have a CLI agent working, like running in a CLI. And then additionally, here I have like a whole chat. And this would be like your standard, you know, chat agent mode. We have a plan mode. Now we have a bunch of different, uh, agents that are here, like my, my, um, Playwright agents. But what I want to point out over here is like on the right-hand side, you can see that I can monitor the work, right? Okay. So, if I was to fire off a new cloud agent, this is our GitHub Copilot, you know, like coding agent, cloud agent. I could say, um, can you add a new pink theme, uh, to the app? And then fire that off. So, I think as I'm working on my application here, I'm going to go ahead and just like delegate this to the cloud agent. So, here I have things working, like I have this working in the, the CLI. So, here's like the, this is actually being run in the CLI, but being visualized in VS Code. I have the OpenAI, uh, CodeX, uh, here, which is running the, the OpenAI CodeX extension. We can see it kind of here, like doing its, uh, thing, loading up, uh, on my machine that's under pressure here. And then we also can see that we have this, you know, brand new Copilot, um, cloud agent here that's adding a new pink theme, but it's doing that up in the cloud remotely, right? So when I think of like an agent dashboard, this is showing me like all of my agent sessions all at once, basically, right? And there'll be more in here as I choose to use them or not use them, whatever I want to use. Like me, I may not use my, the OpenAI CodeX. I may not, right? Like here it is. It's like showing me everything in the application. It can deep link right into the application, into the source code, can show me everything, can do diffs, all that stuff. Like me, I'll probably, you know, use more Copilot stuff because that's me. But the flexibility and openness to have this dashboard so we can all sort of live harmoniously together here is like really, really cool to see. And then you can also hover over the Copilot icon and see things running. So if you have things running locally or remotely, you see them all running here in this little popup. Yeah, someone says, I don't believe I have Copilot. I do have, I have Copilot Pro Plus, but then plus plus with unlimited everything. So, >> yeah, I don't know what the what the Microsoft tier is, but, um, I don't actually don't think it's unlimited because I get like messages from people internally. They're like, "Hey, I hit my limit. What do I do?" Like, I don't, I don't know.

>> Exactly.

>> So, so some of this reason is like some people prefer the look and feel or or like how things work, right? How your, your brain works and how different things work, uh, in general. So like whether you're used to using the CLI, this, I could have just opened the CLI. But here, this is nice. I can actually just like pick the model, do the thing, it'll run the commands for me, or if I want to open up a remote session, right, and monitor that remotely. It enables me to do that. And then I can go to, you know, github.com and I can go over into the, the agents over here. So, let me go and pull this over. So, this is, um, you know, github.com/agent. So, here I can actually see that new pink theme being added here. I can see all my agent sessions happening over here as well that are going on, and then I could actually delegate new, uh, ones over here too. So, if I wanted to work on, uh, yeah, if I wanted to work on MCP badge creator here, I could actually come in. I can actually decide what branch I want to work on. I can also select custom agents, and that's where that sort of agent stuff that that April was talking about. You can set up custom agents. So, I have like Playwright agents in here. You could have like a C agent. You could have a, a TDD agent, for example, that's running and doing that specific work, which is really cool. So, when I select this, for example, we can see that I have these agents that are being defined inside of here. So, I need to update these ones to be, um, uh, compatible with, with this one here. But I can create a new agent that does work. So, if I had a TDD agent, for example, or if I had a, um, Playwright agent, it, it would be doing that work that I'm assigning to the agent, but in the, in the aspect of with this agent frame of mind, right? Just like how we used to use Beast Mode, or we're using Playwright agents, or we're using, um, um, maybe you have a, a, a Vite agent, or you have a, a Nuxt agent, or you have a, a Blazer agent that's like an expert in that area that can go off and do that work. And as April said, you can orchestrate these things together as well, which is really cool.

>> James, while you have that straight up,

>> Yeah,

>> can I double down on something? I want to talk about one of the things that we talked about at Universe, but probably isn't like the biggest thing, but is actually pretty critical. So, one thing we talk about with any agent, any AI, is the context, right? So, we talk to people about how do we put context into this? While you're over there, can you show everyone, um, I can see it on your screen, Spaces, while we're here.

>> Oh, yes. Perfect. Um, I think this is an important one because, so, I think to give everyone a bit of context at Universe, um, we had customers from all over the globe at Universe, and and some customers are early adopters to AI. They're just going, our, our responses aren't great. So, no matter if you're using the different agents, the context capability of what you're giving the AI models is really critical. So, Spaces are great. This is a great way to create a private collection. Um, James, go ahead and create us a new, um, custom space while you're here. We can call it whatever. We'll call it James's Space. We'll call it April is Awesome, whatever floats James' boat. And we can start adding in that context. We can decide what project it's going to be tied to. Um, and this is cool because we can add files, but we can add things like our pull requests, our issues, um, documentation, our wikis, all this stuff into here. Um, and then we can put in our custom instructions files, like right in there, if we want to give Copilot those guardrails in that context. And this is awesome, especially when we have large software projects spanning multiple repos, and we're trying to do a lot more with agents. So, I mean, James, this is awesome. Like, so many people are going, how do we get the context? How do we bring it all together? This is really bringing together the, uh, not just a single repo thing, but bringing all the repos together and bringing collections of data together, which I think's huge.

>> Yeah, totally. So, like, for a good example is like if my MCP badge creator was, you know, having code and having dependencies, right? So, for example, like, oh, MCP badge creator is using this like NuGet package, or it's using this other like thing that this dependency, this API that's actually living in a different repo. I could create a space for that. Correct. And bring them together.

>> Absolutely. And it's, it's this is game changer for a lot of customers, and they've been asking for it because I think when customers, especially come from other platforms, are like, hey, I do this in GitLab. How do I do this in GitHub? Like, we don't have a one for one in a lot of things, and we're really trying to get that orchestration mode with the agents like further down the road than just saying, yeah, let's go to a tree folder structure. Like, tree folder structures are, you know, ages old. We don't, we don't need to do that anymore. We can do it a better way.

>> Yeah, I love that. And this, this is the one thing I need to explore more and more of like inside of my applications to bring it all together, uh, which I think is like super rad, like you're saying, and I think it's something to explore if you're exactly in that scenario. Not everyone's going to be in that scenario, but like for me, it's exactly the same thing, which is like, I work on a team that I have, I'm working on the front end, but I'm working, someone else working on the back end. Like, actually want to share a lot of like common infrastructure, like our instructions and a lot of, um, our common commonalities. So, like the front end can understand the back end, even though that front end maybe isn't in that repo. So, couple questions for you, James. We get to the chat here, but, um, can we clarify, is an, an agent is controlling the other agent sessions? How does that work exactly in VS Code? What's going on there?

>> An agent controlling the agents? Well, that's a great question. Uh, I like to think in chat mode now, or in chat, there's in agent, agent, and, and a plan agent, which is new, which we'll talk about, an ask agent, an edit agent, and agent agents. There's agent agents everywhere. There's like a local agent, right? So, you can think of like anything here, like this is the CLI agent. So, it's a, you know, an agent running in the CLI, which is, which is Copilot. This is the CodeX, you know, CodeX has its CodeX agent, right? So, it's the mode that's here, and then there is like chat, which is like local agents, and then the cloud agent is an agent that's running in the, in the cloud, right? And Copilot in that case is your agent, and CodeX and the other instance is kind of your, um, instance there. However, kind of what, uh, April also talked about too, is that you can not only assign it to like the standard agent, you can define your own custom agents as well. That was one of the big things that sort of, um, um, they were talking about on stage too, which is like, you can define agents for your project or how maybe work with your back end. So, for example, maybe you have a Superbase agent, and you're about to do some Superbase work, and you define exactly how, you know, how you want to work with Superbase, or with Azure, for example, and then you can assign work to those agents to go do specifically. Does that kind of make a little bit more sense? It's kind of like, you know, custom modes we had before have evolved into agents, or custom modes you were saying, hey, agent, you know, agent mode, hey, Copilot, like, work this way. Now, inherently, we're just changing that functionality of how they work to be more agent-like.

>> Yeah. So, it, that's exactly what it is. It's like a tweak to the terminology. Instead of calling them chat modes, we call them agents now. So, Beast Mode would actually be just like, I guess, would just be the Beast, uh, agent, right? It's a, it's an agent, right? Exactly. Another question real quick, and then April, I want to get to you. Um, seems this is a good question. We kind of hit on this, but if I'm using OpenAI CodeX through C-pilot, does it hit my premium request allotment, or do I somehow magically get unlimited >> GBT5 access?

>> I do have that answer for that.

>> I know, I know the answer as well.

>> The answer is it uses your premium requests.

>> Yeah.

>> Yeah.

>> All the billing and rate limits from Copilot service layer apply here. So, so think of it like this is, I just used like, I, James, I do not have a HGBT account. Like, I don't have a Codex account. Like, I don't have it. It doesn't exist. I don't have one. But I'm using Codex here through my GitHub Copilot subscription, which will then use one premium request in this instance for that turn. Yep. And then down here, this is a little bit different. If you're brand new, you've been using Copilot a lot. Here you can see this is context window size. So, this actually has nothing to do with this has nothing to do with like premium requests or usage or that. That's like how much context is in this, in this chat. And what I learned yesterday on the streams is that in Copilot, so if you're inside of like a, a chat here, um, you'll, you'll see eventually, maybe like, it's like summarizing chat, and what that's doing is compressing the context. So, kind of like Copilot and Copilot Chat sort of hide the context window, whereas the OpenAI Co CodeX extension puts that front and center. But that's the thing too, is like, you also, also, you know, you have GPT5 and CodeX here. So, it's what experience do you like? So, I think here, like, what, as an open platform, you know, um, it's all about choices. So, that's what the, the, the VS Code and, and the GitHub team are all about, and having partnerships with, with all these amazing, right, different model, uh, providers, but additionally, you know, the extensions and the ecosystem, and to make sure, hey, like, they're, you know, showing up side by side in our agent sessions, everything that's happening here, too, which is really cool. And I love that you can obviously like see everything happening here inside of here, but also there's this new view, like when you click on a coding agent that's happening, you can actually see it sort of happen and update in real time, and the files change and things. So, like here we can see it kind of like go through. It's adding stuff, and it'll actually give you live previews and things as well when it's like working. So, you get this nicer view. This used to be the sessions view, but now you can go in, you could actually like see, uh, you can open it, you can open in the CLI, you can view the pull request. It's, it's a nice little dashboard here of everything that you kind of see happening, uh, as well, which is really, really cool. So, kind of want to show that off as well. Yeah.

>> So, that is, that's available today, y'all. And again, we're in, we're on Insiders for everything in VS Code here. You want to be there as well if you can be, uh, if you want the latest and greatest.

>> Uh, April, let's go to you. Uh, biggest announcement on the GitHub side, at least. I mean, a lot of these things are blurry, right? There's not a clear line here, but

>> There's a, there's a lot of cool stuff. Um, I think, so the, the thing that I heard most customers, enterprise customers, really excited about was the metrics and the dashboards for Copilot now. So, why, so a customer comes in and pulls in Copilot, commits to how many have ever licenses, and the adoption rate is sometimes really low. Uh, but also what features are they using in Copilot? So, when a, when any company makes investments, and also talking about like DevOps lifecycles and observability, what kind of agents are they using? Where are they having success? So, with the new metrics, dashboards, and the API, customers are extremely excited about this. This is the one that I heard the most. I didn't expect this one to get the most excitement, but literally every customer brought it up to me this week. So, they're excited about that. I can't show it because, um, a lot of the stuff's been under NDA, and I can't live stream some of the dashboards that are going out right now. Um, but I think for, for customers, again, that feedback mechanism, that's been the biggest thing we've heard all week.

>> Yeah, I agree with that. We get this a lot, this question about metrics, and we have some rudimentary metrics right now in Visual Studio Code that basically show you your premium requests, um, how many are being used. James, you want to toggle back to your window just so we could just show people that real quick, and then we'll go jump back to github.com here.

>> Yeah. Uh, over here.

>> You can see that here, right?

>> Sweet. Included on the >> right there, right? It says, yeah.

>> Yours won't say included, but it will say, sorry, it will say a percentage.

>> And then there's a, there's a little bar over to the right. Um, one more. And this one is interesting. It's that little, this little status bar to the left of the frog, to the right of the frog, which you're not supposed to call this. I didn't call it that. April, yeah, right.

>> It's Copilot. It's Copilot. It's Copilot.

>> To the Copilot.

>> It matches. It matches this guy.

>> There we go.

>> Yeah. So, it, that right there, chat, this is actually how much AI is coding versus how much you're coding.

>> Right. So, you can see here, James has just given up.

>> A lot of keystrokes, too many percent. Yeah.

>> Yeah.

>> But this is the dashboard. This is the dashboard. So, I, I recommend going to the changelog if you really want, like, the best place. I'll drop it here. Is like, I just like open every tab from yesterday. I was like, from E8, I was like, boom, boom, boom, boom, boom. But, yeah, like, this is the, this is the new, the new dashboard, and this is an enterprise feature, correct, April, right now?

>> It is. It is. It's part of the GitHub Enterprise, uh, licensing. So, I mean, this is something, you know, GitHub has always been API driven for years, and customers would take the API, connect into it, and build their own dashboards, whether that's in Power BI, DataDog, whatever. Um, but they're like, no, we want, we want better dashboards. So, this is great. Um, there's a question here about any privacy concerns with the dashboard. So, the, the dashboards are, I'll hit this from a couple different angles. First off, they're in your GitHub Enterprise subscription. So, they're ring-fenced into your subscription. So, your data is in your data. Um, the other thing is for any Copilot usage metrics, uh, go to the GitHub Copilot Trust Center, and I'll put that on the screen in a minute. Um, that will walk you through how that is managed in terms of your data, um, your data residency, etc. But also, you can go into the configuration, um, uh, within your Copilot settings and configure the metadata and how much of that sent up, uh, for data. But we, we don't train on private data. I think that's the TLDDR that everyone needs to know. But you have some good granular controls as well. But go to the GitHub Copilot Trust Center, and that will walk you through a lot of the, the bits. But yeah, in terms of, you know, we, we are 100% focused on security with our customers. Um, you know, same thing with Microsoft, like Microsoft builds on trust. It's one of their big taglines. GitHub does the exact same, secure from day one. So, that's huge for us.

>> Absolutely. And, um, so this is, I'm excited about this. We get this all the time on, uh, on various, like, social media, people asking about this. Where can I see better analytics? Obviously, making an investment in Copilot for your company can be pretty big, right? It's not cheap. And so you want to be able to show that, especially when people are running pilots, that productivity is actually improving. And so we're really happy to be able to get this stuff into people's hands because the data is clear on the productivity. It's just not, your data isn't has not been available to you, but it will be, which is really exciting.

>> Yeah. I want to point out a few things. I'm going to do a speed run here for folks as well because then I got to drop in five minutes for, like, call.

>> But this is what I did yesterday on my plane ride home. Like, one, MCP, the GitHub MCP server, completely new and different. It's like fantastic. They've merged and shrunk down tools to be like super unified, beautiful. Here's the VS Code stuff that I'm sure Burke's going to show off. The GitHub code quality is now like packed in, and like the PRs are completely redone. This thing is like wild to be able to see like a full preview of like CodeQL, like linting, everything built in. Now, actually, like when you do a code review, it kicks off an action that you can watch in this graph of like the agent working and working through your code. So you can see what's happening. It's fantastic to see. Um, you can now use Copilot coding agent. You can say, at SL, you know, at Copilot in Slack. You can say, at Copilot in Teams and other places in Linear now, which is crazy. I thought there, so many developers using Linear and Slack that just want to be like, oh, boom, do this thing, which is amazing. This is the control plane and AI stuff that we're kind of talking about. The user, me, metrics that are there. Oh, this is a game changer too. Now, if Burke opens a pull request on my repo, I can now say, at Copilot, rewrite all of Burke's code because it's terrible, and then it will open a new pull request off of Burke's pull request, which is crazy. Also, you can now run Copilot coding agent in self-hosted runners, which is bananas. So, someone was asking about running the coding agent, the cloud agent on self-hosted. It brings me back to my Azure DevOps day. It does really.

>> Let's, let's not say those three words. If you say it three times, you know, it's going to come out of the woodwork.

>> I want to actually point out something really cool about the coding, um, GitHub in pull requests that you went up to. Um, when you're using these agents and GitHub Copilot creates, uh, a draft PR in their tasks, C-pilot can't self-re its own code. Unlike every other developer who's written code and going, yeah, I'll, I'll approve my own PR. We put some boundaries around C-pilot. And I, I think that's a great one because so many times as a dev, you're like, you know what, I'm just going to push my own code. No one's going to know. It's going to be fine. Um, but putting in those guardrails has been, has been really, really good.

>> Oh, yeah. Let me see if I have, like, really quick here, like the, the MCP badge creator. I'm pretty sure, like, I, let me see if I can find the action here and which I ran to kind of show it off. So, you go into actions. This should see it's things are running. You see running, Copilot's running. It's adding a new theme. Then here's the code view one that actually ran. You can actually see. Can I drop this down somehow? I can't. I don't know how to do that. Um,

>> You can see it pretty good.

>> Oh, full window.

>> Oh. Oh, gosh. Make it bigger. Come on, James.

>> Make it, make it bigger.

>> So, you can see it's preparing. It's running Q CodeQL. It's like running this off to the agent. It's like uploading the results. It's cleaning up the artifacts and like reporting it back. So, you can actually watch it in real time, and you can click in, cuz it's like, you know, actions that are running. You can actually look to see what it actually did and what the outputs were as well, which is like really, really cool. Like, if you're, because like before, I was always wondering like, what is it doing? Now, you can see what it's doing.

>> Well, now, and this was announced this week, um, you can, you can modify Copilot on the fly. So, if Copilot's working through its issues, and you're like, you know, this isn't right, you can watch the process. You can go in and modify it on the fly, which I think is awesome because sometimes Copilot's a little too helpful, or it just goes off in the wrong direction. I don't know, it hasn't had its coffee that morning. Uh, being able to kind of interject and like course correct is also been really good.

>> Yeah, exactly. So, like, a lot of folks that are using VS Code, they'll be, they'll be used to saying like, okay, like, hey, things are going on here, and this one might be stuck or something, but I could be like, "Oh, you know what? I didn't want pink, actually, like I wanted this shade of pink." So, I could just talk to it, and now you can actually just also just do it right here in this, in this panel. So, you can put it back on track to whatever it's doing, which is, which is super rad. All right, I got to drop, y'all. This is this is

>> Thanks for joining, James.

>> All right, let's get to a couple questions here, April, and then we'll, uh, we'll move on. I want to talk about something that I really was excited about. So, this one from Greg here. What's the advantage of Copilot versus Claude? So, I assume, Greg, I'm gonna assume the question is like, what's the advantage of what is the advantage of C-Pilot like in VS Code over Claude code? I'm going to answer it that way. If I get it wrong, forgive me. Um, it's a matter of just what, which one you're most comfortable with. So, they're similar. They're both agents. They have different workflows and sort of different ways of working with them. And some people like Copilot agent, and some people like Claude Code, and some people like CodeX. And so the idea is to let you use the agent that you want inside of Visual Studio Code with one subscription, ideally, right? And so, and, and you can use multiple ones. You could be using Copilot and Cloud Code and CodeX and, you know, um, coding agent in the cloud and CodeX in the cloud. You could use them all, right? They all have different strengths. And so ideally, that's choice. Choice is what we're after here, right, April?

>> Absolutely. Um, you know, we want to give people the choice, and I think the big thing is that you see from GitHub and, and Microsoft's been doing this last few years as well, but it's bringing in the third-party tools. You're, you made an investment in tools, you also want the choice. We're not going to lock you into one thing, and that's been the big thing, like how do we, how do we make it so people want to be here, but they can play in the other playgrounds, so they have access to

>> Absolutely. Now, a had said a couple, I had a couple comments about GitHub should allow you to run local models. You can do this today. Uh, there is nothing stopping you from doing this. So, if you go into, and actually, I'll add my screen to the stage here. And let me over here. So, if you go in your VS Code today, let's clear out the chat. My, I'm a little bit of a mess here, y'all. Bear with me. And you go to configure, I think it's configure, uh, models. Uh, you can do it from here. Manage models, right there. And then you have all these different providers here. And like Olama, this is a local provider. Olama has cloud models, too. But like, if you choose this one, you have Olama running locally. Any models you have running locally will show up, right? I think, um, Cerebrus is in here as well. Oh, I don't have the Cerebrus extension installed. So, there's a Cerebrus, um, which does super fast inference. I don't know where Cerebrus, maybe I don't know how to spell. Anyway, should check that extension out as well. But yeah, you can run local models today. There's nothing stopping you from doing this. Run them to your heart's content.

>> Um,

>> Absolutely. And I also want to point out there's some, there's some questions in chat about, um, model usage and feature access and swapping things around. Um, if you're under enterprise licensing, the enterprise controls a lot of what you have access to. So, a lot of customers will only use certain models that are, um, GA, if you will. They will use anything preview or premium models because there's going to be legality around the models and, um, compliance requirements. So, they need to get each one of those models approved. Um, so sometimes if you, uh, want to do stuff locally, that might just be blocked by your admins, unfortunately, and that's a bit of a tough

>> That's what I'm thinking here because, yeah, um, they're saying there's no manage models if you joined enterprise. I assume your enterprise has it turned off.

>> Yeah, and these are all things that are there, and I, and I think of it this way, like, we give all the tools to enterprises. It's their choice to turn it on or turn it off. There's going to be compliance, regulatory requirements. Each countries, uh, each set of countries have different laws. So, I'm based in Europe, and Europe has some very specific things. So, for example, like the Copilot analytics and dashboard, we have to report differently in Germany than we do in the US. Um, and, and some employers just have compliance requirements around what they're allowed to report on internally to employees, especially if you're a contractor, your third-party contractor. Like, there's all sorts of stuff that is unfortunately out of our control, but the features are there. So, um, try to, you know, bring it up internally, raise up the flag pole, write a, write a business case for your team of why you all need it. Um, but yeah, unfortunately, a lot of it's, you know, it's locked out.

>> Yeah, there's some problems we can solve for you, chat. But getting your enterprise to turn features on is not one of them. That's up to you. You're on your own there, although happy to get on the phone with you if it would help. All right. Um, let me see here. Oh, that shrunk up my window here. So, let's talk about something that I was really excited about, which is this new planning mode, uh, which Pierce demoed in his keynote demo. So, we talked about how we had chat modes, and then we re-sort of renamed the, the nomenclature is not chat modes anymore. It's now agents, and that's this box here. So, we have these different modes here. And if you have been following, so here's Beast Mode, right? These are the chat modes, and these are the different ones I have, right? And so you can see that I have, like, a chat, a mode for designer. So, in this case, this designer knows I've given it instructions about how not to design like an AI, but to design like really minimal. So, that's, but this would now be considered an agent. But you can see these up here are the ones we ship to you. And there's this new one called Plan. Uh, and so I want to start here. Planning is a new agent that you can use to generate a plan for building out a feature. And the reason why this is super important is because we, we have discovered, just as an industry, that if you create, if depending on the complexity of what you're doing, once you reach like a moderate level of complexity, you really

You need to generate a plan before you actually have the agent execute. And the reason for that is because there's a bunch of edge cases that you just didn't think of. And the agent then has to make a bunch of assumptions on your behalf. And that, and then it's going to make assumptions about things that you didn't want, and that's how you get a bunch of code that you have to just throw away. And so planning prevents this. This is like the spec-driven dev idea, but this is just at a very basic level of doing it.

Planning also will get you way more mileage with GPT-5. If you take nothing from this stream, nothing else, it's to use the planning model with the planning agent with GPT-5 first and then have it implement. If you do nothing else, you will increase your productivity drastically just by doing that. That's all you have to do.

So, let's take a look at how this actually works. So, in this case, I have an application. Not that it matters. The application isn't really that interesting here. Um, but what do we want to do? I want to, um, go ahead and I want to say, add, uh, field validation to all fields in this app, and I'm in planning mode here. Now I'm going to use CodeX because, as we talked about, I do have unlimited premium requests. I realize not everyone has that. I'm working on a blog post on this because what I want to do is be able to recommend to you which models to use to save your premium requests and still plan. Because when you plan, it's a high-turn, um, uh, it's a high-turn operation, as you'll see here in a moment. We're going to use CodeX. We'll talk about what else we could use. It works with pretty much every model, and in talking to engineers, it actually works really well with cloud models. The planning mode does. Okay, pro tip there. So we'll leave it on CodeX for now, and let's just send this and see what happens here.

So what it's going to do here is it's going to start, um, doing a little bit of research on my repo here, and all this is included in the agent. And I've basically given it very vague information here. I've said, "Add field validation." I didn't say what fields. I didn't say to what form. I haven't done any of that, and I was purposefully vague because I wanted it to ask me questions. So the first thing that it's going to do is gather context from the project so that it can actually ask intelligent questions. Uh, and then when all of this is done, and like we get through our plan here, we'll take a look at how the planning mode, or planning agent, I need to get my, uh, my verbiage right, actually works behind the scenes because it's, it's quite interesting, and it's something that you can build on.

"Yeah. And I think this is, this is a big question that we get from people. You know, we, we are still in control of what Copilot's doing, and that's a very, very important thing. This is helping us alleviate a lot of the design parameters, but giving it some guardrails, and then we can review it, and then execute, and then also understand what we're going to do before we do it. Because I think a lot of the times, well, Copilot back in the day, three years ago, um, there wasn't any of this, right? There was no context. There was no ability to see what was going to happen. It was just output, and you had to kind of roll with it or hate it. You had to love it or hate it."

"Yeah, exactly. And I think this, this also goes back to this idea that you are responsible for the code that Copilot writes, right? If you prompted it, and you are the, you're the behind it, then you're responsible for what it writes, right? And that means that you have to be really sure about what it's doing here. Uh, and so planning definitely helps with this."

So, it's asking, "What specific message and link rules should vanity slugs, uh, obey?" Now, where did it get that from? The app. It went out and said, "Okay, you need to validate vanity slugs." Right? So, it's asking me these questions based on the project. They're not just random questions here. So, we can just answer these and say, um, "Just any valid URL characters." And then, "Should we require at least one valid description sentence per list?" So, we have lists of items. Uh, we can allow empty descriptions, right? And then we'll send it back. And this is what I mean by high-turn. So, each time we're making a turn here, this is a premium request. And I realize that this is problematic for people that have limited premium requests. You can't just do this over and over and over again. So, again, I'm working on how best to do this with something like GPT-5 Mini or one of these included models so that you can save your turns here.

So, it's going to do this again, and it does this until it has 80% confidence that it has enough information to proceed. That is the, uh, that's sort of like the barrier that it has for whether or not it should go on or whether or not it should stop. Now, while this is running, April, let's take a look at what plan mode is actually doing. So, if we go to configure agents here, you can see it's right here. And just like your custom ones, you can just click on this, and you can see it right here. It is, you can see exactly what it's doing. So, there's a couple of interesting things here. If you look at the tools here, it has, these are the different tools that it has, and notice that it has the GitHub MCP server, get issue, and get issue comments included as part of the planning. It also has, um, issue fetch and active pull request. Right? So, it's, if you don't have these MCPs or extensions installed, they have no effect. But what it's doing is it's trying to grab all of the context that it possibly can about your project, including what's actually upstream in the repo.

Let me answer this one more question here. "Should slug validation treat uppercase, lowercase as distinct?" So, was this normalized? I didn't think about that. "Normalize to one case." That's a great example of an edge case. I just would not have thought of. And then you can see down here there's something called handoffs, which we'll talk about in just a second. Let's kind of go through this. So, it says, "You're a planning agent. You're pairing with a user to create a clear, detailed plan. This is the workflow. It loops through gathering context and drafting the plan for review. And then you're planning. You never implement. You just plan." And it, and actually, it doesn't even have access to write code here, which is why it doesn't write. It doesn't have access to the tool to do it. This is quite interesting, April. This is like stopping rules. There's so much you can learn from prompts, but this is kind of like, this is usually we want to keep the agent going, but in this case, we're giving it the parameters. It's, it's like instead of telling it what not to do, "Don't stop," we're telling it to only stop under these conditions, which I just never thought of doing.

"Yeah. And I think a lot of people talk about sometimes Copilot goes too far. So, how do you stop it? And planning is, and this is what it's doing. It's given it a bit of guardrails to stop." Uh, uh, there was a question I saw. Um, so there's actually a question on Twitch from someone about keeping an agent running out of tokens. And I think you've nailed it, Berg. I mean, you have to monitor it, right? So, it's whether they, most people are probably going to have access to the dashboard, see where the agent's utilization's going, and I don't know how real-time that feedback's going to be also on the analytics boards because, yeah, but using the, the more free models is, is definitely the way forward if possible, and then elevate it to a, a premium model if you need.

"Yeah. And again, I want to make a video for y'all on this. Ideally, you would have one premium request here. In a perfect world, you'd get a plan, and then you'd use one premium request to actually implement that plan. And you would use your built-ins to actually make the plan, but I need to actually build that out for y'all so I can see what that actually looks like in practice."

And then down here, we have, it's going to do the context gathering, which you saw. It's using this new sub-agent, which is, it's an agent that runs in its own context window and in the background. And so, essentially, it doesn't use the rest of the chat when the, when the sub-agent runs, it doesn't have any of that. It's running in its own isolated context, which, again, as April said, this is all about context engineering. If you don't, if the agent needs to do research and it doesn't need the chat history to do that, then there's no point in shoving that in the context window and eating up half of your context window.

"Absolutely. It's being smart. You know, the premium models are cool, but they're they cost money. And I think you have to, you know, we've commoditized GitHub and Microsoft have really commoditized these models. However, there's still a cost. They're not free."

"Yes. And actually, it's, I think it's way more than people realize. Like, there's going to be some point in the future when people are like, 'AI used to be free.' And people are going to be like, 'What?' And they're like, 'Yeah, you could just go on the internet and use it without being signed in.' They're like, 'That's crazy.'"

"Yeah. Like free, free electricity."

Uh, and then it's, you know, it, here's a plan here. It's going to generate this plan. And then, so you can go look at this yourself. So, here's the plan that it actually creates, the clarifying questions. And then, y'all, you can modify this, right? If you just click enter, you can set it as writable, and then you can go modify this plan to do anything you like, right? Maybe you want to save the plan to a markdown doc because it doesn't do that. And you would have to add, right, something like that. You would have to add the tool up here to do that, right, which is like, right, I think, or whatever the actual tool is.

So, here, let's, let's, let's kind of proceed. You can see we have one question left. It's kind of stuck on this, on this, uh, slug values. We've already answered this. So, what I'm going to do now is use what's called a handoff. And you can use these handoffs. They're right here. You can do this in your own prompt files, y'all. So, you can see the start implementation. It's right here. This is the prompt. Let's just save this. And so you can see that. And then here as well. And so the idea is that instead of having to like type the prompt, you just hit the button and it just starts the implementation, right? And then you just go.

Now, before we go, April, we have another option here. Instead of just going, we could do something else entirely.

"Okay. We can click this button and we can delegate this to the different agents that we have. Now, I don't have OpenAI CodeX installed. But if I did, it would show up here in this list. So, I could just say, I'm just going to pick the cloud agent because I use it all the time. The GitHub cloud agent is phenomenal. It's one of my favorite ways because I have this plan. So, I'm relatively confident that I can hand this off to the cloud agent, let it open a pull request and work, and I'll come back to it later."

"Yeah. This is how my current workflow. I plan, hand off to the cloud. Plan, hand off to the cloud. And I go to my next task. I don't have time. I'm not going to sit around and wait for it to work and tie up a chat session. I'm moving on. Right. Boom. And while that's happening, as James is showing, it's over here running. And you can see right there. And what's nice about this is at any time you can just jump in and have a look at it. And as April said, you can now, if we open this in the browser, although I don't think y'all can see this because I haven't shared it. Me, my Streamyard. Let's share one second here. This is because this is, you talked about this, April. This is important to show people, the ability to interrupt it, sort of like midthought. So, let me share this here. Put this back. So, this is the pull request. You can see it right here. And you can actually steer it while it's working, right? So, you can see that it's working. So, I could go in here and say, um, 'Please make sure to add a red outline around any, uh, invalid inputs.' My taskbar is kind of covering, right? And then I can send that, and I'm actually directing. Look at this. While it's working, I'm giving this back."

"Like I forgot to tell you this."

"Here you go. Like this."

"This blows my mind."

"Yeah. It's awesome."

"It's so cool. It's so cool. It's able to just move." Um, I actually want to talk about something while we're talking about while your code's pushing. I'm going to steal the screen from you, Burke.

"Yeah, go ahead."

"Or try to. I want to talk about code quality because this was something else that, um, we announced, and I'm going to try not to go into inception here. Um, let's go. Here we go. Share this. Uh, can you see this?"

"I will add you to the stage here. Let me remove the stage. Let's add April."

"Let me in. There we go. Yeah. All right. So, while we're all of us are writing code, and we are using Copilot to write code, um, a lot of us are going, 'Well, I have this code that's forever old, I want to refactor it.' The GitHub code quality got announced as well during Universe, and this is just, it's in preview right now. Um, there's AI findings and there's standard findings. I'm going to show the standard findings. So, what it looks at is, um, you know, what's the maintainability of our code, and what's the reliability of it? And then it will give suggestions powered around AI. This one doesn't have any at the moment. This is just kind of a demo code base that I use quite a bit. Um, but I can go in and see the different rules. So, I can do go double down into semicolon insertion for maintainability issues. Go into it and go, 'Right, we've got some issues here.' And what's really cool about this is these are things that I've generated in my code by hand. Uh, Copilot didn't do it. I can go ahead and open a pull request here. Um, I can assign it to Copilot. If everyone sees my screen, it is coming soon. So, I can delegate all these tasks to Copilot, but for now, I can open the pull request. And if we go ahead and open our pull request, um, it will give me a PR in my repo, and then we can also have it ask it to generate the fix, um, and it's doing it right here in this mode. So, we, this is an easy way to help maintain our code ones that we've written, or even maybe Copilot wrote. And again, this could be Copilot two years ago that you generated this code. So, we can see it's given us an autofix here. I can open the pull request. Um, it will give me a branch name. I can give it a a custom one if I want. I can use the generated one. Open a pull request. We can also check out locally if we want to double down on this. Um, that's another thing we do. So, I'm going to go ahead and commit this change, push it to a branch. Um, and it's going to go ahead and throw me over to our change. So, um, with this, it's a draft. So, I said this before, you saw it in Burke's. It's a work in progress. It goes into draft mode. Copilot is not going to, um, push this code without your visibility. So, this is, you know, Copilot here. Same thing Burke is doing. I can go ahead and mention at Copilot. I can go, Copilot here, and start querying it to do things. Um, you know, I can ask it to, um, double-check its work. Uh, I can ask it to do lots of things and add that comment in right there. So, it will then iterate on that PR. Um, now, again, I said this before, we can't, when I add reviewers, um, I can add Copilot. This one hasn't been really like locked down yet, but it's one to think about. You don't want Copilot reviewing its own code. U, this is where you want the humans in there. Again, I can assign myself to it. I can add my labels to it. But this has given me a really good fix with a good amount of information. I can see the diff, and we're good to go. So, I think the code quality piece is absolutely huge for us. Um, so it's, it's a pretty cool thing. But Burke, how's your, how's your, uh, situation going over there? How's your code making?"

"Yeah, it's, it's coming along. I did want to ask on the code quality thing, April, because Kyle made this statement in the keynote where he said, 'A developer's whole job should not be reviewing code from agents.' And this idea that like, you don't want Copilot reviewing Copilot's code, or is that true? Or do we, like, is it okay for me to submit an agent-generated code and then turn around and tell GitHub, 'Hey, review this code that Copilot just wrote?'"

"Yeah, I've done it. I've done it as a test, and it's actually pulled out issues that it's generated. And I think when we look at agent mode, agent mode's meant to be self-healing. And we don't want to just keep reviewing Copilot code reviews all the time. So, there's a self-healing element. I think we tell people to review the code now to vet it because we don't want developers just blindly accepting code. So, I think there's a human element and an accuracy element that we're still trying to kind of figure out. But in terms of the self-healing mode, we're getting Copilot more and more involved to self-heal. So, those of you that have been using agent mode, it will work through tasks, it will self-heal, and you just autocorrected Copilot. So, I think there's, there's an element here which we haven't quite found that boundary in development. Do we want humans kind of involved? Like Kyle doesn't want, he has a valid point. Doesn't want us just reviewing code reviews all day. Like that's not what we're meant to do. But we haven't quite figured out that boundary yet. And I think that's going to be, um, some really cool stuff in, in the near future of what that starts to look like."

"Yeah, I totally agree. And I, I've actually had some pretty good luck taking code that Copilot has created and saying, 'Hey, review this code, make it more concise, ruthlessly delete things that don't need to be there.' And it doesn't, of course, it has no idea that it wrote that code. It's not like a human being. It doesn't remember. It has no recollection of writing that. It's just code. So, it's more than happy to review it. And especially if you have different models review each other, that's another thing you can do. So, you can have like GPT review Claude's code and vice versa. And they have different pathologies and quirks that make them pretty good. Claude is good at catching edge cases, and GPT-5 is good at cleaning up spaghetti code, which Claude tends to be more prone to write. So, you can combine them. And like you said, I think we're figuring this out, but the important thing is that you can do this now, and you have access to all of these models and agents."

"Right, Leslie, this is the foundation on which we do all this."

"Y."

"Okay, any last thoughts, April?"

"Um, I mean, I have lots of thoughts. Uh, how valid they are at this point in the day is is a whole other thing. Um, let me think of, uh, let me look at my list of announcements and see if there's anything else that really stuck out that we haven't really talked about. Uh, talked about code quality, actually. Yeah. So, you know, we've talked a lot about Copilot, but we've missed the boat on some other things. Um, there's been a huge focus on the software development life cycle and securing your end-to-end software delivery from GitHub and more than just Copilot. We talk about agentic AI, but there's been, um, some good releases around Actions. Um, there was the release, well, using Copilot on self-hosted runners, but also immutable releases just came out and went GA at Universe as well. So, I like immutable releases because it means we can't change what we've built. So, we're talking about generating code, okay, we've, we've built this code, we talk about securing how we generate that code, but we're not talking enough about when we ship that code, how are we securing it? So, mutable releases has just got, um, just got this last week, and it's been, it's been great for our customers, especially in the enterprise space. Again, um, and yes, tame craziness, this is the Levi's GitHub jacket. Thank you for asking."

"Which, which by the way, I don't have either."

"So."

"Well, that's because you're not an official GitHub employee."

"Uh, I'm also jealous."

"Yeah."

"Sia got one. So, Sia was here this week, and, uh, he got a jacket. He got a jacket."

"I saw that. He gets all the good stuff. It's not fair. I'm submitting an official complaint with Microsoft."

"I will say, anyone who is out there right now watching, uh, the coolest reason to come to Universe, and mine is not charged, are the GitHub badges. So, we've released the new badges this year. They're wicked cool. Um, and they're color screens, big LCD screens on it. Uh, but yeah, So, look at the upgrade, April, between So, I have the old one, which will, this will be interesting, so you can see. So, this is the old one, and it has like the E in here. So, you used to just say your name, and now I know yours isn't charged, but like, you can play games on the badge."

"So, this is last year's model."

"Yeah. And this is your shoes. Oh, and it has, it has the frame around it. So, last year they gave out some like 3D printed ones. This year it has everything. And I haven't even, I need to actually turn this on. Um, there's a boot button in the back. Let's see if it does turn on. But, uh, yeah, like last night we were all playing games on them. It was pretty cool. So, yeah, if you do want to go to GitHub Universe, you get a cool badge. You get some cool Copilot stuff. We've got the store going as well. Um, we just had a load of cool swag this year. No skateboards, though. There were no skateboards."

"Oh, man. That's one of my favorite pieces as well."

"All right. Well, um, nice. There we go. So, I think April, as you said, like any other new things besides AI. Yeah, this is, this is really important and a good point. Um, AI tends to suck up all the oxygen in the room just constantly. Yes. And it's not the end of the world. It's, it is not the be-all, end-all. There's still GitHub, very much. And I think you're going to see there's a lot again around the SDLC. Um, there's also stuff around security and governance. Probably what I would call the less fun topics, but the super critical topics. Um, there's a lot more, uh, capability with metadata scanning and secret scanning that's available. Um, also custom images. So, customers want to bring their own images to GitHub-hosted runners, which is huge. That's come out as well. Um, and more dashboard updates. So, there's been nothing that has been, I would say, mind-blowing in that space. It's all been agentic, but there's quite a few of the governance side. There's also some stuff around the governance side for customers, again, just with Copilot, with GitHub adoption, just making it easier to do, um, how you manage your users, how you manage your policies. So, there is quite a lot in that space. Um, again, it's all in the change log for everyone that's out there as well. Um, but there's, there's some good stuff, but yeah, we do, we do tend to overshadow with Copilot releases because it's cool and it's fun, but I think the reality is there's a lot in the, in the change log as well for other stuff."

"Yeah. So, maybe producer, if you could find that change log link, maybe we can throw that up. Um, sorry to throw that at you last minute. That's super important, April. I agree."

"Right. Like the other stuff kind of gets pushed to the back burner. And listen to all the devs at GitHub out there that are working on GitHub and are not working on Copilot. Thank you so much for all the improvements that you're just con. And same for VS Code, right? There's tons of people that are just working on making the editor better, and that stuff is super important. Crazy."

"Can I ask you a question, Burke, because I, I get this one a lot. I get a lot of customers that are, you know, Java developers working in Eclipse or whatever. Um, like I always think VS Code is just the coolest place to write my code. What do you tell people that are like Java devs that have been using an IDE for forever long? Like, why move to VS Code other than it just being awesome and having all the latest Copilot features?"

"Yeah, I mean, I think you should use whatever you love. Um, the, why would you move to VS Code? You would do that if you thought that it provided a better experience for what you're working on, right? But there's a lot of people that have other editors that they use that they like, and they're comfortable with. And I don't think that you should feel compelled to, to just swap out your tools just because something is the new cool hottest thing. That would be my answer to that, right? Like, you can do Java dev in IntelliJ, and you can do it in VS Code, and you should use which one of those you like the best and makes you most productive. And I absolutely."

"Yeah. I just don't think that people should compromise on that. And of course, from the VS Code side, we want to help you be the most productive. And so, if it isn't us, please tell us why. Yell at James on Twitter. He'd love to to hear your critiques. He's back. He's back. Like, James, can I add you to the stage or you just, are you just lurking?"

"Everyone needs to go to Twitter. It's been a while since we. Hey, James, welcome back. We're still here."

"I'm still here. I'm still here. And my pink theme did finish, by the way. So, I'm just want to be very. Some some more question here now that James is back. Maybe we can get to these if y'all have just a few more minutes. I got all the time in the world for you, Burke."

"I just need to catch a plane later today. Just let me get home today. But, uh, I got about another eight hours. We're good."

"So, yeah, when you, y'all, when you steer Copilot like that, yes, you are paying for a request when you do that. Good thing to keep in mind. Um,"

"I just, I just, I just steered mine actually. And like one. Let me actually like, um, let me see if I can. I have that. I have this here. Oh, this is cool. So, I was steering the pink theme. Let me go and share my. That's a great question. And the new visualization is like super nice. Let me pull up my. I got to do region to share. Hold on. I got my ultra wide. So, it's like, oh gosh. So, let me do this. This click 8,000 buttons. I'm, I'm really big into button clicking. So, if I go into here, region to share. Boom. Boom. Boom. Boom. Boom. Cool. Awesome. So, this was that pink theme that I just added, right? And it was like, boom. It just gave me this huge here. But I navigated it down. And I wanted some changes to where it was going. So, I just like talked to it like April was saying earlier. And then you can see two premium requests used. So, it kind of, that tells you no hidden, no hidden fees here."

"Well, that's nice."

"So, here's a good question. If you're using auto mode, so if y'all don't know what auto mode is in, um, Visual Studio Code, yeah, James, you can show. I'm going to show auto mode."

"Auto mode. Close this."

"Oh, it's gone. Where's it at? They moved it."

"Uh, I think it's in. Hold on, James."

"There it is. I was, I was clicking on the auto. Wait, no. What I really want is auto agent. You pick the agent that you think I want."

"I mean, you kind of want that. Like, pick the best agent and the best model for the job. I don't want to have to know about any of this. But, there's 10%."

"So, the question is, um, in auto mode, will it pick, will it take into account when you've really nearly reached your premium request limit?"

"Not at the moment, but we are improving auto mode all the time. Right. I think the north star here is that it should take into account everything."

"Right. It should help you get the most mileage out of your subscription."

"It's really, right now, capacity and performance, and that's like what is taken into consideration on that. And I, I feel like it, I, when I use it, it mostly bops between like GPT-5 or Claude Sonnet 45, but it is a 10% discount. So, you would get 10% more premium requests if you went down that route, uh, in general. And, and, and you can, if I was to fire this off. So, let's say I went, I have this, um, have this plan, I wanted to put together. So, let me go into plan mode, and I was to like bop this in here. I want to like add Nougat packages like onto this page. Oh, I should probably go to main, though. Okay. And like plan this out. Like, you can, you should be able to, when it's done, like see what it, like, so for auto mode, like you can't really necessarily see what it's using in this moment, but when it's done, you'll be able to hover over and see what it actually used. But the idea is like, you shouldn't care what it's using, I guess. Like, I think that's some of it is like, you know, it's auto, and you should hopefully trust it to do the."

"I think she would tell you upfront though, right? Like now running this with instead of after the fact."

"I don't know if you, you see this too. Um, I don't know if you showed this off, but this, like the group, like these all ran in parallel, like all these reads, cuz see how they're nested."

"Yeah. So, like I did, you show that already? I wasn't watching the stream, but like."

"No, we didn't, I didn't call that out specifically. That's a good call out."

"Yeah. So, so this all happened, like this is that, um, that sort of sub, I think it's the sub-agent runner that is doing it, basically, like in the tools, like there's this run sub-agent, cuz it can like grab the context from all of them, basically, and like grab them all. So, when it's doing this research, it's actually doing this nested, and then it's actually putting it down there. My assumption is here that this is using a Claude Sonnet model because I would see thinking, and I would see the output of the thinking if it was like a GPT model."

"So, yeah, that's a clear marker there. If you don't see the thinking, then it probably is the cloud models. We did look at the plan mode prompt, James, and you're absolutely right. It is using a sub-agent to run that research so that it can parallelize."

"Async."

"The other thing that we didn't talk about that was pretty cool that was released this week, um, is querying Copilot from either Teams or Slack. So, let's say you're in a, in a, in a Slack channel or Teams channel, and you're talking to some folks, and you're going, 'Oh, wow, yeah, we need to assign a task to that. At Copilot.' And you can start querying Copilot. That got released this week as well. That is really cool. You can also do it in Jira. So, for those of you still using Jira, um, now, I say that, I think it's, I think it was released this week. I hope I didn't miss Steph on that one. Um, I think that's in preview. I should probably look that one up. But, yeah, being able to query Copilot from your favorite, um, you know, commun, bit to communicate. So, yeah, definitely available in Teams and Slack. So, we use both of those at GitHub."

"I'm sure April is on some Slack somewhere."

"My favorite part of Universe, actually, was like, was like Kyle was like, 'And anyone else that's not, you didn't see on stage, and I like showed a bunch of logos, like, feel free to reach out, like, come on into our ecosystem, right?'"

"Absolutely. There's, there's tons of third parties that, uh, actually spoke with customers, like, 'How do we get involved in this?' Um, you know, 'Build it, and we will bring you on board.' It's pretty cool."

"Yeah, I would, I think I'd just close by saying here that it was so encouraging for me to see both OpenAI by Anthropic there because ultimately that's what we all want is to be able to have access to all these things. I want access to all the great tools, right? I don't want to have to pick and choose between them and be siloed. And so that was super encouraging to see. I was very, I was really impressed by that. And I had no idea that was coming. I didn't know if, you know, what other people knew, but I did not know that was coming. That was very encouraging."

"I have to pretend I don't know."

"Yeah. There you go."

"All right. Well, thank you. Thank you so much for joining. Uh, April, thank you. I know it's very early day for you. I hope you get, uh, a nap, uh, this afternoon, if that's something that you do. If you're a big napper, I don't know. Some people don't nap at all."

"We'll see how the day goes. We'll see. But no, thanks for having me, guys. Really appreciate it."

"Yeah. And James, thank you for coming fresh off the airplane straight into the. You just ran through the door. Oh, let's go."

"And thank you, chat, for being here. Uh, hope you watched the Universe keynote. If you didn't, go back and watch it. It's really, really good. I mean, I was just blown away by all of the new stuff that came out this year. It's even more than came out last year. And consider this, chat. Consider this. Last year at this time, we released edit mode. That's where we were 12 months ago. So, here's the question. Where are we gonna be 12 months from now?"

"That's how fast this train is moving. All right, y'all. Good to see everyone. Have a great weekend. We will talk to you all next week. Bye."

"Later."