What the hell is going on with GitHub?

Welcome to Syntax. Today, have Kyle Daglan. He is the COO at GitHub, and I'm just gonna start it off right away with the first question. Kyle, what what the hell is going on with GitHub? You got the downtime and everything like that. The people are mad, and you're gonna see them in the comments. And if I started this with any other question, we would get roasted alive. So Totally. What's going on? Yeah. I mean, like, not only are all the developers showing up, but developers are bringing, you know, 10 to a 100 of their agents with them. And I mean, like, the traffic... I've been a GitHub for thirteen years. Yeah. We've never had growth like this ever.

Because, you know, it's not just that, like, a human, you know... I was I was talking to someone on the weekend. I might, you know, code and maybe push, don't know, eight to 10 times, you know, between kid's soccer games. Yeah. Now my agents are doing the vast majority of that and likely pushing up, you know, 150 times like this past weekend getting ready for universe. Right? Yeah. Yeah. Times 225,000,000 people is enormous. And so we've had to fix the underlying systems, not just for scale. Horizontal and vertical scale, we've kinda done that. Yeah. We now needed to go, agents don't act like humans.

They break things in ways that we candidly like didn't because the behavior is very, very different. Yeah. You know? And so I think we've done a lot of good investment. Like, our uptime has been much better. Like, not perfect, to be clear. We still have work to do. But, we've made some really good work, and now we're, you know, talking about changing all of underlying Git infrastructure and what that can give us. So Yeah. More to come. Like, I appreciate the patient support and honestly, just like, I get tweeted at or I see tweets and I'll respond like, if we're down, that sucks. Yeah. Yeah. Yeah. Sucks for me, sucks for you. Sucks for devs. Yeah. And we, you know, we gotta be up. You know, we gotta meet everyone's expectations. Yeah. Two things here. First, can you give us a... Just like a... Some numbers as to, like, how much that spike has? Every chart in the last two years has been like that. Yeah. Like, how how much more has it been going up? Yeah. Yeah. So I mean, like, when we were at Universe last year Yeah. You know, got on stage, and I was like, you know, this year, we've got, like, a billion a billion commits, you know, and it's gonna be a big deal. I think in early March, we were talking about, okay, we're like a billion to 2,000,000,000 a month.

Right? And now we're at like about 7 to 8,000,000,000 commits a month. And so like, it is just growing incredibly quickly. It's really hard for us to measure, like, how many agents, you know, are on the platform But, like, we're adding more users per second than last year, you know? It was one per second last year. It's more than one per second this year. Oh, man. I know. It's it's hard to, like, really understand that. Well, that's why we, like... Yeah. Yeah. So, originally, I asked the events team. I was like, it's really hard to understand these numbers. So I said, why don't we drop ping pong balls? Because if we drop ping pong balls, you'll understand, like, here's five, and then here's, like, 50,000 ping pong balls. We couldn't pull that off. So you got that visualization of the actions jobs. Myth... MythBuzzers could have done it for the record. MythBusters. Yeah. We're taking all that money and putting it into the under Okay. Yeah. Yeah. Yeah.

I remember the MythBusters spent a lot of money on ping pong balls, so we actually appreciate that. You see? Okay. So, you know, but, like, that's the that's the truth of it. It's like the numbers just feel like is a billion or a lot or a little something we talk about, you know, internally a lot. Yeah. It's enormous. There are more contributions being pushed up this year than the last two years combined. Oh my gosh. So it's... It... Numbers are just, the numbers are just crazy, which is cool because to me, it's a lot of us either doing more things that make us excited in all our passion projects. Yeah. Yeah. Or new people coming in and, know, giving things a shot for the first time. And if you want to see all of the errors in your application,

you'll want to check out Sentry at sentry.io/syntax.

You don't want a production application out there that, well, you have no visibility into in case something is blowing up, and you might not even know it. So head on over to sentry.io/syntax.

Again, we've been using this tool for a long time, and it totally rules.

Alright.

Yeah. So the infrastructure is getting a huge update. Can you give us, like, a little bit more so... Yeah. You're not just plugging in more computers. You are. Oh, yeah. Yeah. Yeah. I'm curious about, like, the actual technical details of, like, what are you actually doing? Yeah. Yeah. So, you know, obviously, yeah, like, we've got 4,000,000 more, you know, cores going in, mostly, like, to talk about, you know, GitHub Actions runs primarily. You know? That's a big amount of use. Yeah. 45,000,000,000 GitHub actions run? Like, yeah. Yeah. Like, there... There's so much use of actions, obviously, because as code is being built and then it's being tested, it's all showing up in GitHub actions. Yeah. So that's a big part of it. But I think on the underlying infrastructure, we're making a big shift by currently, we're using, you know, storage and compute side by side with Git. They're like together. Okay. And we use like spokes, which is kind of what everyone has been using to do a quorum of, you know, when you push up, you gotta get to a quorum of servers before we consider this repo to be stored.

The problem with that is, is you have to lock for the moment in which you're looking for quorum, which can slow everything down enormously.

And so, we're basically separating out the compute from the storage.

So now we're using a, like, blob store, you know, for the underlying code. And then we're using a sort of more optimistic approach when it comes to refs. So we don't need to have full quorum before we can say we have the ref. Mhmm. We're gonna store the files. And then over time, we can ensure that that ref is Yeah. Yeah. Showing up all the right It's like a little bit more that it is just at least not locking while it's waiting Yeah. For the ref to be stored. Okay. Because you can also look at it across, like across refs. Meaning like, if I'm working on a branch and you're working on a branch right now as we push up, we're still waiting. Like, we're waiting on both versus like, I can look at it at a per ref level and say, hold on, like, my branch can work fast, your branch can work fast, and so forth. Like Peter mentioned, you know, from Open Claw, like when we've been working with him on this, it gives us so much advantage because we can tell you so much more about the repo. Maybe you want to like really aggressively like poll or ask questions about it. It's hard to do right now for the same reason that, you know, the write and read locks kind of happen. The other pieces like we can also like do way more with the code than we could previously more easily. Meaning like with agents, a lot of companies and people aren't... They're basically just doing a super shallow clone. You know, they don't actually want all the history or they just want these files. We can provide that all to you when it's just coming out of Blobstore way, way, way, way, way more efficiently Yeah. Than, you know, looking across a set of servers that are ultimately both computing and storing the code, which is just a lot more, you know, computationally intensive. So how long does, like, a big change like this take to go from planning to to completion? So, we've been working on the migration for... Like, well, I should start by saying latency is the problem. So, like, right now, we've been in data centers of our own for quite some time. And, of course, we use some cloud compute, like, for, you know, actions runners, etcetera. But Mhmm. We've needed to get almost everything into, into the cloud, in this case into Azure, for us to be able to take advantage of this new underlying infrastructure. Because otherwise, the hop kills you if you're incredibly busy, of a repo. We... Like many, many, many years ago, we used to have data centers both on the East Coast and on the West Coast. And so in spokes, that's Yeah. Really difficult Yeah. Because quorum takes long every single time. Oh, yeah. And so now that we're in a single or moving towards at least the, like, the full cloud, I guess, I would say, across multiple regions, the goal is to have that done, you know, January, February, March. And then that's when you'll be able to fully experience the benefits. We have your repos. We've write a blog post of the 100 customers that are stuck, like, waiting to be fully migrated because they're in the worst case scenarios, you know, that we need to do more work on. But once the repos are over, which the vast majority are, and all the services are over, then we kinda give you the the best the best experience that GitHub can. The people that are in worst case scenarios, those are... What are those? Just large repos or busy repos or what? Incredibly busy large mono repos. Okay. Yeah. Like almost exclusively. You know, monorepos have benefits, especially, you know, in this agentic era. Yeah. But they also have unique pain points and get land.

And when you're very, very, very busy, you're experiencing the worst of the latency when you're trying to, for example, always merge to main. Mhmm. You know, you experience, the problems that some other, you know, repos won't. So we can make the move for those folks, but we wanna do it when we can, also kind of like give them the full benefit of the rest. So the cloud versus make the move and then very, very, very busy, but also with latency. Mhmm. Yeah. Because the requests are going to two, data centers. And... Yeah. I knew monorepos.

We're trouble. Know?

Scott's a monorepo hater. Yeah. Yeah. I like my stuff in... I don't I don't have this question on you, but I'm kinda curious about this. A lot of people are saying now, like, the the CI

actions, whatever you're using Yeah. Is is the new bottleneck because, like, you can just spin off these things. Do you have any any thoughts on that? Yeah. I mean... So I I think the interesting thing is, like, every time we've hit a bottleneck thus far, we've kind of been able to, like, throw something at the problem. Yeah. You know? It's like, we'll just go past this. And I think the problem with GitHub Actions or, like, you know, CI as a bottleneck is that, sure, you can keep putting more CPUs, but at some point, like, you do not want to. Like, it's expensive, like, for for the user, like, you know. And so I think the the reality is there's all...

It's the first time that we are all gonna need to decide, like, what does testing look like in this era? Yeah. And it is not likely to be all the same tests that we've been doing. It's... I don't necessarily believe it's no tests either. Right. Like, there's probably something in the middle that we're gonna have to actually grapple with a computer science at least or software development Yeah. Problem, you know, and change how we do it. Yeah. Either to be way more selective or, you know, we shared like our proof driven development research from GitHub Next, is basically like, can you create a proof that the work was done and compute that instead Yeah. Of actually running unit tests or just doing the verification loop like, you know, OpenCLAR like, is moving away from, you know, all of its automated testing and moving towards that verification loop. I think it's gonna be some collection of those things.

Because otherwise, you know, even if you had unlimited cores, you're unlikely to want to do unlimited builds because it'll catch up to you regardless.

Yeah. Yeah. I haven't even seen that myself in my own personal compute. Just the the strain on the system of running all these checks all the time. And and we had a a meetup last night, and you would not believe how much of the conversation was around testing. What are you guys doing for testing? How are you fixing it? Like, I don't know. I don't know if I have any answers yet. Anybody knows yet. Yeah. But one question we had last night was

someone said, like, are pull requests dead? Would... Issues? No. But a lot of people are are closing pull requests anymore partially because they don't want the slop that comes in. And Sure. Partially because it's just, like, it's not helpful because I can I can also generate the answers to that? The helpful part is the the actual issue. Any thoughts there? Yeah. I mean, I think that poll requests are for, like,

closed group projects, meaning, like, this could be open source, but, like, you're not accepting PRs or, like, in the enterprise or, like, at work, I think still serve a purpose as, a transactional record of what was done, why it was done, how it was done. Yep. You can also attach all the exhausts to it so you can look at the agent exhaust to it so you can kinda look at that later. I think the thing that I was talking with some open source maintainers about was like, if we gave away, for you to basically, like, contribute tokens to you and say, like, this is the feature that I would want, and here's, like, a kinda, like, PRD, you know. And, like, instead of me just giving you homework, that you will have to pay for Yeah. I can go in, like, here's my subscription.

Mhmm. You know, like, go run it. Because I I constantly hear from open source maintainers what you said, which is like Right. I know how to do it. I have the harness. I have the system. Just let me... I... I'll agree or not to the intention of the feature. Yeah. Let me go do that. Yeah. Versus looking at your work and basically asking your agent to redo it or spend so much time verifying it that you're effectively redoing it. I do, like, wonder as we kind of move into the new era where GitHub can help us matchmake, not in a sort of like bounty or pledge way, like, with your, you know, but instead just say, I'm convicted.

Here's like a, you know, a token with like a nonce or something that you can go use to go do this one thing, know,

a Interesting. In like a new way for us to kind of build the open common without just giving us all more homework. Yeah. Yeah. Go get yourself something nice too. Go get yourself something. That's the other thing. In the same breath, these open source maintainers are saying, hey, Anthropic, OpenAI, can I have? Can you sponsor me? You know? Can I have open source tokens? So that's that's an interesting idea. I hadn't thought of that. Yeah. The GitHub Copilot app is looking great.

One thing that we're always talking about is, like, it's it's great, but then we also have all these other subscriptions and So stuff like I think one thing that we always ask is, like, is ChatGBT sign in and, like, other OAuth providers ever coming to the the Copilot app? I think for individuals,

I really would like to do that. Yeah. You know? I think that, like, you know, we're trying to figure out how to give that sort of, like, same experience that, you know, I use every personal agent under the sun, and I'm dragging my subscriptions across Open Claw and Hermes and whatnot. Yeah. I would love for devs to be able to give that, like, to give GitHub Copilot, and potentially, honestly, even like Copilot... GitHub Copilot Fusion a shot by just bringing your subs and letting us use them to, you know Yeah. Do the the mixing and being able to use the app. So, like, you know, there's more to be done, like, in trying to, like, figure out if this, like, is...

But not just technically feasible, but what, kind of, like, downsides there are for the user. All of our AI goes through a gateway that we call CAPI or, like, the Copilot API.

It's how we kinda manage where we go and, like, we have lower latency and, like, lower error rates going through that. So you can divvy up between a variety of providers. We couldn't give that to you if we're going through the subscription. We'd have to kinda always go, like, direct, and, there's some downsides to the experience. But, yeah, I mean, you can bring your own key right now. There are ways to kinda, like, work around it, but I would love there to be a first class way to just go, like, add the sub, add the sub, add the sub, and start to use it. Yeah. And I I just know along the same lines,

Fusion was mentioned. So is Fusion a model router? If... And if it is, like, I think we've heard both that model routers aren't super effective. So, like, what makes Fusion effective? Yeah. Yeah. Yeah. So

the complexity of is it a model router is that can it model route? And the answer is yes. Okay. Is it a model router? I would think no. Cool. So go... GitHub Copilot Fusion effectively uses a multitude of models to produce a single request. And it has three ways in which it does that. It can do that through a single model route. So, like, the task is so straightforward, I'm gonna go to a single model. That's effectively just model routing, looking at the task intent.

Then it can do two different modes. It can do cascade, and it can do critique. What cascade does is it starts with the lowest model that it believes it can use to implement the task, and it starts drafting an implementation.

And then it basically grades itself. Is this implementation looking to match the user's intent? If not, it can upgrade models. So it can then jump to a more, like higher intelligence model, potentially across multiple families. And it can do this multiple times for different parts of the task to ultimately give you the end result. And so in that way, it's not going like, you know, I'm asking it to do work and it's gonna choose a model. Maybe chooses the wrong model. It's using all of that, you know, using each of those models to get to a better result at the end of the the sort of cascade of models. Okay. Yeah. Critique is the third one, which basically says, hey, I'm gonna start working. I'm gonna use another model to check the implementation of that model and then sort of work into a loop. Something I think we've all done, you know, cross modeling. Modeling up. Or something else. And so those last two modes are the two that we currently have, but those will use multiple models to produce a single result, which gets us towards the fusion model idea versus simply, you know, like an auto model router Yep. Where it's choosing the intent and then choosing a model. That's how we get such better performance out of, you know, the multitude of models where we're, you know, two points off of OPUS five five max and ultimately can be up to 77% cheaper.

For GitHub and our customers, we're usually between like 50 or 60% cheaper Yeah. And 40% faster because those cascade modes and critique modes, create fewer turns. Okay. The model doesn't need to sort of like think, think, think and keep working. It can go, I'm wrong. And then, know, escalate to another model in the chain. Interesting. It's it's really it's really cool, but it's also quite different than if you've used like auto modes or other, you know, more kind of one to one model routers.

So... And and this is exclusively

inside of the Copilot app or is this something you could use in other harnesses? You you can use it across any place where you're using the GitHub Copilot harness. Oh, cool. Yeah. Because some of this intent is handled client side. Okay. And some of it is handled server side. And so we kinda need both in order to do it really well. Cool. Yeah. So any of those services though, you know, the GitHub Copilot apps, the one we showed today, PS code, CLI, github.com,

the cloud agents and so forth. We got a lot of surfaces. We, Yeah. The screen on stage is like, I don't know, probably sixty seventy feet wide. And it's like the craziest projector. And they filled it all up, and I just wondered, Scott, that's a lot of surfaces. You know? Like, it's not just the Copilot app, but it's it's in Versus Code. It's in the mobile app. It's on github.com.

And And there's, like, more that we didn't put up there because, like, you know, we support a ton of, you know, third party Tamagotchi integration.

Yeah. You got your Tamagotchi.

Basically. Yeah. You're part of that. Yeah. That's great. Yeah.

Let's talk about npm really quickly and just, like, the supply chain in general.

Npm has been feeling some creaks with all of this as well, probably because it's the biggest place to put stuff.

But I'm curious if you have any thoughts on, like, what the future of software dependencies

and the whole supply chain might look like. Yeah. You know, I think we're all kind of experiencing the month we've had of, you know, security, palooza of, like, finding, you know, unique and novel attacks. And I think, like, npm has been, one of the examples of where a package registry, you know, has been plagued by that.

We've been doing some work to solve that problem kind of very directly, which is, you know, we've invested in infrastructure where we'll take a package, we'll explode it inside of a VM and look at it to ensure, like, that we're finding potential malware that's within it. That's one way that we're doing this work. Then we have a couple of things that we didn't really talk about in the keynote, but now we can look at, you know, like, the package dependency tree. So, we can look at basically, like, as you're doing a a a like a dependable update or you're looking at, do I need this actual package and should I update it? We can track all the way down whether you are using that package, are using the effective code, etcetera. The thing that's more important, I think, for most of us is, like, the package firewall now. So, like, the ability to only receive packages that have been trusted or been published for a period of time before I go use them and be able to have that. I think that, you know, this is one place where we're using AI to help find the problems as they come up. But the reality is that so are the attackers. And a cyber model that helps us defend is also ultimately a cyber model that can help attack, which makes the process a lot more complicated. Yeah.

But like we're working to find more and more ways that we can give mostly, you know, like, open source maintainers or people who are publishing packages Mhmm. The same tools without going through all of the pain of

accessing these cyber models, you know? Because it's been quite painful in recent You think we're gonna get to a spot where we just, like, don't need as many dependencies as well? Like, we're coming from... We're JavaScript devs. Right? Sure. We npm install it. If if we need to do, like, a capitalized letter, we npm install something. Yeah. Talk about a ping pong demo. We could do that for our npm dependency. Yeah. But, like, a lot of people are just saying, like, I don't I don't need the dependency. I'll just vibe code my own version of that. And I I don't know if that... If that's that's gonna be the case. Any thoughts? Yeah. I mean, like, I think

I think the fact that, like, the sol... Like, the solution to do I need these dependencies is I'm just gonna write my own code, and therefore, I won't... I will be more secure and be able to be well... More well maintained. We've lived that life. Yeah. You know? And it doesn't work out the way we all think. Yeah. I do think that, like, you know, the benefit, but the problem with, AI code gen is that, yeah, it's one click away for me to be able to go and just, you know, generate something. I I kinda laugh with, you know, Frank, and Paulus from, you know, Home Assistant every time somebody's like, I'm using I'm using AI to, like, build a tool to, automate all my lights, and it only took three hours, and I Yeah. You know, 10,000,000 tokens.

And I'm like, just download download Home Assistant. It's right there. And AI's good at writing the YAML files Yeah. For Home

A 100%.... Than I am. Yeah. And so I do think that there's something that we can do both within our harness, but maybe through skills and whatnot. There... It's just... Sometimes it's just better just to go get a package. Yeah. Go get a package and let the agent automate either the use of it, the installation of it, or the installation of the entire kinda app. Yeah.

That way we can secure fewer of these endpoints rather than every time we have a, like, you know, zip compression algorithm shift or something Mhmm. You're all gonna do that. We're all gonna do that in our own projects. That's that's just,

not... Doesn't seem feasible to me. Yeah. And and in regards to npm, I think it can feel like... You know, GitHub gets a lot of attention. Right? And obviously, npm is owned by GitHub. Yep.

I think it can feel a little bit more quiet on the npm front. Is npm something that is a focus for GitHub? Like, you you almost never hear it mentioned that often. Yeah. Yeah. No. I mean, like, we we absolutely have people that are working only on npm that are focused on npm.

We also work across like the Microsoft sets of package registry providers too because like what... Know, whether it be kind of like official or unofficial or supported by either GitHub or Microsoft, like we work with them as well, both to secure, to fund, and ultimately do the work. So I think it's just a...

When GitHub acquired npm, you know, in order to continue to support it, we sort of suddenly became an owner of a single technology and, like, language where, like, we're so, like, you know, whatever language you want, whatever package... We don't care. Whatever tool you want, to be honest. Like, you can use any tool you want with GitHub. I think that's where sometimes we don't talk about it because our DNA is so open, you know? But we're we're not hiding anything. Like, we're absolutely,

like, you know, keeping it working and Yeah. Improving it, like I mentioned earlier. Yeah. So we had some questions from the audience. I think that was important because we're gonna get a lot of comments. So we want to make sure that like we get what the audience is going to be asking. So we had a question from Sasha here who says, please ask about the recent novel code storage architectures that being pursued by Anthropic with Scion or Cursor with Origin or Epic with Lore. And how is GitHub thinking about the future

of code storage within their own platform? Yeah. So I mean, like, we kind of shared the... Our thinking of, you know, how we're doing code storage. I would say that the the approach that we're taking, while it has some kind of like novel benefits, broadly matches the idea of, like, okay, we're gonna store the blobs in a particular place, you know, and it's gonna look more like blob storage and so forth. I think with Scion, like, Anthropic, you know, was chatting with us as they were building it and, like, ultimately, when they were going out, they describe it as well as like, you know, hey, we're building this for our exact way of working and at the scale that we're at. And they're like, don't... Like, don't... Do not try this at home. You know? And I think that, like, it's possible that that will be a thing, you know, that we'll all go do when we're doing 30,000 PRs a day. Yeah. You know? But at the same time, I think that, you know, when I talk to, like, open source maintainers or customers about what's your favorite kind of like alternative version control system these days or like one that you're looking after, we've always, GitHub included, been trying out different things or building different ideas.

I'm committed to continuously improving Git until there's obviously a better thing. And I mean, like, materially better, not a 10% improvement better, you know, before we would go, hey, it's time to go look at other options.

Because I take it very seriously when we change the API of developers or the Internet, GitHub. I I I directly remember the day we changed an API and Xbox Media Center did not work that day. Really? Oh, yeah. It was back in the good old days when I was, you know, working on APIs and platforms. Yeah. And so, like, you know, that... I... That's that's a real thing, you know? Yeah. And so just kinda going, we're gonna, like, throw out the the Git API effectively.

Yeah. That's not a a nothing decision, you know, for... Yeah. That's big. And you'd have to change the name too. You couldn't keep you couldn't keep the GitHub. Yeah. Yes. We... I've been I've been there since 2013, and, like, the number of times there's been, like, a really passionate discussion of, like,

like, should it be the hub? The... You know? Oh, yeah. We're rebranding rebranding to the hub. Yes. Oh, yes. We're gonna stick with GitHub. No. I don't like that. Man, that that story, we'll have to hear that story some other times because we do a spooky stories every Halloween of just terrible stories like that. Ryan asks, what are they doing to make it even easier to ship successfully with agents? Git, as of today, seems like it's very human centric, and the industry hasn't quite oriented around the best source control patterns for agentic devs, especially when you can throw swarms at things.

Like, totally agree. You know, the GitHub Copilot app is kind of where our focus has been, but we also have, you know, the GitHub, desktop app, you know, that that folks have been using over the years. And I think the goal would be that when I'm building an application and someone was telling me, you know, their spouse was building an app and, they had built it and then they came in and said, hey, you gotta put this into Git and that'll like make your life a lot easier. And so they go do that and immediately are on like a, you know, 45 to 90 degree, you know, hill climb to go figure out, like, everything I need to know about Git to go do this. Yeah. I think that, like, either with a set of skills or in a sort of, like, app experience, we'd love to be able to just say, like, progressively disclose to you the Git features that you need to know about while we'll choose the right ones kind of like under the hood. There's no reason why you need to know all these esoteric pieces of Git in a day to day use these days, especially if you're kind of like learning how to code or building side projects.

And so we have some experiments on it. If you go to like githubnext.com, like our research group, you can kinda see some of the things that they've been working on Mhmm. Over the last year. But it's definitely a goal to not have to teach you all the tiny bits and bobs. Even just like like a couple years ago when Versus Code got a sync button that just committed and pushed or pulled. You know? I was like Yeah. Yeah. Beautiful. No. No. No one cares. No. Just sync. Just sync. All we want. You know? So...

Totally. Agree.

Cool. I think that's all... Do you have any more here? I don't have any more here. We got four minutes left. So Yeah. Anything else that we haven't touched that you're you're passionate about you'd like to... For the audience to know? I mean,

I don't know if you all have been able to have a chance to, like, walk around or if they've locked you in the podcast studio. Oh, no. No. We've been we've been cruising. Okay. Okay. Like, you know, I think the thing that's, like, super important to me is that, GitHub, like, yes, we're always trying to push, you know, whatever the sort of frontier of software development is. But like, it is a... It's about the people. Like, it's about like meeting devs, seeing what they're building. And yeah, you might use a 100 or a thousand agents. But like when you talk to Peter about the things he's doing with Open Claw and then the tweets he tweets about, like, hey, why am I being stopped from doing those things? Like, it's about, like, what Peter's ambition, what Peter's passion is in Open Claw. Yeah. My favorite thing, and I'm I'm sure it's all over social medias or you could pop a photo in, but, like, the record store over here Yeah. Where we took so many of the different, like, open source projects and made, like, various records of them is just, like, such a... I don't know, like a a a like a passion point of mine that, you know, we we want to celebrate your work. Like, we want to celebrate the idea behind it and empower you to do that, like, as agentically as you're excited about, you know, and without hitting a scale limit there. But, like, in this era where we're all just talking about, like, which model and which benchmark in the harness and what the hell is a harness and, like, what are we supposed to be doing with it and so forth, it's just kinda, like, nice to be able to get together with, like, you two, the folks out there, and just, like, hear what you're actually excited about Yeah. And not what we all kinda... Like, talking about our hammers and screwdrivers and We had that meetup last night. So refreshing.

Am so jealous.

Talk

to people, folks. Yeah. That's great. Oh, it's great. Yeah. Yeah. And everything here is always about the people. I think every year we've come has been really great. And from Ashley making sure that we are all hooked up with our Yeah. Our swag boxes.

Like Unreal. The GitHub stars program, all these things is just so well executed. So yeah, we really appreciate it. Awesome. Wes, do want to do sick pics? Sure.

Okay. We have this thing, Onset Tax, which is called sick pics, where you get to pick something that you think is sick, Whether it is a chocolate bar, a little gadget that you have, a piece of software, literally anything in your life. Ryan Doll, creator of Node. JS, picked flying a kite. Yes. So do you have a sick pick in your life right now?

I have a, like, a leather bound, like, notebook that's from a company in Amsterdam. Like, I can send, like, the link out or something. Yeah. Yeah. Yeah. That, like, is just, like, really bespoke company that builds these, like, like, beautiful little leather books. I, like, love it because every time I, like, carry it out, it's, a nice little reminder of, like, the real world. Yeah. You know? And a little bit of permanence versus, you know, just, like, I'm not gonna be, like, start from scratch, but do it right, you know, in it. And so, I carry it around and, like, the team will sometimes make fun of me because, you know, everyone's usually like Yeah. Dictating around their phone or on a laptop.

But I find it really, really makes a big difference to just, like, taking a beat and getting my thoughts down on paper before, you know, candidly, snapping a photo and sticking it into an agent. Yeah. Beautiful.

Awesome. Well, thank you so much, Scott. I appreciate your time. Yeah. Thank you. Alright. Peace.