Welcome And David’s Background

SPEAKER_00 0:02

Welcome to Uplink, where we explore the world of digital infrastructure, uncovering the technology fueling AI and cloud innovation with the leaders making it happen. Thanks for joining us here in San Francisco. I've got David, who's the founder and CEO of XE.dev, which is a really cool company. It's not your first time sort of creating a company and scaling it. And I know Tailscale was the previous company that you were a big part of. I think you were were you CTO at Tailscale? I assume we did a lot of the coding on that side.

SPEAKER_01 0:37

That's right. I was the co-founder, CTO, and you know, worked on the original product. The three founders were all engineers by training. So two of us spent a lot of time doing a lot of programming. The third jumped in sometimes. Awesome. That's uh yeah. That's uh uh, you know, this is uh, you know, uh Tailscale's uh um in very good shape. I stepped back at the Series C so that I could play with LLMs. Oh do something new. So cool, yeah.

SPEAKER_00 1:01

And that's uh you know, starting something new around that. Why don't Tailscale's doing exceptionally well as as well? Um uh and before that you were a Google as well.

SPEAKER_01 1:09

Oh yeah, yeah. I worked at Google back in the day. I spent I spent longer there than I expected to. I think it must have been some seven years actually.

SPEAKER_00 1:17

These are great place. In in the bay?

SPEAKER_01 1:19

Yeah, yeah. I started in Mountain View and then ended up in New York for a bunch of it. Oh yeah. So it's uh but yeah, I learned a lot there. It was very educational. And were you coding and developing in your time there as well? Or is it yeah? It was an engineer role there. That's uh yes. It's uh I I still like to program, it's my favorite thing.

SPEAKER_00 1:36

Yeah. So well, so now you've found an Xy.dev. Um we came across you, uh, I think you've started just playing on our platform, which is uh cool. That's how we find lots of interesting folks, but you've had incredible scale, I think, most recently. Um we've seen, and I love your perspective on this, but we saw a massive change between I think like January, February, March, something around that time, where people just started coding an insane amount comparatively. I think Claude must have released something sort of late last year, and um certainly in inside our business we're seeing it, but externally what we're seeing is companies that have just sort of starting to scale so quickly. And maybe you can start there and then explain what XC.dev does and and

The Opus 4.5 Coding Shift

SPEAKER_00 2:20

sort of that that problem that you're solving. Um, yeah, and and how that came to be.

SPEAKER_01 2:24

Yeah, I I think you're right. There was a change in the industry, and uh, I mean, we can see it in all of our graphs, right? We know it's there. Uh, and I I would say that the thing that did it was Opus 4.5. That was that was the turning point. When before that, you know, my co-founder and I, Josh, spent you know, m more than a year before that uh pulling uh pulling programs out of models. Yes. Uh and it was uh yeah, you're just the quality of the programming. That's right. Late 2024 is a painful experience trying to make it work. Early 2025, you could really see like something's there. Like we can make this work. We spent, you know, we spent a long time trying to get it out. When you know the first thing we built was a uh coding agent-like thing around this. Uh and we we spent a lot of time thinking about like how to let agents do a lot of things simultaneously, which led us to Is this before you're in telescope? No, no, this was after telescope. This was before we launched Xe Dev. Uh, we were building a product called Sketch, and it was a coding agent. Um, hilariously, we started on it before Claude code came out. Oh, right. Uh yeah.

SPEAKER_00 3:28

No, we were we were you're inventing Claude.

SPEAKER_01 3:30

We were very well, we didn't. Uh we uh we over-engineered something that didn't quite work. Okay. Uh but the problem we were trying to solve was really around how to isolate uh a coding agent in a way where it could uh work in a clean environment and mess things up. Because what we saw with models at the time, the pre-4-5 models, is you have to give them a lot of leeway to let them get any good result out of them. And so we built a Docker.

SPEAKER_00 3:56

What do you mean by leeway?

SPEAKER_01 3:58

You gotta let them be root and you gotta let them run TCP dump and like let them install packages, like app get install stuff. You know, they they have to be able to just just give them freedom to work, to do their thing. Yeah, they get really good results if you give them more freedom. Uh, but of course, you give them enough freedom to delete your production database, they will. Uh and so you know, it's uh it's this and it's like kids. Exactly. It's in the carcass. No, it's exactly that. It's uh, you know, it's the the new intern who showed up is super enthusiastic and wants to help.

SPEAKER_00 4:26

Uh yeah, and he gets the keys to the core.

SPEAKER_01 4:29

Yeah, exactly. And so this is the balancing act of doing anything with agents is the more you give them, the better they get, uh, and the more you give them, the more damage they can do. And everyone's walking this uh this impossible tightrope, and all of the uh what happened, I think, with 4.5 is we reached a point with model quality where the constrained environments that Claude Code gives the model to work in and uh and then later codecs uh uh became functional. And before those existed, if you gave agents more freedom in a full VM, you could get more out of them, and that's why we were out there building that stuff. And we started actually with Docker, just doing containers. We quickly discovered that doesn't work because uh the um there's not enough freedom for them. I see.

SPEAKER_00 5:14

Yeah, then the VM gives more capability to spin things, could do things as opposed to a can a container is far more limited in terms of what it can deliver. Is that right?

SPEAKER_01 5:22

Yes, that's it exactly.

SPEAKER_00 5:24

Uh you I'll give you the layman, uh, my view of it, which is uh non-technical uh from a coding side. So yeah, bear with me on that, but it's probably helpful for the case.

SPEAKER_01 5:31

No, I appreciate it. I uh I jump right into the weeds every time.

SPEAKER_00 5:36

Uh the uh So a VM is far better than a container for an agent, in effect. Um is that and so that's I claim that is true. Yes. Yes, that's right. And so I think that then links back to you were probably

Freedom Versus Safety For Agents

SPEAKER_00 5:50

building a very, very efficient way of delivering a VM, like a very memory-optimized, very um, I guess a low-cost way to deliver that because of how you've optimized the actual software itself to deliver a VM. Is that sort of an appropriate statement?

SPEAKER_01 6:05

Well, I think that's I think that's it exactly. The problem is we now we need a lot of VMs.

SPEAKER_00 6:09

You just need unbelievable amount of VMs. It's funny because I I talk about you guys a little bit because I was like, you were solving a problem that didn't uh well that wasn't a problem um not that long ago. I think people didn't really care about the size of a VM because you just you don't need to optimize, you just like stick that in there. But all of a sudden, when you've got millions, I don't know, billions of agents at some point, all needing their own little VM to run these components, um, all of a sudden, every little optimization makes a big difference. And so the cost reduction and probably performance that you bring is solves that problem. Is that is that it is that sort of what's happened or I think that's pretty I think that's a pretty good description.

SPEAKER_01 6:47

Uh I know one or two people who would want to split some hairs there. Yes. And uh uh there's there's been a lot of people working on VM performance before us. Solving solving performance performance means many things, depending who you are. And so a lot of people have worked very hard on the scale to zero problem, and that's been around for many years. This is the uh I have 10,000 VMs, and most of the time 9,500 aren't doing anything. So I want them on disk, not doing anything. And people have built very sophisticated systems. Yeah, there's a there's a lot of ways to describe it. Uh scale to zero is language that comes out of the the platform as a service world, and they under the hood they do a lot of these things. That is not what we build because it's not what we need uh for agents, because the result of agents is we have more active idling or semi-idling programs, and we need full environments. Scale to zero always required uh making some compromise in the the computing environment uh so that you knew you could turn it off. We have to keep them on all the time. And so we're solving a slightly different problem, and I agree it's a problem that didn't exist beforehand. And it's very much it's entirely because of agents. So we started with containers, we moved to more isolated containers, we used a uh a virtual machine called Gvisor for that, uh, which also turned out to not be big enough to, it wasn't a general enough computer to solve the problem. Then we ended up with full-size VMs, and then we ended up with the problem of, well, now I need 50 of them. Yes, and uh expensive. Yeah, if they're five bucks a month and I'm just building small programs, this really adds up. Yes. And so we developed a uh a model where uh your VMs end up sharing a compute pool. So we can sell you a compute pool, and then you can run as many VMs in there as you like, and just the way you would containers on a computer.

SPEAKER_00 8:37

And is it the number of VMs is just dependent upon the performance that's getting punched out of each VM? Um, is that basically how I would think about it if I'm consuming it? That's right. And so in effect, I'm only paying for what I use in terms of that, but you can just deliver so many more VMs. And if this lazy VM function, I guess, has occurred or an agent's not doing much at that point in time, it's not utilizing any of the platform. So you're not really having to pay that much for that VM because your other VMs are starting to scale and it's sort of I don't know. Yes, it's it becomes that component. That's interesting. Yeah, and it's uh it's but you also also optimize a lot of the memory side as well, was it? I think you were telling me. It's mostly memory work, yeah.

SPEAKER_01 9:14

That's that's the challenge with an idle VM, is it's not using a lot of CPU, and so that's not your challenge. It's they use disk, but disk uh scales in a variety of ways, and we can handle that. Uh, the challenge is an idle VM is consuming RAM. And there are traditional VM techniques for this around uh you know, ballooning ballooning has been around for a couple decades, which is a way of asking virtual machines to give up their memory. Uh, we have a couple of new tricks uh for paging out VMs semi-aggressively and for packing them into machines, and that helps a lot. And that's that's what makes it actually work.

SPEAKER_00 9:48

Uh so right now, what is consuming your VMs? Like what is the um is it Claude's create is someone's creating a bunch of agents and they're just scaling through that? Someone's making a decision to use your VMs, or is we've even seen like Claude Code make decisions to use your like you know, almost like Claude-led growth or something like that. What are you seeing? Yeah.

SPEAKER_01 10:09

Yeah, are we right now are mostly selling to people. Uh, I don't think Claude is bringing us people yet. I certainly would like it to. Uh I've I've seen it recommend us, but uh it's not, you know, it's not doing it en masse yet. Uh the for us, there are three interesting categories of users. One is individuals, uh just programmers who have a lot of side projects

Compute Pools And Memory Optimization

SPEAKER_01 10:30

now. Uh, because you know, as a programmer, I used to have an Apple Moat filled with side projects I'm never going to build. And uh the nature of being a programmer is there's 10 things you want to build for every one you actually get to build. Now I get to build several of them, which is great. That's uh that's how I know agents work, is uh I have a lot more side projects. Uh the the second category is uh businesses who are doing plat internal platforms for their teams. And so this is they have a set of developers or developer adjacent users who they want building apps inside their company uh with agents, basically. And they want a platform that makes that easy. And why would you put an agent in a VM in that instance? You put an agent in a VM because the sorts of programs you're writing are small enough that the best way to develop them is to develop right in production. And so the VM is actually the serving infrastructure for your app for your company. It's like it has a website, it has a TLS certificate, we take care of all of that. Uh, you run a server on it uh and you serve out something just the way you would a spreadsheet, uh, you know, a little app for managing time off or on-call rotations, those sorts of things. Uh and your agent lives in the VM with your software and you work on it directly in production.

SPEAKER_00 11:43

Is that like a sandbox-y sort of thing, or is it like a I don't know, how do you think it's yeah, I I avoid the word sandbox, but uh a VM's a sandbox, sure.

SPEAKER_01 11:50

It's uh it's uh sandbox in the sense that it's it's a it's a fraud environment you're internet, but agent can't escape it, yeah. Agent can't escape. Okay. Yeah. Trapped. So that's that's uh those are so individuals, uh companies with teams who are developing apps for themselves, which actually are very interesting because a lot of them have small engineering teams and then larger non-technical teams. And it's the larger non-technical teams that are building apps. The smaller engineering team is supporting them, building support infrastructure. So this is a whole, you know, it's it looks it's more like Microsoft Office for a company in a sense of it's uh just the way everyone creates Word documents. Now everyone's building apps.

SPEAKER_00 12:26

Yes, everyone's building an app every time it spins up and instigates a VM. So if they weren't using your platform, what are they doing? What's what's it what's the answer at the moment? They Or they literally pay to spin up VMs.

SPEAKER_01 12:39

Yeah, they they have to build on some other platform at like five dollars a month per VM or more, yeah. That's one option. The other uh the other option is to build uh build the infrastructure yourself. Yeah, more more than one. Yeah, I don't know if you've if you know this about engineers, but I'll uh it's it's very common to look for an engineer to look at a product and say, oh, why is that even a product? I can build that in a weekend. It's just X and Y glued onto Z. I I've done this, I I know it's a thing people do. And I've heard people say that about you know Apple.

SPEAKER_00 13:10

CTO does that all the time. It usually builds things that we have a problem to solve and weird things. We built an org shot just the other day. Uh I was struggling with uh whatever uh platform it is that runs every workday. Couldn't get an org shot, so we went and vibe coded this thing and five. It's awesome. Yeah, that's uh that is that is what exe dev is for.

SPEAKER_01 13:30

It's uh and so there are apps.

SPEAKER_00 13:32

Y CTO is a customer of yours, no doubt, somewhere.

SPEAKER_01 13:34

That's uh that's that's uh if not, uh send them our way. Uh it's uh yeah, there are there are other for some subset of apps, there are other platforms that exist today you can use for those things. If you have software engineers on staff who can take the time to build the platform support around it to hook up your production databases in a safe way, uh to integrate it into uh the the elements of your system. It's it's the it's the the best building platform for that. And so that's what it's really for. There is there is a third category of people using it, which is really interesting, which is new, which is the majority fits into that second category, I think.

SPEAKER_00 14:14

Yes, that's right.

SPEAKER_01 14:14

But the this new category is very interesting, which is uh companies building products where the it's a traditional SaaS looking product for users on a website or something similar or an iOS app. But what they need is a VM per customer under the hood to run agents in or something like that, usually run agents. And they have to build a lot of infrastructure on existing platforms to do that, or that we can provide it for them. Got it.

SPEAKER_00 14:42

Yeah. And yeah, this is the So you've you've coupled the VM to the infrastructure as well, and that's how you deliver both of them. It's not like a software that you run in AWS or

Three Customer Types Using Xe.dev

SPEAKER_00 14:50

something like that, or I've made it a lot of different things. No, no, we use it would be probably very expensive, I'm assuming, or how does that play out?

SPEAKER_01 14:56

Yeah, we we run it all, uh is the idea. And uh yeah, we start the VMs and go, they're then they're yours and you do what you want with them. It's uh and our job is to start them quickly and uh manage the pool for you so that it's a lot of people.

SPEAKER_00 15:08

So if I'm a user, how do I pay you for that? Do I make a subscription for a pool or do I just pay on demand, or how does that all flow to you?

SPEAKER_01 15:16

Yeah, we're doing a couple of different things because you know we're early, right? We've got to figure out the problem. Yeah, we look exactly. Pricing and packaging is most of my day right now. Uh talking to customers, figuring out. It absolutely is. You know, I c I'll I've got to.

SPEAKER_00 15:27

We've got product market fit, now we've got to figure out how we price this thing.

SPEAKER_01 15:30

Yeah, I'll never complain about getting to talk to customers, especially when they're the ones sending me emails. It's a wonderful thing. It's uh yeah. The uh uh I've been on the other been in the other place where you you're not talking to customers.

SPEAKER_00 15:42

Customers doesn't matter what price you put on it, no one wants your product.

SPEAKER_01 15:45

Yeah, not a not a good place to be. Yeah. Yeah. So we actually we're trying two things. And one is the traditional on-demand, like you're paying for the compute you're using sort of model. Uh I don't think I I think people are familiar with it, but I don't think it suits their actual use case. I think what they need is to pay for a compute pool and say, like, I need this many CPUs and this much RAM, and then they run what they need on top of that. And I think that ends up being a much better deal for customers because they're effectively reserving compute.

SPEAKER_00 16:16

Yes, I was gonna say it's probably the important part because at the moment it's hard to get um quickly, as you know. Um, I think that's probably the right way. And I think you've seen a lot of the AI, um larger AI companies. I think um OpenAI is just going down that part with r reserving. I mean, that's today, tomorrow it could be a different product discussion, but um reserving instances seems to make sense. Uh or at least guaranteeing maybe an instance of availability. Or maybe you could put yeah, a cost per is a cost per VM, or is it more as just the instance of the compute and you can run as many VMs as you figure out what it is?

SPEAKER_01 16:53

We don't that is that is the pricing innovation we've got is we don't charge per VM for anything. We charge for the underlying compute. Uh the challenge is we're probably innovating too much on pricing on top of that, and we should just make that one innovation. And that's that's why I spent my days on this.

SPEAKER_00 17:08

Like you could just work out how to charge for a VM, but the problem is you can scale the VM, I suppose, as well.

SPEAKER_01 17:12

Yeah, I don't uh I don't want to charge people per VM. Uh, but you know, what we could do is price the underlying resources just like VMs on AWS or anywhere else, and then uh then have as many VMs on that as you like.

SPEAKER_00 17:25

I think that's a fair people would find that a fair way to pay. Yeah, maybe it's like tokenesque or like you just pay as a how you get the token out of it, but the equivalent of the Yeah, that's uh I mean you can You can slice it a million ways.

SPEAKER_01 17:41

You can do it per second. Yeah, yeah.

SPEAKER_00 17:43

And then then it gets up sort of token-like.

SPEAKER_01 17:44

Yeah, that's true.

SPEAKER_00 17:45

That's uh the problem with per second is people then can't guarantee that they've got that capacity with you or something, and so then that gets tricky. It's also hard for you to invest in the capacity as well if if you don't have that. So it's 100%.

SPEAKER_01 17:58

That's uh this is why I think a model where people are paying per month means that they're doing some degree of reservation, but they're not making a multi-year commit. Yep. Is really nice. I know you guys do per month for a lot of work, and I think that's great.

SPEAKER_00 18:12

We do hourly, monthly, yearly, multi-year, yeah, whatever. We can sort of break it down. But usually if people need large quantities, yeah, we we we would do like multi-year contracts for super large. Um Do you run into issues with people doing bursts on hourly? Not an issue for because you

Pricing Without Charging Per VM

SPEAKER_00 18:30

own the compute, like you're you've got it's all yours, so you can do it, you can push it as hard as you want. Uh we don't care, in fact, that's a good thing because it means you're utilizing the platform itself. But you could get to a point where you know we've got 30,000 CPUs of like in the pool, and if everyone starts consuming them all, you could run out of CPU. That's what I mean. Yeah, it's a lot of the physical component. We keep procuring lots and filling the um just buy more. It's a green answer. Yeah, well, we we actually have what we call we actually have had to split the company into two most recently because of this problem. Before, people were scaling sort of, you know, 10%, 20% every year, maybe it's every month or whatever, but people are doing like 100x in a week. Um, and and so what we've gone is all right, we always want to have the pool availability, and then when a customer wants a lot more, we'll do a contractor piece where we go and procure a large quantity for that particular customer and scale that, manage that for them. Um, and customers are needing to do that anyway because they need to shore up demand. They've also got this problem that things pricing keeps changing. So to try and get an actual consistent price for them for some certainty, it's actually easier for us to do like a larger contract for someone. So we've got thousands and thousands of developers that sit on our platform constantly spinning up per hour, whatever it is. Um, and then we've got companies that have just gone and scaled their company out of nowhere very rapidly, and they've gone, actually, we need a hell of a lot more. And so we can actually solve that equation as well, which I know, I know you guys are doing, and and and and we get to see your we see all these smaller companies that have come from nowhere, we see their sick their incredible success. What I am astounded at is how quickly that success is happening at a scale that I mean, we've been in the tech industry for a long time, been in that understood the startup space for a long time, watched quite a lot of startups, seen them sort of scale, and this is just unbelievable. Like I've never seen anything so fast. We're changing before, where like you're scaling so fast, and you were like, Am I scaling fast enough?

SPEAKER_01 20:24

It's like, well No, I I genuinely don't know. It's so it's so hard to see through all of the the noise. Or uh I I I guess I should be here, hearing through the noise, shouldn't I, not seeing through it.

SPEAKER_00 20:36

But yes, that's a head down, charge through it, and just see what happens. It's uh it is distracting because you look over here and someone's announced something and someone's announced something as well.

SPEAKER_01 20:45

But yeah, and you have to wonder how much to believe some of that stuff. There's a lot of there's a lot of marketing hype going on right now, which I'm trying to ignore. A lot of chest beating. I I do uh I have heard some private numbers from some other companies. Uh it's clear people are doing really well right now. Like and I think this all comes back to Opus 4.5. Yeah. That was change. Yeah, and it was a very uh uh in retrospect, I think it's really clear. Uh, well, we made it a lot easier to write computer programs. We made every engineer worth, you know, an engineer and a half or four engineers or some, I have no idea. But we definitely increased the productivity of engineers, which means there's more programs. And if there's more programs, we have to run them, and that means we need more compute. And and uh, you know, if if you take that as a given, then this is not a uh transitory effect. There's just more computer programs to run now. Yeah, and so this is a fundamental shift in demand.

Handling 100x Demand Surges

SPEAKER_00 21:40

But I I cannot see it throughout. I only see it accelerating. I almost see that also um whilst the adoption seems fast, probably to you and Silicon Valley, particularly SF, there's so many corporates out there that have got thousands and thousands of developers who've not yet been unleashed into the um yeah, and that and they're probably withheld or even. they're they're so worried from a whatever um appliance standpoint they haven't even been unleashed into that space and they will at some point or slowly and the more that that gets unleashed the more scale that that comes. So I've only seen probably the first wave of like what is a set of lots of waves coming. And um yeah so you these problems are going to get exacerbated. So companies like yourself solving this particular piece is just going to continue to scale. Could I ask you something about um are you public on who your investors are and how you've are you yeah I think we've spoken about some of them.

SPEAKER_01 22:30

Oh yeah we're in your yeah we announced we announced our series a uh in a in my usual convoluted way uh a few weeks back. Congrats. And uh thank you yeah we our major investors are uh Amplify uh CRV and HeavyBit which are cool uh investors I've known for many years uh and uh stepped up quickly and enthusiastically to get around again they were with you in the TailScale although you at least come across in prior or Amplify weren't but I've been yeah I've known them for many years.

SPEAKER_00 22:59

But CRV and HeavyBit were both with us for Tailscale and wanted to uh get in again uh which I'm very grateful for no well I mean you're becoming a serial entrepreneur is just like an annoying word isn't it yeah that's uh but it is good but too too many syllables for me it's uh five dollar words on my you need to dig that down and put it on a tiny VM and yeah yeah yeah yeah and call it yeah all right I'll give me a lot one last one before we we wrap up give me some what's your predictions on the the market as a whole like what do you

Code Review Breaks And QA Becomes The Bottleneck

SPEAKER_00 23:30

see so hard to predict but let's just say by the end of the year do you see a new cardiom release that makes it even faster or better or um I don't know what where do you where do you think the world's headed in the short term? Long term's pretty hard. Or maybe long term's easy. I don't know which one you no no long long term is impossible for me.

SPEAKER_01 23:47

I actually couldn't help it yeah yeah the uh I think you actually hit on something very important a moment ago which I actually gave a talk on recently which is that uh there's a large number of engineers who uh can't take advantage of the productivity of these agents yet and there's a lot of reasons for that and many of them are uh process related and so you know that can't do can't do shit yeah there's also engineering processes behind this too so we have these traditional processes around code review in engineering uh which no longer work with agents and companies haven't adapted at all yet no one has a plan i i've speaking to a lot of people at big companies about this we we've got we've got this exact problem I'm keep going but um yeah we we we face the exact same problem yeah it's uh it's clearly a problem and a lot of engineers at larger companies don't even realize what they can't do right now because they they're stuck in this trying to if you take an agent and you put it into these processes you make everyone's life worse uh and you don't feel more productive at all and that's where a lot of people are stuck right now and you know we I was programming before these processes were invented uh and like they weren't well I mean you know they they they existed in in some places but uh they hadn't they hadn't gone mainstream yet uh and it's clear to me we have to change those things and so in the medium term there's gonna be a lot of reorganization around how engineers at large companies work totally and to relook at the whole thing yeah the first companies that actually get on with it and figure it out are the ones who are going to win the productivity first from agents and I haven't seen a single uh any company over 200 engineers yet that's actually figured it out and so it's so hard there's a there's enormous uh there's enormous productivity gains coming in the future yes uh in the near future when we figure this stuff out yeah so I agree with you I'm looking forward to that we have our brotlenecks become QA so the amount of codes getting produced but then to review the code we used to have a few QAs and they could withstand it now it's impossible.

SPEAKER_00 25:49

So we've actually had to sort of in the short term I don't know if this is right we've got our coders reviewing each other's code because we've had to turn everyone into this like there's so much coming through no one can withstand it. And so I don't know maybe maybe agents get so much better at QA that this problem goes away.

SPEAKER_01 26:04

I don't know but they can definitely help. Yeah we we've invested very heavily in end-to-end testing as a technique which in general before agents we didn't do very much of uh and now it's the basis of how XI is built and that is effectively moving a lot of QA into automation.

SPEAKER_00 26:21

We've got a really interesting customer called Blacksmith I don't know if you've ever come across how I've heard of them. I don't know much about them. Yeah they do CI and they do it incredibly quick you can put the entire code base in and run a test very quickly on high performance infrastructure and then sort of so you can constantly just run these tests instead of test against the code base so that you're not waiting half an hour and just constantly so every agent.

SPEAKER_01 26:42

So this is I mean that they do every this this problem will be solved in some way by someone cool and innovative and um that's a that is a big problem in the industry in the software industry right now is the the GitHub actions bottleneck everyone is stuck on. And so that's actually what blacksmiths is solving for good someone someone

Who Should Try Xe.dev

SPEAKER_01 27:00

needs to fix this. 25 person organization nice that's I mean it's crazy you have to start somewhere no yeah yeah that's great.

SPEAKER_00 27:08

Well congratulations on what you've built we're really excited by it. How do we help share your your message out there wide or who who is who should we tell that should look at your product anyone coding or anyone using an agent or what we anyone who wants to try building a program and sharing it with people with an agent should give us a try.

SPEAKER_01 27:26

It's it's very easy to just you know jump on the side exactly give it a try it's uh it takes takes five minutes to get something set up yeah yeah it's a pleasure thank you David yeah really appreciate your time thank you cheers