Are you spending too much time on Twitter waiting for CallCode? Like, that was the original launch post within the first week. You know, we had, like, 10 founders using it for fourteen hours a day. And we're like, this is more traction than the thing we just built for four four years. It's kinda like start up university.

It's getting beaten up over and over again.

I'm joined today by Louis from Vibe Kanban. In this episode, we talk about why everyone's wrong on AI pull requests, why Louis walked away from 6 figure deals, and do we even need software developers? Enjoy the episode.

I think, like, one of the rules we've got at Webcam, Ben, is, like, don't build anything that the Clawd Code team is probably gonna build in the next three months. Mhmm. You know, because, like, I think if you take a twelve month view, you could arrive at the conclusion, like, almost everything is gonna get built by this team. But if you take a three month view, you're like, okay. Well, there's so many other things that are obviously high priority, and you can just kinda see the direction of of of where the product's moving.

And then if you if you put everything through the len through the kind of filter of, like, does it pass or fail the the twelve month test? Everything fails. And then you're like, well, we can't build anything. We may as well go and become carpenters or something. You know?

Yeah. So maybe that's helpful So

you think of, yeah, like, they're not gonna build it imminently.

Yes. Yeah. Yeah. Everything is going to get built at some point by the AI labs is is like that may or may not be true. But if you just accept that and look at the kind of competitive landscape more as, like, there is a sweet spot of things where it's a good idea, and therefore that means it's probably on the lab's road map.

But it's not an immediate thing that they would work on. I think that's kind of the the intersection.

Yeah. I mean, like, lovable, I think, is the obvious one. I think at the time, everyone was like, they're just gonna build this way. This is a fad. Like, why why haven't they, do you think?

Do you have an insider?

I mean, I don't know if the answer's a product answer because when you look at ChatGPT now, you can go on. You can say make me a website, and they've got a button that says preview, and you can see a website. And I can even send a link, you know, to anyone with my preview, and it's it's like a working website. Right? But people still choose to use Lovable, and there are some product things that you can't buy a domain through ChatGPT yet.

Maybe they'll have that. You know? Who knows how soon? But I don't know. I I think I think a lot of this comes down to you as a founder versus product manager number 50 at a big AI company.

And you see it time and time again, especially when it comes to distribution, you know, this kind of founder led I'm in a post on Twitter and, you know, like, what's, you know, what's the kind of OpenAI equivalent of that? They're they're promoting their kind of core product. They can't do the founder led marketing for 50 products at once, can they? So I think it's a combination of focus and just how distribution works in the current environment.

I guess if we thought of it of, like, why wouldn't Google build this? It it feels like there's an obvious answer that maybe we're holding or at least I'm guilt like kind of holding them in this kind of mythical place where it's like they can do anything and I don't know.

Yeah. Yeah. Holding OpenAI.

OpenAI. Yeah. And they're kind of they are a lot of the other labs. Maybe not I don't know. Because it feels like they were on role.

Maybe maybe it's less now, but I don't know.

Yeah. I don't I don't know. I guess what are they most interested in as well? You know? Maybe we live in a massive bubble, and these companies really focus on what what does AI look like for the billions of people, not just the kind of actually quite small minority of people that want to build a website.

Right? And so the Sora app is, like, a good example of this bet. You know? I don't know how many Americans have TikTok, but probably, like, a 100,000,000 people have TikTok. Right?

And that's a bet that they're much more you know, if they can win that, I guess that's just got the potential to be a much bigger thing than whatever their equivalent of a lovable clone would be. Right? So maybe that's another way to look at it. And they're also kinda unencumbered like, it doesn't matter to a big company with existing distribution like OpenAI whether they get there first in the way that it does as a start up. Like, as a start up, you can't come from behind and and and and win.

You have to be early on something because you're betting that you can kind of get distribution, get to escape velocity before the incumbents kind of take over. Apple's great example of this. Right? Weren't first to the smartphone, weren't first to well, certainly not first to AI, but as soon as they roll out Gemini and Siri, like, that's gonna be the way that most people use AI suddenly. Right?

Because they've already got the distribution. So it's it's not I don't think OpenAI is sitting there thinking like, you know, damn. We we missed Lovable. Think they're just they're not actually it doesn't matter whether they ship it this year or next year to them. Right?

Because they've got

the distribution. Then they can come from behind.

Exactly.

Yeah. Scaling DevTools is sponsored by WorkOS. If things start going well, some of your customers are gonna start asking for enterprise features. Things like SSO, SCIM provisioning, role based access control. You could spend ages tearing your hair out, building these things yourself, or you could use WorkOS.

Will, what do you guys do?

My name is Will Stewart, cofounder and CEO of Northlink. We're a self-service developer platform, and we help teams deploy their most critical workloads into their VPC.

And you guys use WorkOS?

We use WorkOS for our SAML and OIDC integrations. It's a pretty exceptional product. It it makes everything regarding authentication pretty seamless, and, it's been instrumental for us to onboard our enterprise customers much faster. Building integrations with lots of different SAML providers is really challenging. We can do that for cloud providers.

That's our job. But we don't wanna do that for, the other type of IDP. This is quite interesting because internal developer portal platform, and then there's the identity provider. So IDP has three different meanings, and we're internal developer platform and WorkOS is IDP for authentication. But, yeah, it's a great product.

Thanks, WorkOS. Back to the episode. Okay. Speaking of the value of being early. So when we first met, you're working on Bloop.

Yes.

And you were I think you were very early.

Yes. Too early.

Too may okay. So talk to us about the stages of earlyness because I felt like maybe Vibe Kanban was like the the right level of early.

Yeah. Well, I think all the things we did before Vybe Kanban just meant that when we started working on Vybe Kanban, it was really early because we had just been immersed in the bubble because of all the stuff we'd built before. So I guess just to recap, started the company in 2021, training small embedding models to do code search. And this is in a world before GitHub Copilot had been released, before GPT three or maybe GPT three had just come out, and certainly pre chat GPT, you know, ClawCode, all of that. So we had to adapt very quickly when the GPT three Turbo came out and chat GPT moment happened, and we started moving up to the kinda application layer and just building software on top of other people's models so we were no longer training our own or fine tuning our own, embedding models.

And then we moved in services for a bit. So we're doing legacy code modernization for a year. And at some point, that wasn't really working. We just saw the sales cycles were gonna kill us relative to how much runway we had and and, you know, the milestone that we would need to meet to kind of get funded, and and raise the next round. So we were looking for a new opportunity to work on, and that's when ClawCode came out.

And we thought, okay. Well, clearly, this is gonna be a big shift. And all the big companies in the space that have already got a foothold have ridden a big shift. Now you look at Cursor, it was the first to kind of implement that concept. Same with Devin, same with Lovable.

And probably many outside of the coding space as well basically cemented their lead by just being first and executing really well. And it was just a kinda overnight decision to just build something that we'd been playing around with internally, which became Vibe Kanban. So we saw that there were a bunch of problems with just how you use core code. Like, if you want to run two of them, you have to open two terminals, and that just doesn't feel like the future Yeah. Of of coding.

And you very quickly find that your bottleneck when you're working with coding agents is your ability to cognitively balance review and planning. And so how do I kind of create an interface that makes it intuitive, you know, that I should be reviewing something that AI has just worked on rather than me looking at kind of green text on a black background terminal trying to remember, like, okay. Well, what is this doing? Having to run some commands. It's like, let's try and remove the cognitive burden and just surface.

Okay. AI's done some work. Human, you just QA this or look at the code or whatever. And so we launched it to the YC community. And within the first week, you know, we had, like, 10 founders using it fourteen hours a day.

And we're like, wow. This is, like, more traction than the thing we just built for four years. Yeah. Yeah. And I guess that gave us the confidence to, like, double down, and then we we cleaned it up a bit, released the open source version.

We're now approaching I think I think we'll we'll hit 20,000 stars today, actually.

Wow. Okay. Cool. Which is which is cool. Amazing.

Yeah. And it's very different as well. Just very maybe very quickly on on the community thing because we we had an open source community around the first product that we released, which was the the code search engine. But, you know, had a lot of stars, had decent downloads, but was not that interactive. Like, people weren't contributing, and they weren't logging into the Discord.

We had, like, a 100 people in the Discord. And now we've got thousands of people in the Vibecan Band Discord. It's like night and day. You know? It's like catching a fish.

You always think you've caught a fish. But when you do catch fish, there is no doubt in your mind that you have caught a fish. Yeah. And it's a bit like that with the community with Vibecam, and, like, we it it feels so different to everything else we've we've done previously that yeah. It's it's it's really exciting.

Yeah. There's a like, this is the cringest thing I'll ever do on this podcast. But there there's a there's a Lana Del Rey song. Oh, god.

She's gonna sing it.

Yeah. Yeah. And one of lyrics is like, if you know, you know. Yeah. If you have to ask, the answer is no.

They get it. It's like

It is a bit like that. Yeah. I don't have to ask.

Yeah. So I I feel like this is one of the things that, like, I in a minor way, I felt this for a few different things, and I've I've but I've had so many other things where it's just like no one gave a damn. And I think there is that huge difference between, like, stars, people liking it, people going, oh, this is cool. But, like, versus, like, people spending fourteen hours a day

and stuff. 100%. I guess it's a tool that, you know, you use it's it's such a core part of your workflow, you know. It, like, is in the same category of things as your IDE. Right?

And so if something pisses you off, you're going to want to change it, and you're gonna you know? And so the ability to just make that change, especially with Claude code and and all the other coding CLIs, Now, you know, we have loads of people opening small changes that are just, like, quality of life improvements, things like that. And so I think it is important that, you know, we have that, like, really close feedback loop with the people using the tool for this type of product. Because if you're using something for fourteen hours a day and, like, something's pissing you off, you're gonna churn. So I think this lends itself quite well to that model.

Is it kind of fairly obvious what you spend your time on versus I would imagine before it's like, how do we how do we market this? How do we get people?

Yeah. Like what is your question?

Like what are the steps? Well, guess the question is like, does it then is it once you find it, is it like relatively straightforward to just know what to do, what to work on, and just hard work versus that's what I would imagine it's like. Is that fair? Or

I think I think the road map is kinda clear. So so maybe we can look at this in a few ways. Like, one is is what do we build? And I would say the backlog has been pretty consistent for the last few months, and we've been just working on that. And it's a question of how do we how do we build it faster?

How do we hire the right people? At the same time, you know, there there there are some forks in the road, especially around how far do we go down code review, you know, things like that in a world where that's obviously, like, an incredibly competitive landscape, and there's lots of people who wanna kill GitHub, and then GitHub wants to, like, improve their thing. And, you know, and then on the other end of that, there's how deeply do we want to integrate with, like, existing project management tools? How much do we wanna build up into that and kind of become a competitor in that space? So I guess, you know, it's there are some decisions, but it's not fundamental, like, product market fit questions, I guess.

It's more resource allocation at the end of the day for us. And in terms of marketing, like, something we're very keen to do is get teams using the product more. So we currently have a bunch of people working in companies where, you know, they have, like, a team within a large company and everybody uses vibe Kanban on the team, but we don't really have great teams support yet. So I can't see what your call code does. You can't see what mine does, etcetera.

But that that's gonna check. Maybe by the time this comes out, like, we actually will have shipped that. It's it's kind of dropping imminently. So that's the the kind of the the big unlock is once we release version one of that, be a bunch of new marketing that we have to do because everything else has been this product led growth stuff. And then suddenly, we're selling to teams and so it's bank maybe hitting up the phones.

I mean, I I think the foundation of all of the marketing will always be the the open source, people adopting it as an individual, and then hopefully spreading that within their team. But this is something that we might experiment with is, like, just jumping on a flight to San Francisco and sitting down with a laptop in in a bunch of people's offices and just showing them what the product is and what it can do for their team.

Yeah. And is it like, are you seeing it that it's devs and will always be devs that you're building for? Or is it yeah. Yeah. Yeah.

I think

there's I think there's so much to build for that space. I think this is one of the failures of other tools in the space is that they are too broad. And the workflow if it's all about if all human skilled work becomes basically plan and review work that has been done by AI, then that's very domain specific. Right? Like, the way you review code is very different from the way you review architectural designs or whatever.

And the speed at which you can perform that job as a human is gonna be completely bottlenecked by how solid the experience is, you know, and how tailored it is. So, yeah, I'm I'm very bullish on on profession specific orchestration tools. I don't I don't buy that there's gonna be one general kind of suite of tool that wraps AI for all professions.

Okay. Okay. And you you're thinking that devs will continue to be need a head end. Okay. The big question though.

This feels like rage bait.

No. They're like, gone.

We are so far away from a world in which AI is able to, like, plan its own work. So, you know, like, there are some interesting startups where they take you know, they add a feedback form to a website, and then they spin up call code in the background to, like, deliver what people type into the feedback box or whatever. But we're so far away from that being useful that you definitely need somebody with a product mindset to kind of, like, take these inputs, which are, like, user feedbacks and pass them into outputs, which are actual tasks. The two are not the same thing. Right?

You can't just automatically turn every user request into a task. That's one end. The other end is the review. And, you know, there was some invisible threshold that was crossed, maybe with, like, GPT five and Opus 4.5, where everybody suddenly realized these things are more than 50% reliable. But there's a big difference between these things being reliable enough that humans don't need to be in the loop reviewing things, And it's like a 99% problem.

You can't, you know, you can't take humans out of the loop of reviewing software for a neobank or, like, the average business if you have 90% accuracy. You need 100% accuracy. So I think software engineers are gonna be very relevant and people are gonna be building more. And so, like, we'll just need more software developers. I don't know how you kinda solve the the short term lack of usefulness of of junior devs.

Like, it seems very obvious if everybody's just reviewing code, then you need experts who know how to review code in the driving seat of this stuff. And that that's the real problem with junior developers is, like, they're just not able to you know, they maybe don't command enough experience to be able to give companies the trust that they wanna hire those roles right now. But I think it'll come back because ultimately, at the end of the day, we're gonna be building more software. Software. We're gonna need to review more, and then we're gonna need to onboard more devs to the industry.

So I think we're in a weird in between phase before that kicks in.

Yeah. And how how do you think it's changed the way that devs work with vibe cam vibe cam down versus like, say, in the properly old days? And then also even just when people are using like one called code instance and just

The properly old days being what, like, 2021.

Yeah. We could say we could we could finger it from that, I suppose. Or maybe we could just start from everyone knows what it's like.

I guess one of the things that is is interesting, especially with having, a big Discord community is there's no common way. Like, you ask me how has it changed, and the reality is everybody's doing different things. There's not one way. The way I advocate for because this is just how I found my own work to be quite productive. You can imagine it's, like, full stack, Rust on the back end, TypeScript React on the front end.

So I think this workflow kind of works fine for both of those things, is to paralyze your work as much as possible and try and get the AI to work for as long as possible before it needs to yield back to a human. So if you look at the the the brief history of, like, AI assisted coding, you know, you have GitHub Copilot, which takes a few seconds and gives you a line of code back. Then you have Cursor, which can complete a paragraph in, like, you know, 2023. Obviously, you can do a lot more now, but, you know, just talking through the timeline. Then you have, you know, poor code eventually, can do multiple files at once, read them, edit them, and then also it can run scripts.

So it's starting to run the type checker and other scripts. And what is the direction of travel? Well, it's like you've gone from a few seconds waiting to do something as a human after AI has done its work to now having to wait a few minutes. And, you know, this year, I expect AI is gonna be able to, like, reliably start a dev server, probably open a browser, click around on some stuff, test the change a bit more thoroughly. Probably not all the way, but, you know, first pass, second pass of the testing, I expect will be there.

And then you're into, like, ten minute, twenty minute, maybe hour long territory before you as a human actually need to kind of take back control and do whatever your task is, which probably review something. So I'm I'm looking slightly to, like, with that framework in mind, I'm like, okay. How can we prepare ourselves for that world? It's like, okay. The current capability almost works well enough with, you know, with with with tasks that can run for five or ten minutes.

But I think, really, I guess, the workflow that I advocate for is something that I expect to become the default once these things are running for a slightly longer period of time. Then there's all the other weird stuff like, you know, Ralph loops and stuff like that and people using, you know, like, agents that command other agents that sort of break things down. And to be honest, I haven't spent a lot of time really looking at that in too much detail because I think we're so guided by just what is what are the majority of our users doing? They're basically using ClawCode as it's shipped in the in the CLI. And to kind of focus on, I guess, the minority of users would be to to maybe build a subpar experience for the majority of our users.

So

Yeah. For instance, yeah, chat I was experimenting with ChatGPT Pro, and that definitely has, like, a 20 sometimes timing. Like, how are people actually this is maybe very tactical, but how are people not getting very distracted? And I I find it very hard to just like rifle off a really long task, which I which I have no idea how long it's gonna take. And then I'm like, I often just basically wait and just mess around to be Yeah.

You're on Twitter.

Kind of. Yeah. Sorry, Damon. Yeah. But, you know, you kind of go off and you do something else.

And I sometimes I what has helped me is, like, having, like, sound effects, like, in the terminal and stuff. So you get, like, a little ping when it's done. But, like

That was the original launch of Vibe Kanban, I think, was are you spending too much time on Twitter waiting for call code? Like, that was the original launch post back in, like, June 2025. And I think the very first feature we added was a cow mooing at the end of each execution run so that you would remember so that you would know to, like, get off, you know, whatever it is you've doing. Go back in. And we tried lots of sound effects, and everybody sort of, you know, every every sound effect faded into the background except for the cow mooing.

Really? Yeah. And a rooster crow.

So Okay. Okay. Okay. Okay.

But, yeah, I think I think there are loads of things that people can be doing while they wait for AI to do its thing. The obvious thing is, like, think through when this is done, what am I gonna do next? Yeah. And so that's what we talk about when we when we talk about planning is, is, you know, AI generally is as good as the input you give it a lot of the time. So if you can be organized and write down, like, a really detailed spec, you're gonna get better results.

And so, you know, kind of guiding the the human to do that. Okay. You've just started something. Now let's go and write another ticket. You know, that creating a workflow around that just means that, you know, you're you're less likely to end up wasting time looking at the logs or on Twitter, whatever.

Review is obviously the other one. So, you know, I've probably got four or five instances of coding agents on the go at any given time. They're not running for twenty minutes. They're probably running for five or ten minutes at a time. And one of them finishes.

I'm starting a dev server. I'm testing the change. If the QA goes well, I then move on to the code changes, and I do a I do an AI review now, and then I do a human review of that. And there there's always things you can be doing. There there was kinda really interesting experiment that somebody took the database that Vocabulary use under the hood and created a visualization, like, almost like a calendar of all of the different things that you're working on.

So create like a calendar event to say, call code was running on this task here. And you can see if you've got four things in parallel that you've got, like, all these different calendar events overlapping as you kinda go back and forth. And it's it's quite easy to keep yourself busy if you just accept that you you have to run more than one thing in parallel. If you just run one thing in parallel, you're gonna end up on Twitter.

Yeah. Yeah. Yeah. Okay. This is great, actually.

So okay. So Vibe Kanban, 20,000 GitHub stars, as you said. Massively popular there. How have you found the kind of open source versus SaaS kind of play.

Yeah. We just want to build a great tool. I know this sounds cheesy, but honestly, it's it's just there is only one thing that is important, and that is figuring out, like, what is useful. And so the more feedback we can collect, the more we can enable people to open PRs and make changes, the faster we will, you know, we'll achieve that. And and as a small company in this little tech outpost called London, you gotta look for these, I guess, unfair advantages that, you know, we otherwise might not need if we were based somewhere like the faucet of AI culture, San Francisco.

And so, yeah, it just gives us a gives us a huge leg up. And we obviously do have a SaaS version in the works that will probably be out by the time this is published. But it's just a it's just a hosted version of the open source product because the philosophy is the same. It's like, okay. Well, people should be able to improve that as well.

There's no reason why the Teams version of this, you know, won't be something that people spend fourteen hours a day like they do in the individual version of this. And so there's gonna be things that they wanna change, and we wanna be able to solicit that feedback in the easiest way possible, which is be open source.

Yeah. One of the things that I've seen a lot of people talking about with open source projects recently is AI pull requests. Have you had this

No. Issue? It's great. It's great. It's like somebody took the time to, like, you know, tell Claude to do something and then spent some money on tokens to create a pull request.

I mean, it's like it's almost the new bar for what I consider a serious complaint or product request is, like, a kinda half baked PR. And I'd say more than half of the things that we ship from the community, we we take the kind of the AI slot version, and then we just clean it up and we take it over. And, like, sometimes maybe 50% of the code survives. Sometimes maybe less. But it is so much better than just, you know, seeing people on you know, who just complain on Discord.

And I guess the difference is when people use Claude to, like, try and make the PR, it also means they're able to test the change. And so there's often some refinement, or they they don't even open it because they, like, ask Claude to make the change. They test the change. They realize, oh, this is bollocks versus having some sort of weird academic conversation on a on GitHub discussions or Discord. It's just so much it's such a better filter than, than than that to just be able to build the feature and discuss that.

Okay. Why why do you think this is a different view to what a lot of other people have been saying?

I don't know. Other people are wrong. It's very simple, isn't it? A failure of them to be organized enough to deal with their community. I mean, hypergrowth is tough.

Right? Some of the you know, we're we're at a we're at a scale where we have over a 100 open pull requests, which I wish was smaller, but some of these projects are growing much faster than that. They've got thousands, but you you have to allocate resources to fixing those things. Otherwise, you know, you you have this you have to balance solving the urgent and the important things. And not everything urgent is important, but, you know, if something is is really annoying somebody and they're using your tool fourteen hours a day, like, you wanna get that patched even if it's not strategically useful to you.

So we we try and balance doing a a mix of

both of those. Yeah. That makes sense. Have there been any other things in terms of open source that, like, you'd be saying to other people? You've been doing it for a while now, like, managing open source community.

Yeah. Well, well, we I guess the only interesting thing I could possibly say is, I guess I mean, having had the original projects that we worked on pre pivot, open source, and the ones that we work on now, post pivot, like, I guess, just trying to just get as many channels as possible to understand what people actually want. So we got a feedback box in the app. We got GitHub pull requests. We got GitHub issues.

We got GitHub discussions. We've got Discord. You know, I regularly email users to just solicit feedback. We post on Reddit and religiously read the comments. And, like, it is if people see that you're actually responding to the stuff and you just no.

Half the time somebody sends in a problem and I just reply with the pull request, and then they can watch it get merged and get released. Mhmm. Like, you actually need to kind of go out there. It's not it's not enough to just be building SaaS and then accidentally make your GitHub repo open instead of closed and then call that open source. You you get stars, but you won't get a community.

So you need to be I guess it's the the the breadth of channels and the responsiveness that I think separates this project from what we worked on previously that kind of may have looked like open source, but didn't feel like a real community.

So you weren't proactive in the previous

No. It was it was, like, source available. You know? We were we I don't think we had that many people opening PRs. We weren't really encouraging people to open PRs.

I don't think we had a Discord for most of the life of the the first project. We definitely didn't have Discord discussions on GitHub, and we weren't really going out there and soliciting feedback in in that way.

Why why do you think you didn't do that? It just didn't occur to you, or was it like a strategic like

Young founders. You know? Yeah. I I I feel genuinely, I would I would explain the last five years of my life is kinda like start up university. It's getting beaten up over and over again.

Yeah.

But you know what? We're tougher now. Yeah. Yeah. Yeah.

Payoff is here. I know. It's here.

Yeah. Well, kind of on beating up, on taking beatings. Well, the last episode we did was on, together, was on, you know, modest modernizing legacy code. And it made a lot of sense to me. Like, you know, it was like, this sound this sounds great.

What what do you think were the what were the challenges there? Like, why why are you not still doing that, I guess, is the?

Well, I would say a couple of things. The I mean, one of one of the the the easy things to admit is that I don't really like selling to banks as a person, to be honest. I think I tried it, and it and there are so you know, I've been in the room with people that are really good at this, and I I actually think it's, like, something that I just don't really particularly enjoy. Yeah. There's there's you know, with Vibecam, we're going much further, I guess, leaning into selling to the types of people that we are who are driven by technology and excitement about that rather than risk.

And, you know, it's a it's a much more involved process to kinda persuade a big bank that you wanna convert all their cobalt to Java as much more than just the technology. And it's a skill. I I I respect people that can do that, but it's a skill that we would have had to spend much more time learning how to do. And at the end of the day, small startup, limited runway, looking at one and a half year to two years sales cycles, we could have stayed in the game and waited to see what was gonna you know, whether some of those big deals were gonna convert, but we might have gotten to the end of that road and found out none of them converted, which is often the case with enterprise sales. You know, we had one customer who, they were we were negotiating, and this was right during the first wave of, like, Trump tariffs against Canada.

And they're a Canadian company, and we just came in one morning, and the guy we were dealing with had been fired. This company that had paid us 6 figures already to, like, do some work with them, and then they just shut the whole thing down overnight and reawked. So, you know, it's those cases where you're not losing deals to competitors or you're not losing deals because you built a bad product. You're just losing customers due to internal chaos at this company that's that's really hard to deal with. And I I guess at every, you know, every every startup eventually ends up having to cater to that audience, but that as a starting place was just quite difficult for us, I think.

And so this feels like it's back to our roots. You know, we know how to talk to developers. It's a tool that we would use. We like, we didn't have a mainframe. This wasn't a problem that we faced.

And so maybe we struggled to empathize with that user. Like, if I'd spent ten years in a bank working on modernization, maybe I would have kind of approached that differently. But the perspective that I have as a developer now, you know, allows me to build five kanban in a in a kind of authentic way. So, yeah, some thoughts.

Yeah. I think there's because we've been thinking a lot about product market fit and Jason Cohen, the guy that founded WP Engine and and another one, he he wrote this, like, really good article and I think it's like a lot of it was about the the obvious stuff you think in product market fairs like the poll. But I felt but he also talked about this, like, personal bits or personal

Founder market fit. Founder market fit.

Yeah. Yeah. Is that something that sat I mean, it sounds like that was a big part of this one.

Yeah. Definitely. From from a go to market perspective, low found a market fit on COBOL modernization. I would say the technology is there. If you are if you are a bank listening to this and you wanna convert your COBOL to Java, like, you definitely can do that with off the shelf tools.

It's a totally solvable problem. I think the the problem that you have to solve is what do I do with all these developers that I'm gonna have to fire? And how do I do this in a way that is, you know, risk free or nothing's risk free, but in a way that is kind of satisfactory from a risk perspective. And then just, you know, you have to fight the sales machine of these kind of legacy incumbent vendors who have been selling to your company for forty years. So, yeah, it's definitely kind of more founder market fit, I think, on that one.

The business model would make sense, I think, if we had another vested interest in selling to those same companies, namely banks. So for example, if you're AWS, you can do the modernization of somebody's mainframe, and then that company is on your system, and you can charge them for the next decades on EC two instances. Right? So it's great, great thing to do because you're you're getting workload out of a data center somewhere onto AWS. Or if you're Accenture, you know, you can sell a bunch of transformation.

Oh, well, we're gonna do, like, workforce whatever. And, you know, we're gonna you know, you look at it from the wider perspective of business transformation rather than just what's this kind of technology that we're trying to introduce here. So I think, yeah, it's found a market fit, does it work for startups, but the technology's there ultimately. Yeah. Yeah.

Okay. That that makes sense. I mean, it's it's really cool now to see the whole journey that you've been on. And yeah. So what what are you most excited about with Vibe Kanban over the next Well few months, I'll ask you because

yeah. We might all be dead in a year. So, you know, I guess the thing I'm most excited about is is is when this stuff becomes multiplayer. I think very deeply about how there isn't really any good interface with especially with coding that kinda combines, you know, the raw ingredients of, like, me and my team and coding agent and a great interface. And we've had some of our customers sort of early customers who've kind of faulked Vibes Kanban and built interesting, you know, mutations of what if three people could, like, work on the same conversation, but it's specifically about planning.

Or what if, you know, we kind of compartmentalize the work in a way where when ClawCode finishes running, it doesn't matter which developer picks that up. Like, you just go to the pool of developers, and and whichever task lands gets matched with the next available developer. Kinda, guess, the delivery or DoorDash for our American listeners, ification of, of of coding, things like that. Yeah. Or the the Mechanical Turk of Yeah.

All our sales, sales leads and stuff.

Exactly. You just get you know, the the auto dialer just starts dialing, which is maybe slightly depressing, but it's more the process of experimentation that I think I'm excited about with with that stuff.

Okay. That's cool. That's really cool. Okay. And then if you were giving one piece of advice to someone listening about the journey you've been on, someone that's a bit someone that's in their first iteration starting out like, yeah, that hasn't been to start up university like you have now.

No. Well, hopefully, you won't have to go to five years of start up university.

I'm

still in it. I'm an I'm a older student now. Well, just don't die. I mean, I know it sounds stupid, but, really, like, we wouldn't be alive today if we hadn't have raised all that money in 2021 for the original idea that allowed us to stay alive. I mean, I'm I'm very grateful to our investors who've been incredibly supportive and have backed us each pivot along the way.

And, you know, a lot of the original team are still here as well. And I think the reason why we've been able to kinda maintain that support from the staff and the investors probably because we've been very open and quick to jump from one idea to the next. You know? You gotta put everything within an experimental framework. Mhmm.

Okay. What are the success I know it sounds simple, but, like, you need to kinda constantly remind yourself to just put everything within this framework of, like, what would this look like if it was working? And then be ruthless if it's not working about cutting that and moving on to the next thing. And when you get to the point of when where you're you're you're kind of thinking, okay. Well, maybe this isn't working.

Just being very quick about, you know, moving on to whatever that is. Because that's when companies die. When you're when you have a team that is working for a long time on a product that everybody knows is dead. And it's just so demoralizing. And so, yeah, just getting that energy.

Like, you know, I say to everyone, you know, it's like, you can go and get a job anywhere. You know, you're all very employable, but it is my job to kind of make sure that it's still interesting on Monday morning, and that's that's completely true. So yeah.

Yeah. That's amazing. And, where can people learn more about Vibe Kanban?

Run MPX Vibe dash Kanban and go through the beautifully designed onboarding process that has just been refactored.

Okay. Nice. Very nice. Very nice. Okay.

Awesome. Well, thank you everyone for listening. Thanks, Lou, for coming back.

Thanks for having me.