The Exchanges

Q I was going through Some of your tweets as well. And you said async remote agents will happen. And I just wanted to double click on that first. Can you expand on how you think about this and which tasks will be automated first?

A So the idea of async remote agents, if you were to look at a product like Windsurf, Windsurf is actually a local agent that runs, runs locally, and it can do very long running tasks. Like you can migrate an entire code base from one version to the next, and it should be able to do that without you having to touch Your keyboard, right? Once you, once you instruct it to, which is very, very powerful. But this idea of being able to do it asynchronously is also very powerful. You can imagine like just firing something off from your phone and it just generating a, like a pull request for you. But I think one of the complexities of this is there's a building products in our category is I think there are three things that really are important, right? Uh, there's latency. So how long it takes to get the response back. There's quality, right? Which is what is the, what is the correctness, right? And then correctability. How quickly can you make a change?

AI assessment note: “You can imagine like just firing something off from your phone and it just generating”

Q What do you mean by that? It wasn't where you placed importance in terms of that initial UI, UX being seamless?

A Yeah, I think for us to go out and build Windsurf, we actually felt that we needed a massive, like, technical breakthrough of, like, actually a new product experience, and I think that was the only reason why we actually decided to go out and build Like actually have our own ID. I like, if we didn't do that, I think we're the type of company that would not just go out and build UI UX just for the, like, you know, for a much better experience by default, unless we had great technology. I like to think of ourselves as like, we're a technology company and I think we ship product to maximize the amount of technology that our users can consume productively.

AI assessment note: “we're the type of company that would not just go out and build UI UX”

Q To what extent is the quality of your customer experience defined by the software that you create, the UI, the UX versus the model progression that lies beneath it being open AI anthropic?

A Yeah, I think, I think you can't downplay the fact that, that the latter is very helpful, right? The frontier model is getting, getting more and more capable. Every new capability unlocks maybe a new type of product experience that wasn't capable in the past, but I think it's a little bit of both, right? Um, maybe, maybe to put it, put it kind of, uh, kind of only like we were the first agentic IDE experience, right? Uh, for when we launched Windsurf, uh, we were, we were the very first ones. Obviously now that idea is not very, it's not non-consensus. Like it was the obvious thing to do, but these models like Sonnet had existed for quite a while, but it actually took us some work to actually put together Uh, like using these models, but also using our own models internally to make this a usable experience with the UI us to be able to review all the software very quickly before people actually this, the product experience kind of took off. So I think it's a, it's maybe a little bit of both, but I don't want to underplay how much the models actually affect, you know, the value people get from these experiences.

AI assessment note: “I think it's a little bit of both, right?”

Q Given the commoditization of that model layer, as we said, and no one being able to run away, to what extent the model providers have to move into the app layer to have any form of differentiation?

A Maybe in the, in the API side, I think there are lots of ways you can build better APIs for a lot of different applications. And it's kind of like, maybe you can, you can, you can kind of focus on the applications you do support. And an example of this is like, let's say you were, you were kind of working with a company like us, Um, and we were using a model provider. You know, when we output code, maybe a lot of the code looks the same as existing code, and you can actually optimize from a latency perspective to help our workload versus another workload that exists for images out there. And you can be a much better kind of vendor to us than another company. So there is maybe specialization you can have on different axes to provide a better experience. But as you can see, you, you are seeing like these model companies actually going up to the app layer too in whatever apps they see are like valuable to them. Um, so I think you're seeing a little bit of both. People have too many opinions about what the right way to do things are, and I think you, you have to be rational on, on what the best way to win is, and I think that's the only thing that matters.

AI assessment note: “you are seeing like these model companies actually going up to the app layer too”

Q What do you believe was your non-conventional insight when you made the pivot to windsurf? Yeah.

A So, I mean, I could even start before this, like the first idea that we had, even like our company name has changed twice now at this point, it was called, uh, exa function. When we first built GPU virtualization technology was that GPU workloads were going to power The world, right? That was like sort of the, the mission there. A lot of us at the company had previously worked in autonomous vehicles and AR, VR, um, at the, at the company. We had a fairly small team that was actually accurate that GPUs were going to become very popular. Like we were, we basically said NVIDIA is going to sell a lot of GPUs. The thing we were wrong about was how diverse the GPU workloads were going to be, right? We thought a bunch of different model architectures would be run. There were going to be hundreds of different model architectures and we were wrong. All of them ended up being transformers. And in a world in which all the architectures look the same, um, there's very little reason for us to be a differentiated infrastructure provider in that category. So we needed to pivot at that point. But that's some, one of those things where you don't know going into the market that that's how the world is going to turn out.

AI assessment note: “I could even start before this, like the first idea that we had”

Q I was going through Some of your tweets as well. And you said async remote agents will happen. And I just wanted to double click on that first. Can you expand on how you think about this and which tasks will be automated first?

A So the idea of async remote agents, if you were to look at a product like Windsurf, Windsurf is actually a local agent that runs, runs locally, and it can do very long running tasks. Like you can migrate an entire code base from one version to the next, and it should be able to do that without you having to touch Your keyboard, right? Once you, once you instruct it to, which is very, very powerful. But this idea of being able to do it asynchronously is also very powerful. You can imagine like just firing something off from your phone and it just generating a, like a pull request for you. But I think one of the complexities of this is there's a building products in our category is I think there are three things that really are important, right? Uh, there's latency. So how long it takes to get the response back. There's quality, right? Which is what is the, what is the correctness, right? And then correctability. How quickly can you make a change?

AI assessment note: “You can imagine like just firing something off from your phone and it just generating”

Q cursor user and I switched to windsurf and it was like better or just as good or like he switched to windsurf. For this one project. Um, I've no idea if he stayed. I'm not making a comment. Tom, don't kill me. Um, but, um, my question to you was like, how do you think about switching costs and retention? And is it that easy to switch between the two?

A We need to innovate on features and product fast enough that we are able to build differentiated experiences for, for folks. That actually takes a lot of work, right? Like, as I was saying, like most of our products that we end up building don't end up working. And even though we're shipping releases every couple of weeks, some of the more complex pieces of technology that we've shipped, they take months to actually ship. So I'll give you an example. A couple of weeks ago, we shipped a model that was comparable to some of the frontier models at the agentic workload. And our models we want is much faster, much cheaper for us to run and capable of operating over a large code base as well. And we were able to do that by learning from our users on how they use the product internally.

AI assessment note: “We need to innovate on features and product fast enough that we are able”

Q Is there anything that you can structurally do to increase the response time of an organization to move faster post pivot?

A Yeah, I think, I think the answer is for a startup, one of my true beliefs is that companies don't succeed because they do many things well. They succeed because they do maybe one thing really well. Like, it's kind of like what I said, I think having one good idea already is like a miracle for most startups, right? And when you think about it from that perspective, when you think about a company, um, think about two companies, you know, any product has some sort of exponential curve at which it's able to grow, right? There's some R value attached to a product. Uh, the right thing for a startup to do is always focus on the product with a higher R value. It never makes sense to be diverting your resources to work on two different products that have different exponential growth curves, right? If they're completely disjoint. So when you do a pivot, I think the right answer is you should just give up what you were doing in the past, because the thing you were doing in the past probably will have no sort of relevance to your business going forward, right? The revenue multiple on that business is not going to look at all like the revenue multiple of the other business. And I think if you can go cold turkey and actually go and Make your organization completely work on the new thing. You're more likely to succeed, but that's scary, right? You need to tell people, tell your investors, te…

AI assessment note: “go cold turkey and actually go and Make your organization completely work on the new thing”

Q Given the commoditization of that model layer, as we said, and no one being able to run away, to what extent the model providers have to move into the app layer to have any form of differentiation?

A Maybe in the, in the API side, I think there are lots of ways you can build better APIs for a lot of different applications. And it's kind of like, maybe you can, you can, you can kind of focus on the applications you do support. And an example of this is like, let's say you were, you were kind of working with a company like us, Um, and we were using a model provider. You know, when we output code, maybe a lot of the code looks the same as existing code, and you can actually optimize from a latency perspective to help our workload versus another workload that exists for images out there. And you can be a much better kind of vendor to us than another company. So there is maybe specialization you can have on different axes to provide a better experience. But as you can see, you, you are seeing like these model companies actually going up to the app layer too in whatever apps they see are like valuable to them. Um, so I think you're seeing a little bit of both. People have too many opinions about what the right way to do things are, and I think you, you have to be rational on, on what the best way to win is, and I think that's the only thing that matters.

AI assessment note: “so I think you're seeing a little bit of both”

Q What, what do people who believe this to be true get wrong? What are they fundamentally mistaking?

A I, I actually, I never understood this idea of like a single person is going to be able to build a billion dollar company. I've never kind of understood it. I mean, because what is a company? A company is the sum of the discounted cash flows over time. And, uh, and like down, down the line, like you're going to need people probably, or if the AI is so good that you don't need people, like at some point, I would just wonder like, if like, what are you really contributing? And is the AI building the whole thing? In which case, like, I don't know, like actually that would reduce the That would increase competitive pressure because everyone else will have access to the AI too.

AI assessment note: “that would increase competitive pressure because everyone else will have access to the AI too.”

Q Dude, that is a great shout. Um, I haven't had that before. If you had to make windsurf open source tomorrow, how would you feel and what would you do instead to win?

A You know, I, I feel like that's a, that's a question I've been, I've been getting, uh, recently. I don't think it would affect, affect much if we had to make it open source, because most of the work that it's kind of like the thing I said before, right? You know, Google is, yeah, they have a front end, but the front end is just a single box. And if they open source the front end of Google, I don't think anything would really change about the world, right? That, that box of the UX for that was not really the big deal. Uh, I am aware of the fact that Copilot has done that, but most of our logic is actually in the backend. Ultimately. So I don't think that it would change a ton. I also don't think it would change our adoption, um, of the, of the product. Uh, like most of the differentiation has nothing to do with the fact that the front end is actually open source or not.

AI assessment note: “I don't think it would affect, affect much if we had to make it open source”

Q What do you believe was your non-conventional insight when you made the pivot to windsurf? Yeah.

A So, I mean, I could even start before this, like the first idea that we had, even like our company name has changed twice now at this point, it was called, uh, exa function. When we first built GPU virtualization technology was that GPU workloads were going to power The world, right? That was like sort of the, the mission there. A lot of us at the company had previously worked in autonomous vehicles and AR, VR, um, at the, at the company. We had a fairly small team that was actually accurate that GPUs were going to become very popular. Like we were, we basically said NVIDIA is going to sell a lot of GPUs. The thing we were wrong about was how diverse the GPU workloads were going to be, right? We thought a bunch of different model architectures would be run. There were going to be hundreds of different model architectures and we were wrong. All of them ended up being transformers. And in a world in which all the architectures look the same, um, there's very little reason for us to be a differentiated infrastructure provider in that category. So we needed to pivot at that point. But that's some, one of those things where you don't know going into the market that that's how the world is going to turn out.

AI assessment note: “So we needed to pivot at that point.”

Q Do you think in your category brand power is important? People associate with being a windsurf user or a cursor user. Do you think brand has that same power that it does in other categories?

A You know, I, I would, I would like to, obviously, I think having a large brand is just helpful. It enables you as you ship more and more new products for more and more people to be able to use it and not have to build that, build the user base up from, from scratch. But if I just was to remind everyone, you know, kind of like a year or two ago, probably in Silicon Valley, GitHub Copilot was the product that everyone, everyone was talking about, right? Uh, at that time. And if I was to talk about it a year ago, right, actually a bit over a year ago was the Devon launch. And I think for three to six months, I was not hearing anything but Devon, uh, from, from everyone.

AI assessment note: “I think having a large brand is just helpful.”

Q cursor user and I switched to windsurf and it was like better or just as good or like he switched to windsurf. For this one project. Um, I've no idea if he stayed. I'm not making a comment. Tom, don't kill me. Um, but, um, my question to you was like, how do you think about switching costs and retention? And is it that easy to switch between the two?

A We need to innovate on features and product fast enough that we are able to build differentiated experiences for, for folks. That actually takes a lot of work, right? Like, as I was saying, like most of our products that we end up building don't end up working. And even though we're shipping releases every couple of weeks, some of the more complex pieces of technology that we've shipped, they take months to actually ship. So I'll give you an example. A couple of weeks ago, we shipped a model that was comparable to some of the frontier models at the agentic workload. And our models we want is much faster, much cheaper for us to run and capable of operating over a large code base as well. And we were able to do that by learning from our users on how they use the product internally.

AI assessment note: “We need to innovate on features and product fast enough that we are able”