Kleiner Perkins' Leigh Marie Braswell on the AI coding wars, Windsurf, and what comes next

intelligence, AI agents, all the news of the day. And we'll bring her in. And we got some we got some good Chirons for her coming up. We'll see if we can get them up on the stream. What do uh what are we what are we talking for her? The term she tyrant herself. Lee Marie, are you there? Can you hear me? How you doing?

Oh, what's up, Kugan? How's it going? It's good. Uh what's uh what's new with you? I'd love to get your initial take on uh AI week. We got Microsoft Build, Google IO, Open AI IO, we got Claude 4 from Anthropic. It's been a massive week.

what has stuck out to you as like the most interesting, the most underrated, the most worth digging into uh of the news of the week? Well, I guess just to start off and it's so good to see you again. It's been it's been a while. So, yeah, thanks for having me on the show.

Um yeah, it's there's never a dull moment really, you know. Um yeah, you know, you know, I feel like every day there's there's something really, you know, significant that you need to be paying attention to if you're going to try to make money investing in AI companies. Yeah.

Um, and so yeah, I mean I think the big thing that's actually there's so much this week. I mean, right, you know, starting with some of the stuff you said, the best time to be a technology podcaster, investors close second, but podcast is really where you want to be.

That's where the action is because every day there's new news. Anyway, yeah, what is sticking out to you? Hey, Jordy. Um, hey. So, so there's um the I mean the IO acquisition in particular, that news. Um, what an incredible video. What an incredible team.

Um, I mean, just really showcasing OpenAI's ambition to really, you know, get into get into a lot of stuff. And I mean, obviously there's a lot of speculation about what the device will look like. I was actually looking My favorite, sorry, sorry to interrupt.

My favorite part was uh John and I are, I guess, exhaul monitors. We're in the community notes program and so somebody was trying to community note the OpenAI launch video and the suggested community note was this video uses AI. The these two scenes aren't real.

If you look at this section, it's like you're trying to gotcha an AI company for using AI in a marketing video. Like what is this? That's very silly. What do you expect, you know? Yeah. Yeah. Yeah.

Um, so I mean, yeah, I mean, super excited about what potentially that could that could imply for just, you know, OpenAI's ambitions, especially as it relates to consumer consumer hardware.

I was looking it up on on 03, but 03 kept then kind of giving me some different answers, but it might be, and to be fair, I don't, you know, kind of know the current state of things inside of IO, um, Johnny Ives company, but it might be one of the largest, if not the largest sort of pre-product acquisition of all time.

So, you know, I mean, $6. 5 million is a lot of money. Yeah, we've been noodling on a couple points. It's a couple points for a key exec. Exactly. It's 2% of the company, which when you think about bringing in an executive at that level, that like comp for that industry, it's not that crazy to get to that number.

And then today in the Wall Street Journal, Sam Alwin was quoted saying that if they can nail it and get a major consumer device, that's a trillion dollar opportunity.

And that sounds crazy, but you look at Apple's market cap, you look at, you know, Android, Samsung, like, yeah, there probably is a trillion dollars in market cap. There was also there was also the the tinfoil hats saying like, oh, he's trying to dilute the profit.

It's like, you know, that they're issuing billions of dollars of stock to employees like all the time, right? like this is not some like you know crazy conspiracy. I I I do want to get your take on the actual device.

We we've seen some pre- render some AI images around like a pendant with a camera on it, the her model, maybe it's an earpiece. I imagine you get pitched a lot of AI devices. We've seen the rabbit, the humane, the friend.

Uh, are there any other uh any other ideas that you've seen that could potentially be um just AI devices, even not from Open AI, but just in general? Because Apple's got it pretty well covered. They got your ears, they got your eyes with the Vision Pro, they got your phone, the watch, the iPad, the laptop, the desktop.

Like Apple's there's not a lot of white space in the Apple catalog, but um it'll be interesting to see if somebody can come up with something that's new. Have you seen anything that's interesting? Oh man, I really wish I could I could tell you about it, but yeah, we actually have a company in Stealth.

I can't say much, but there's certainly different, you know, the show when they're ready. Yeah, I I'll be super excited too, but people are being really creative.

Um, and then also, I mean, just generally, and I'm sure you're seeing this too, just given like the pace of h how much is happening and then also the competitive nature of these markets. Like it's clear that there's this big prize maybe bigger than ever before. Yep.

And so if you got a really good idea, like and if you don't have trouble raising funding, getting customers, or hiring people, like you know, why why talk about it too soon? So, yeah, I'll be excited to to share more on that.

Um, yeah, there's also I feel like people have just been generally willing to fund hardware, but also kind of categorically bearish, right? People are just like, oh, like hardware is really hard. Like I'm going to make this bet, but like it's really hard. Hardware is hard. Used to be about supply chain.

It used to be about all the things that can go wrong. But then also look at working capital. We're at a point where like Aura and Whoop are like big businesses that I you know I don't know like ultra specifics on either but I I bet that Aura will be a really big company in 10 years.

From my perspective it's almost harder. It's not it is hard to make the device and manufacture it and deal with the fact that 90% of your revenue comes in Christmas.

But um like the real hard part is actually changing consumer behavior and getting someone to take off their Apple Watch or add a Whoop band to the other wrist, which is I think the predominant pattern for most Whoop users. But um what what what are you looking for in hardware startups that you're looking at?

What are the pitfalls that you want to see founding teams overcome before potentially backing them? Yeah. So I'll I'll say I don't have a massive background in hardware.

I mean I will say I was I was at scale you I worked relatively early there um for you know and was an engineer there for four years and then a PM um and we worked with a lot of hardware companies there so I mean the main customer of scale back in the very early days was self-driving car companies um and robotics companies um and so I have seen just through to working with them as customers you know some potential pitfalls and I mean you know the the quip is that hardware is hard um you know that you've got all of the challenges of building a company and then you've also got to just manage supply chains and you have to be a lot more careful about, you know, the economics.

You know, software is usually relatively high margin or it's easier to make it high margin. Hardware obviously much harder when you've got to figure out where all these materials are coming from and ramping up.

So even if you have demand like ramping up in a sort of scalable way and then also implementation um can generally just just take a really long time and it distributes at the speed of the internet like you can go viral and get a 100red million people on a website in the day.

you you just can't ship a hundred million things anywhere in any reasonable amount of time. It just takes time. Um like maybe one day. Um but yeah, it's like interesting to remember like Whimo was founded in in 2009. Yeah. Right. So like just as that like it took overnight 20 years to now become this overnight success.

But yeah, you mentioned robotics. I'm interested in the pools of training data that are out there. Uh we've been joking about scale AI. Are they going to put everyone in motion capture suits eventually to get to the humanoid data sets? There's also a lot of work being done in in simulated environments.

Uh is there a data wall that we're going to run into with humanoid robotics? How are you thinking about the data challenges in robotics generally, whether that's in the humanoid side or or elsewhere? Um because it feels like the humanoid narrative is taking hold. People are funding companies. It's getting big.

But if it's a 16-year journey like it was with Whimo, we could be, you know, in for in for a long one here. Definitely. Definitely. It's been interesting. I mean like I do think there are a lot of really good arguments for why we'll see robotics pro uh like progress accelerate.

Um and that's really you know obviously the data is more challenging to get you know it's not like you can just scrape the internet like we did with a lot of these these LLMs and the data is kind of like there and ready for the taking for at least like some of the early transformer based models.

Um though with an asterct like as as we're now moving into agents I actually think you have to be more intentional about creating different types of data.

Um but then now with robotics too like you know you've got to figure out ways to get real world data um whether that's you know internally or using somebody like a scale um or you know there's a lot of really exciting work going on in world modeling and simulation um and so you know I think hopefully a combination of being really intentional about the data both on the sort of like human collection side and then also some of this work that we're seeing in simulation and world models um will hopefully mean that you know we don't all hit this data wall and we'll be able to to sort of scale these systems more effectively.

However, I mean with robotics, it's always a little little bit challenging to tell what that exact timeline is. Yeah. Interesting. Um, we were talking earlier about uh the agentic, the coding market, the AI coding market.

I was kind of breaking it down to you in it three markets, is it two markets, is it synchronous work, asynchronous work, is there a consumer, proumer, bottomup enterprise, top down enterprise, sales motion? um how are you thinking about just AI coding generally as a market as the businesses uh evolve?

Uh just give me kind of where you're at and then we can kind of tug on that thread in a bunch of different ways. For sure. My favorite my favorite topic of conversation or definitely one of my favorite is is uh AI codegen um near near and dear to to my heart.

But um yeah, as a disclaimer, so I'm I'm on the board of this company Windsurf. Um so Agentic IDE AI powered developer tooling uh platform. Yep.

Yeah, it's been it has been just utterly sort of shocking to me like you know when you invest in a company you expect like okay you know thing hopefully things will go well like hopefully market will be there and I just don't think anybody um I don't think anybody really saw this coming just in terms of like how widespread tools like Windsorf would become.

Yeah. Um and so right now, you know, you're certainly seeing like there's this sort of segmentation specialization. You know, you have tools like Windsurf and Cursor um that are, you know, proumer.

So you've got people even with basically no coding or definitely no coding experience that are able to get up and running very quickly with these tools and really sort of like talk to it and it acts like this sort of programmer that will, you know, like respond to what you want to do.

Maybe it comes back, maybe it comes back with an error, but it can maybe autonomously fix that error once you kind of give it some more guidance.

And you know, it's just been incredible seeing like, you know, even even though I talk about Windsor all the time, my father-in-law, who I never mentioned it to, not a programmer, just like once he was on a on a we were just talk all talking, he's like, you got to go, you're a VC, you got to go check out this company, Windsurf, I'm like using it to like, you know, help me out at work and create apps.

And I'm just like, wow, this is incredible just how widespread he's like, you know, outside of Silicon Valley, all of that.

So, so for those types of tools, I mean, you don't have to speak about wind service specifically, but just how much of it is just a ton of people on small consumer plans versus bottom-up enterprise adoption that gets rolled into a much larger consumption-based plan or even like a seatsbased plan like how is how is the market for AI tooling in the enterprise evolving in terms of pricing?

Because we've seen Salesforce is talking about we want to price things not based on seats but based on resolutions of customer service tickets. You could imagine charging someone for lines of code that get approved in poll requests or seatbased or usage based or just pure inference based.

So how is that how is that side of the market just evolving broadly?

Yeah, I mean so what we're seeing is you know there's certainly the proumer segment for these tools but it very quickly um goes up you know bottoms up into the enterprise and in particular with Windsurf I mean from day one we've been quite focused on the enterprise segment um and so it's you know we're both we both have a large enterprise sales team and then these tailwinds from now all these proumers and then you know sort of like professional developers that that pick it up on their own and bring it in into their company.

Um, I think pricing it's it's always interesting. It's something that we constantly think about. It's a very competitive market. So, I mean, you know, you also have to kind of pay attention to to what other people are doing. I mean, per seat easy to understand.

Um, but then, you know, obviously a lot of these models co, you know, cost money to run and so you've got to figure out some sort of either credits or usage system.

Um, I don't think, you know, in coding where you get to the point where it's, you know, let's price based on outcome like, oh, well, if we get this done, then then you pay. um like maybe you know I've seen sometimes in like customer support and things like that.

Um but yeah, certainly having like a usage element I think is pretty pretty pretty cool. Yeah.

What's your reaction when an entrepreneur pitches you and they're like, "Well, we're we're building agents to go after this labor this like end labor market and and the the TAM is, you know, a trillion dollars because they're just adding up like global payroll for like that type of work.

" Like what's your reaction to that? Is that the wrong way to think about market sizing if ultimately these are software products that can be delivered very inexpensively which will ultimately create competition and drive maybe pricing down for that end service. Yeah, I mean def definitely definitely some good points.

I mean generally generally I'm pretty optimistic.

Um I think now just like kind of living through the codegen market where you know just to use this as an anecdote you know GitHub copilot came out my first kind of pessimistic reaction was just like oh well this is like so useful already and GitHub's got all these advantages okay this like market's done and they will be like okay every developer like uses it and then it's kind of over and and it just the opposite of this happened like it turned out that there's a lot of white space in the market for teams to innovate and to create really great product experiences um whether it's a startup, whether it's a foundation model company.

Uh and then also just like I think the market blew my mind. What was the secret? What was the secret to either the the challengers to GitHub Copilot? I I haven't really used them.

Um is it is it more driven by uh flexibility around which model you're plugging into and and and GitHub Copilot kind of locking you into their model and not moving quickly enough or is it something more on the user experience side? So I forgot the exact date.

GitHub copilot started I think open AI only given the sort of relationship but it relatively quickly moved to an option. So it's so you can put in cloud 3. 7 sonnet which everyone loves. Yes.

And I think actually recently um you know I think I was reading this this morning but you know you know GitHub announced like okay for these particular features we're going to go with anthropic only like particular backend features. So it's clear like you know there GitHub is not locked in to one model provider.

I think you know what cursor, wind surf and others have done is just create you know different product experiences. Um and obviously get still doing very well but I think what this all has told me and kind of back to your original question was these these markets are actually much bigger than we give them credit for.

Um, and so when I'm kind of sort of being pitched maybe a ridiculous TAM, sure, it's probably good to like verify that and really drill in on that. But I mean, we are in the world where now software can do work.

And so you do have to kind of, you know, you don't just get the sort of TAM being okay, it's the, you know, incumbent software spend. I think you have to also realize once the product gets that much better, people will want to buy more software and then also think about it as a work replacement.

So you kind of get all these kind of bonus markets that you know TV you know end up being just like this hu huge thing. How do you how do you how do you process the tension between model providers kind of figuring out if they're wanting to end the own consumer or be a platform.

I mean the the most obvious example was windfropics new models.

uh which seems uh you know I'm sure you can't comment on that specifically or maybe maybe you want to come in with put him yeah put him in the hot seat but um but yeah I I it just feels like it feels like it feels like a very real tension that's going to have to get resolved quickly because these are all big companies now or at least the main players are big companies.

There's billions of dollars of revenue on the line and uh everybody's in a you know constant race to be at the edge. You're right. I pro probably shouldn't say too much. Um I mean what I what what I what I will say yeah it's something that I think is going to have to be you know addressed in the next year or two.

Like I mean I think it's just very clear that cogent in particular is this is this market that everybody kind of wants to get into. Yeah.

You know, I do ultimately think that, you know, I'd be shocked if in a year or two we don't see codegen products that are both, you know, synchronous, so like assistant, you know, I can talk to it and from the same interface, I can delegate task.

I just think that that is something that a lot of people are building towards. It's like there isn't a great experience today that kind of has both. Yeah. Um at least, you know, sort of a generalized both.

But I mean ultimately like all these companies are I I think going to in the in the sort of like even medium term come up against each other and ultimately what you want is sort of this distribution and I think you know all these companies too are realizing you also want the model um and you want to be able to you know give yourself really favorable favorable economics.

So yeah, it's it's going to be a super exciting even, you know, probably probably 6 to 12 months, I imagine. So legal the the sort of legal AI space has been heating up. Kind of a weird, you know, interesting comp. In many ways, it felt like cursor had like a ton of mind share and then wind surf just kind of exploded.

People realized how quickly it was growing. At least that was my perception. And then uh Harvey similarly has been more in the sort of uh cursor bucket and that like tons of attention, bunch of rounds back to back and then Lora uh came out of feels like it came out of nowhere and they raised a big series B this week.

Is there are you do you have a a a kind of prediction or a take on what the next category to see these sort of battles play out? I'm sure you guys already have some bets in whatever category that is, but she's already on the board of four companies that are going to dominate it. She's like the next four categories.

Actually, I didn't jump to the board meeting in exactly what you just said. Yeah. Yeah. Um, you know, I'm I'm I'm very lucky in that, you know, I think I mean, even before I joined, so I've only only been at Kleiner Perkins for two years, and even before that, a lot of my partners made really great app investments.

So for example, my partner Moon incubated Glean um which is now you know kind of the enterprise knowledge discovery um platform in this kind of new new AI era. Um so certainly like we've got a lot of apps in the portfolio.

I mean Harvey we you know did their series B um Ambience which is a medical scribe also in the news recently. Um we did their seed in series B. Um so and then you know we have some some investments in sort of sales AI as well.

Um, I work with this company Nooks, which has started with a power dialer that people really like and is now expanding into other other parts of assisting sales people with AI. Um, I'd say like those, you know, those categories plus maybe add customer support, which we don't yet have kind of a generalized play in.

Um, those are kind of like the the battleground spaces. These are where it's going down right now. Like it's clear the models work for the use cases. It's clear there's a giant prize at the end. And so you've got, you know, inevitably a lot of companies competing for it.

Um, in general, especially with software, you know, I think it's basically impossible to be the only company in your town.

Can you can you fund a company that will just pick up the phone when I get like a robo AI call that's like trying to do some social engineering hack or whatever and just pick up the phone and waste their time really inexpensively. That's I feel like that's what there's definitely some cyber security companies out there.

But yeah, I think I've I think I've actually been pitch pitch I would have to find the name, but yeah. Um, what what do you think about competition from the hyperscalers?

Obviously, uh, Microsoft Build and Google IO happened this week, and there's one narrative that's like, "Oh, so many startups are now just bullet points in a Google IO keynote.

" Uh, but Ben Thompson wrote shortly after Google IO that, uh, Google only cares about search and they only win when it's directly impactful to search. And so, you might as a startup not need to worry if you're doing something that's more in the labs or experimentation phase for Google.

in Google is more just almost doing it as a demo of their of their platform and then they really would be fine with another company just buying a bunch of TPUs on GCP and and and building a real company on top of uh their innovation.

Um but how are you thinking about startups competing in these new AI markets with the legacy incumbents? Yeah, it's it's it's a it's a really good good question and it's something that I think about a lot. Um just given like obviously you know somebody like Google has giant advantages distribution money.

I mean Google literally created the you know first Transformers paper.

Um so I mean certainly certainly it's something that whenever you're investing in in an app company in particular you know you you need to I think at least think very deeply about what's what's the road map of of all these companies OpenAI Anthropic etc.

Um, I mean certainly certainly seems like sort of areas in search in sort of like social and and proumer that you know seem like kind of the most under the microscope with these big companies. But you know it it wouldn't shock me if they tried tried to compete in some of the other app categories we talked about too.

So yeah, in general though startups I I like to bet on startups.

focus is important and then you know I think if you move fast if you're very strategic there are a lot of good sort of like wedges you can sort of start with and then and then from there what about M&A markets uh in particular I mean my my reaction to the wind surf acquisition was and it's going to be kind of hard for you to answer this but um was was basically that like whoa like maybe uh that whole meme of like oh don't build a chatgpt wrapper or whatever uh maybe that was overblown maybe that there's going to a lot of app layer companies that get scooped up.

Obviously, OpenAI is moving very quickly. So, they're kind of early to this, but you could imagine every hyperscaler needing application layer AI companies to slot in, especially if they've had distribution and they've actually scaled revenues and they have a really great team.

Um, and it feels like unlike in the previous era like 2010ish, like the hyperscalers are more sclerotic than ever. It's been 15 extra years.

they're less founder mode and so uh I don't know what your read is on the M&A markets broadly but um how are you thinking about uh the hyperscalers approach to staying on the leading edge of these AI application layer companies? Yeah.

Um I mean I think it's important to remember and I have to remind myself that Chad GBT is less than three years old I think. So it's like this is happening so fast.

Um, and so when you think about like I'm a hyperscaler and I see a market that that like makes sense for me strategically, you know, is it really faster for me to go and try to build it internally or is it faster for me to find a really great startup perhaps with expertise that I don't have um and maybe you know some other advantage whether it's the product or the distribution or or something else um and then you know explore explore partnership conversations with them um and so I mean you know the one one you know acquisition ition I I can talk about uh is uh I a company that I worked with um at at Founders Fund and and still on the board of Neon is partnering with Dre.

Cool. Um and it makes sense because it's like, you know, you've got this team of database infrastructure experts and then you've got this company who clearly wants, you know, their customers to be able to realize the full potential of AI agents. And so you do need, you know, a database and memory um for those agents.

Uh, and so, you know, it it's certainly now, you know, now I'm trying to think in my mind like, oh, well, what what are the other categories that that that hyperscalers might uh might might be looking in and and what makes the most sense for each one?

I mean, the the the big one to me that I'm trying to figure out is just video generation, which we had I wouldn't call it a Giblly moment this week, but V3 was was close. Took a little bit. It just takes a lot more work and it's harder to get a consistent quality output.

But and there isn't a theme where everyone like the Giblly thing was like just take your profile picture, take your whatever's in your camera roll, two words, boom, one shots and you have something cool. With VO3, you have to be more creative with the prompt. Yeah.

The interesting thing there is I can it's so easy to imagine in the future where consumers are using video generation tools daily just for fun. Oh, totally. Yeah. There's there's some like consumers just like making memes on the internet. Then there's like kind of proumer SMB I'm going to make like an advertisement.

Yep. And then there's the like Hollywood studio that's like I'm going to make a movie or independent filmmaker. But where companies kind of fit into that.

I can see a world in which hyperscalers are kind of dominating like Meta for example figures out a way to get really good at at at video generation because it fits into the meta ecosystem or YouTube uh you know with with V3.

And I think that from my point of view, I can see some of these more independent foundation labs that are focused on video having to really move up market to actually dominate because that feels like an area that like Google's not going to necessarily build the thing that creates the next blockbuster video. Yeah.

But if you're competing for the proumer SMB, like feels like a really tough place to compete purely from a cost standpoint and with YouTube or Google's sort of YouTube data advantage. Yeah. Yeah. I don't know any any any reactions to V3 this week. Oh, I mean super cool. Uh and we were playing around with it.

My husband's a founder and he was like trying to make this new new video for something that they're developing internally and oh my god, just insane. Uh so it's yeah, really cool.

And I mean I I I kind of agree with some of the points that have been made, but then on the flip side, I will say like especially in video, there's so many really great models and they excel at a variety of different things.

And so if you're like kind of a third, you know, a startup that is not connected to I mean Google obviously is developing their own models.

So if you're not connected to any one of the particular models and you can offer like kind of a best-in-class like okay, you can pick from you know many of them um maybe maybe that's an opportunity or if you're sort of like very very specialized.

So, um, yeah, I mean, generally still still optimistic just kind of given at this point there's not, in my opinion, a ton of platform risk. You've got so many of the hyperscalers and labs like that are, you know, both closed and open source offering offering these types of models. Yeah. Very cool.

Well, thank you so much for stopping by. Uh, we'd love to have you back. This is fantastic. Yeah, this is great. Super fun. All right, bye. See you later. Uh, really quickly, let me tell you on an absolute tear. We barely got to Neon. Another $1 billion acquisition. Fantastic.

Uh, but really quickly, let me tell you about Figma. Think bigger, build faster. Figma helps design and development teams build great products together. Get started for free at figma. com. Redesign how you design, explore your ideas freely and and iterate quickly. Let's also tell you about Vanta.

Automate compliance, manage risk, prove trust continuously. Vantis trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation. Whether you're pursuing your first framework or managing a complex program, go to vanta. com to get started.

Thank you, Devanta. Thank you, Devanta. Anyway, we have our next guest coming in, Stephen from Lambda Labs. We've had Lambda School on. We're waiting for Stephen. Should we hit some timeline? Yeah, let's hit some timeline.

Um, what's uh did you see uh Yohi uh Pocket is shutting down and so this uh uh we need a sound creator basically soundboard that says moment of silence. Yeah. Vibe coded a quick free open source prototype with some AI features called Keepyard. It