AI to AE's: Grit, Glean, and Kleiner Perkins' next Enterprise AI hit — Joubin Mirzadegan, Roadrunner

During my sales career, probably the number one thing that used to break my back was that the underlying software, like Salesforce CPQ and others, just to create a quote and get it approved, is horrific. If you think you've seen bad software, you haven't until you've seen a 30-second loading screen to get from one page to another when you're trying to close a deal with 2 days left in a quarter. This is standard across the industry.

At every job I was ever at, I used to get yelled at because I would ask people to turn something around within a day or 2 because I needed to get a quote out the door. I thought, “Oh my God, I actually think you can abstract away a bunch of the complexity with these LLMs.” It's unstructured and structured text that you can reason with and do stuff with, right? That's why coding is such a great use case. That's why Harvey is such a great use case, because you have all this case law, and then you can point the LLM at it and reason with it. Then you build a bunch of enterprise features, functionality, and workflows on top of that. This is a very similar problem in nature. So that was the light-bulb moment: “Okay, I think we can actually build something better.”

I feel good. I feel comfortable in this seat. How are my levels?

What's that?

How are my levels?

Yeah, you're good.

All right. I have to ask you: how was the Zuck interview?

Yeah, it's very interesting. Are we recording?

Yeah, I just started recording.

I think they obviously had an agenda coming in, which was basically to raise the profile of CZI and Priscilla Chan, with Zuck being a supporting character, right? I think they accomplished their mission. My quick hot take, in a single sentence, is that if Priscilla Chan gets half of what she wants to do done, she will have more impact on humanity than Mark Zuckerberg. Facebook will just be a funding mechanism for the greatest bioresearch work done in human history.

Wow. Were you nervous?

You must be nervous. I wasn't nervous because of the sheer amount of prep that the CZI people put into us.

To understand what it's like with the executive staff of a 100-billionaire—I've never dealt with someone like that.

Yeah.

They are so good. They prepped us so well, so I felt like I knew exactly what to expect. You mean Zuck's team isn't randomly letting people walk in off the street and ask whatever they want without any pre-existing knowledge? I had to interview 3 times to even get in that room. It was very fascinating. We were very honored to be picked by them because we're not a bio-focused podcast.

No.

But the whole point was to reach out to engineers, and there we're on a stronger footing.

Yeah, that's awesome, man. I mean, it feels like a breakthrough, doesn't it? You've had some studs on.

Yeah. We just had Fei-Fei Li today. We just released it.

From World Labs? Was that the one you came from?

World Labs. No, we just released it, so that was a couple of weeks ago.

Who else have you done?

There's a bunch. Greg Brockman—

You've done some amazing people. Is it surreal?

Yeah, it's surreal for me. I started off as just an independent creator—let's see where this goes—and now we're legitimately in that tier of podcast that gets invited to everything. That is very surreal.

Yeah.

It's pretty exciting. Who would have thought?

I mean, for you, I've been following your progress for a while. I don't think I had caught on when you were still a CRO podcast, but anyone following top founders who really wants real stories eventually finds their way to you, and you get really good stuff. So congrats.

I appreciate that. I think the transition from the CRO to the founder happened around episode 70 or 80, because it was only CRO for a while. The thing that I found refreshing—and I'm curious if you've seen this—is that founders and CEOs have the authority to speak in a different way than somebody on the executive team.

Yeah. They can just talk. And so much of what I want to do is have an earnest and honest conversation. It's harder to do that when you're thinking, “What is my boss going to think?” Whereas, if you're the founder, you can just speak.

Yeah. I think it's also nice for distribution because, obviously, that's the more famous or public-facing person, so people do want to tune in. You probably caught my attention for one of those episodes. I don't even remember which, but you've had so many. Even among the non-CEOs, I would highlight your Emilie Choi episode from Coinbase.

Yeah.

That was the first one.

Yeah, that one was so raw. You went there with all the politics questions.

Yeah, I did. I appreciate it, man.

It's a labor of love.

Mm-hmm.

Doing anything every week for 6 years—

Yeah.

I mean, I guess it started every other week, and then it became every week after the transition from chief revenue officers to CEOs.

Yeah, exactly.

Doing anything for that long every week, you better enjoy it.

Yeah. I've always told myself that the minute I stop looking forward to sitting down with someone and talking to them is probably the minute the show should be over. Well, it hasn't happened yet. It hasn't happened yet.

I was talking with Allie, one of your partners, and they said you even had to justify your purchase of a RØDECaster just to support your work. Doesn't Kleiner see the value in this?

In the early days, when I joined Kleiner Perkins, I was quite young.

Yeah.

I was definitely figuring out what was going on in venture.

Can you say a little bit about what you did before?

Yeah. I grew up in startups and then in sales, and I had a great run. Those startups ultimately ended up getting acquired, the last one by Palo Alto Networks, which is a big cybersecurity company. They asked me to move to the central United States and build out their public cloud business there. I took that business from 0 to quite a bit in a short period of time.

Kleiner heard about the work that I was doing and got in touch to see if there was an opportunity for me to work with founders once they've built the product. What do you do next?

Yeah.

I remember thinking at the time, “No way. Venture sounds amazing, but isn't that the job I'm supposed to do at the end of my career? This sounds incredible, but maybe later on.” Anyway, we got to talking, and it became very obvious that there was a unique opportunity here.

Fast-forward: one of the things that ended up happening was that I was working really closely with Arvind Jain at Glean.

Mm-hmm.

It's actually the last incubation that we've done here. Arvind is maybe the most genius product and technical mind that I've ever worked with. Go-to-market is not native to him, I would say. He and I were doing a lot of work together to figure out, all right, we've built this incredible Glean product. At the time, it was called Sio or Scio—I still don't know how to pronounce it.

We were running all these routes together: “I think you should do this. I think you should do this.” Eventually, it became very clear to me that we actually needed somebody to run the routes, because I couldn't do it and he couldn't do it.

Long story short, I started figuring out, “What leaders do I know? What sales leaders do I know?” I didn't know that many relative to a bunch of other people who were doing the job I was doing, who've been in the industry for 30 years, who are at the tail end of their career, and who do have this coaching tree of leaders. So I couldn't really help him hire somebody.

I realized then that I needed an excuse to get to know people I didn't know today. That was the genesis: how do I figure out a creative way to get to know these chief revenue officers and help them tell their story? KP was like, “Maybe prove it.”

They came to you, you came to them, or they came to you?

Who?

KP.

Oh, no. I went to KP and said, “I think we should start a podcast that interviews CROs.” They were skeptical—skepticism would be generous. We don't do a lot of talking as a firm. We generally let our portfolio and our founders speak on our behalf.

There were other venture podcasts, but most of them are pretty cringe, if I'm being honest. KP was skeptical, and I realized I also hadn't done a very good job articulating what I thought it could be. So I recorded an episode anyway with my old boss at the time, sent it to some of my partners here, and said, “Hey, this is what it would sound like, if you're interested. I still have a job here, so let me know.”

They said, “Oh, this is actually better than we thought.” They had to feel it. It's like a product that they had to actually feel.

To Allie's point, we were like, “All right, let's just do 10 and see how it goes.” After 10, we were like, “Oh, this is actually kind of working.”

I'm getting to know these CROs. It's getting easier to get to know them. Then I made a commitment to myself that I was going to get to 100. I told myself, “I will not make a judgment on what this is or isn't until we get to 100.”

Yeah, it's the same. It's really weird that this number also appears when Marques Brownlee talks about how to start being a YouTuber, because you just don't know what you are until you give yourself room to experiment.

My observation is that—I don't know if you feel this way—it's a very vulnerable feeling. Even if you're the one asking the questions, not answering the questions, you're really out there. You feel very exposed.

And the comments are a vicious place. So, for me, I was like, “All right, until you get to 100, you don't really know what the quality of the work is.” You're making a precommitment that you're going to tune out the noise, because otherwise you start overreacting to what any single person thinks about any given episode.

Usually, I think that's why most podcasts don't make it past 5 episodes: people start to be like, “Oh, maybe it's not good.” They start reacting. Each of these slices feels more real, and so that's why I made the—yeah, that's why I decided to do it.

Yeah. You started audio-only, right? So you didn't—

Well, the beauty of a podcast is nobody can talk back [laughter] in the comments, because there are no comments.

I guess with audio iTunes reviews—

With audio—

Right. Right. Right.

Yeah. Yeah. It was audio-only. It was actually easier when it was audio-only.

Yeah.

In many ways. And now everyone has to be video. Actually, I think the core reason why we wanted to do video was it became very obvious—for me, I listen to more podcasts than most. I'm quite voracious about listening to podcasts, and I realized my behavior changed once I got YouTube Premium. If you turn your phone off, or whatever, it's just black, just noise. Then when you turn it back on, you flip the home screen up and it's video.

I wanted the multimodality where, for example, if I'm cooking, the video's open, or if I take a shower, the video's open. Then, if I'm on a run or something, I just click it off and it's just the audio. It was becoming more and more obvious to us then. I think it's quite clear now. I suspect you'd agree that you need both. You do need both.

I think the question is, can you just transition to video without any change in the format whatsoever? I think that has difficulties. For example, we are a technical podcast, so we can show code, diagrams, and demos of the product. Do we spend the editor effort and money to put that into the video on the off chance that the 5% of our audience watching on YouTube actually sees it? I don't know. It's not super clear.

The other, more relevant thing is, when you look at people like All-In or Dwarkesh, they started video-first and then went audio. It's kind of a video-first mentality. I feel like somehow, when you start video-first, it translates better to audio than the other way around. My only knock on video is that it's, by default, more produced.

Yeah.

And so much of what I want to do is just have a conversation. The minute that you have cameras everywhere with lights all over the place illuminating something, it feels more noticeable to the guest. It just feels more produced. Therefore, they start to imagine themselves as if they're on TV: “Here's what I want to sound like when I'm on a podcast,” as opposed to, “Here's what I just sound like.” You know?

As soon as you see cameras, you start acting differently. It's like, well, this is not how you are in real life. It's just a different thing. That's my one knock on the video format.

Yeah, I noticed you also did the thing to me where you just start the conversation, right? You don't have, “Well, here's the intro, here's your birth story, here's your origin story,” which I try to do sequentially a little bit. But that's one of your tricks, right?

Well, I would say I go through an extreme level of detail to make sure that the guest feels very comfortable when they sit down. One example is how you're greeted at the door, water, all those things. The second is that recording just starts. There is no, “Okay, are you ready?” Because the minute that somebody says, “Okay, are you ready? Go,” you climb up. You're like, “Okay, I'm going to be the guest that I want to be.”

It's like when you're lying in bed at night before you go on a podcast. You're like, “Okay, how am I going to sound? What am I going to say that's going to make me feel smart?” You know what I mean? Make me sound smart. You start to build this idealized version of yourself that you want to project to the world, which is not real.

I start with the temperature of the room. I like the temperature to be cold. I don't want people to feel like they're sweating or hot. It feels kind of cool in here, right? The way that the lights are—you'll notice the lights are all up, not down. I think it's important not to make it feel spotlight-y.

Yeah.

If that makes sense. The way that I do prep—and then for the guest prep, for example—I will have read everything about them. I'll build my own mental model of who I think they are, and then I'll spend the conversation poking at that mental model.

I never give the guest the questions, because if you give the guest the questions, all of a sudden it's a rehearsed set of conversations, which is not how real life goes. I go through a lot to make sure that it feels real.

Yeah. It's funny because we get asked a lot for questions up front. For example, the Zuck pod was very, very, very well-prepped and screened. Sometimes you just don't get the interview if you don't do that.

I won't do it.

Yeah.

I just won't do the interview.

Yeah, we're more flexible there.

Because, look, the secret is the PR team is going to screen the questions, but you can go off script. You can ask follow-up questions.

So it's really not that bad.

True.

It's really not that bad.

That was Creative Corner. I do want to get you on the professional side, but I love indulging in Creative Corner.

Coming to Glean: Glean is obviously Kleiner Perkins' most recent incubation and success, and it's done super well. What's something that you realized working there that made you understand, “Well, here's what works in applying AI to the enterprise,” or whatever—selling AI to the enterprise?

Yeah, I think at the time, in the early days of Glean, when Arvind came to Mamoon with this pitch of doing enterprise search, it was the eye roll of the industry. If you asked any chief information officer or venture capitalist, they had been hearing that same pitch for a couple of decades, where everybody had promised—the best companies in the world, Google included—that they were going to solve enterprise search once and for all.

I give Mamoon a bunch of credit because he realized that if this problem was going to be solved once and for all, it was probably going to be somebody like Arvind who could do it. On the technology side, this was 2018 or 2019. LLMs had not been birthed yet.

In the beginning, it was a pretty serious slog with Glean, because you're asking these systems to crawl through an organization's entire corpus of data, do it with all of the permissioning, do it with all of the auth, and handle this multilayered cake of protections to make sure that I never see what I'm never supposed to see. I never see swyx's comp data, for example.

Getting that right is insanely hard. Then let's just assume—which, in Glean's case, they did—that you get the technology right. Then you have to figure out how you get past the eye roll of all the people who are default skeptical.

Yeah, the category is just dead to them.

The category is completely dead to them. Then you have to figure out how to deploy it. Do you deploy it on-premises? Do you deploy it in the cloud? There's all the security. Then you have to go through an insane number of hoops because this is pretty sensitive information that you're indexing. You have to go through all of that.

I would say the thing about Glean is that now it's become one of the obviously great AI companies that's in the heart of the hurricane. [Snorts] Back then, it was extremely unobvious. Extremely unobvious.

Let's double-click on that, because I feel like we just did the breadth of the problem, but let's talk about getting past the categories that have that sort of default rejection. What do you do there? What did you learn? What did you try that didn't work?

The advantage that Arvind had was that he was previously the cofounder of Rubrik. He noticed this problem at Rubrik, left Rubrik to start Glean, and he basically built Glean for Rubrik. The bet that he was going to make was that what Rubrik wanted, in a rough-and-tumble way, was probably what the rest of the world was going to want.

So it was like their core design partner. I think he had access to all the right people, and he knew all of the systems. That is really important: having these early believers who are willing to co-develop the solution with you and give you unfettered access to get things done. It was still an insane effort to do it, but I think it made it a lot easier. That way, at least you can show, “Hey, this is working in production for somebody.”

Yeah.

Right. I’ll pause there. Does that make sense?

Yeah. It’s a design-partner process, and I think it’s a pretty common go-to-market strategy for early-stage enterprise companies. That’s your ability to communicate what you’ve built and then figure out how to get that through an organization.

It’s like passing a bill through Congress. You have this thing, Glean, and you have to get all of these stakeholders inside a customer aligned and up to speed on what’s in the bill—the product. You have to make sure you’re helping them manage their own process and organization.

This is no joke. I don’t think people realize how hard it was back then. I think it still is very hard. What’s weird is that you have a special expertise, and I think the engineers who are listening maybe don’t appreciate the work that this involves.

It is almost like a military mapping of the organization. You have to understand who your champions are, where the resistance is, and how you want to prosecute a campaign to go to market in a very targeted way. I think if you’re an engineer, it’s even harder now than it used to be.

The reason for that is that this technology is so new for organizations that they both have to figure out, “How do I use LLMs and AI within my own org?” and, “How do I use your product within that ecosystem?”

Right? They’re first trying to figure out how to use the underlying stack that’s changing underneath them, and then how to use your product on top of quicksand. That’s a really hard problem, which is why you see so many companies doing this forward-deployed engineer motion.

What they’re going in and doing is saying, “Okay, number 1: Here’s how we think about LLMs and where you can get the best use of them. Number 2: Here’s an engineer that we’re going to forward-deploy into your environment, and we’re going to co-develop this solution, custom-fit for this org.”

So you have to do a lot of handholding. The reason is because we’re just so early to AI right now. Of course you have to do a lot of handholding, and of course you have to surround these customers with a bunch of technical resources to make it successful.

Yeah. Are there buckets of regrets that you have? There’s also the question of whether you spent your life working on things that you care about. Even if the finances didn’t work out, are you still happy you worked on it?

If I did not like the mission, the job, or the people I worked with, that would be a bigger regret than the pure finance side. Obviously, finance does matter. I always think about it as, in terms of ranking, you should probably put people, products, and money in roughly that order.

If you’re evaluating a company—

Yeah, as an employee, obviously. If you start—

People, product, money. Okay.

Because product is either the shit, then people, and then the money. If the first 2 don’t really work, then basically no amount of money will make up for it.

Yeah. Okay.

Right.

I can buy that stack rank.

Yeah. I would maybe reshuffle it a little bit, but I could buy that.

Maybe I would say people—

And then I would double people. I would probably add it again.

Sure.

And then maybe market, product, money.

Yeah, market’s really good. I guess I might bundle market with product, or just assume that market is a given, given that I only work in dev tools. But yeah, that’s an important distinction.

Coming back to you—or just the general learnings from KP before we go into Roadrunner—I also wanted to touch on the other conversation you had with Varun from Windsurf, which I really enjoyed. Shout-out to him. He’s going to listen, obviously.

Love that guy.

Another interesting company that I was finding parallels to Glean in, in the sense that you have to get people to hand over their entire codebase. It’s a very tough go-to-market. I’ve heard from multiple Windsurf folks that you went there and did sales training and cheerleader sessions.

What is it that you do at Windsurf, or in general at KP? You can use Windsurf as an example, but use that to tell the story of KP.

When I joined 6 years ago, my charter was twofold at KP. Number 1 was, “We have this group of CIOs and customer networks. Can you help us manage it?” Number 2 was, “Our founders need a lot of help on sales and distribution. Where you can, can you help them there?”

Mhm.

That was the core charter. Then I realized that in order to help founders with go-to-market, we needed to help them hire. So that was the excuse for the podcast, right? I was like, “All right, I need an excuse to get to know these people so I can help these founders hire great CROs.”

Then that all started to work. We were like, “Great, let’s double down on helping founders with sales.” So we hired somebody on my team, Liam. Then we were like, “Great, let’s double down on helping folks like Varun get access to world-class customers.” So we doubled down on that and hired somebody else.

Then we were like, “Great, let’s help founders build their demand-gen funnels,” and a bunch of stuff on the marketing side. Okay, hire Suzanne.

That was kind of the first 3 years: How do we help? KP generally invests in technical founders. That’s a majority—not all, but a lot—and those technical founders are generally exceptional at product and engineering. They usually have never closed a deal before or created a top-of-funnel demand flow, right?

We wanted to, as a firm, really help build that muscle for KP founders. Varun is a great example. He’s an amazing engineer, but he’s never had to actually build the machine that is sales and marketing.

I think this is something people don’t appreciate about Windsurf. They look at the product and understand that it’s an IDE, but actually there’s a sales machine that is one of the best I’ve ever seen.

Yeah. Yeah.

And you helped build it. Tell me more about it.

I think—and maybe we can quantify it as well, right? Something like 0 to $100 million ARR in 7 months, something like that. 8 months.

It was the most torrential growth I think I’ve ever seen in a KP company. It was insane.

That’s a high bar because—

It was insane. You have some pretty good companies. It was insane.

Yeah. Tell me more.

We’re seeing companies like Harvey and others following a similar path. You know better than anybody that coding is an incredible use case for AI right now, but I don’t expect government or Fortune 500 companies to adopt at this kind of rate. That’s why I was seriously miscalibrated.

I did a podcast with them on the day of Windsurf’s launch, and even then I was like, “I don’t know. This seems like a Cursor clone.”

I think what Windsurf got right was probably a couple of things. The first was a commitment from the founders that they wanted to both build a Google-class product and Salesforce-class distribution. It was a true commitment from the beginning.

Yes.

A lot of founders will say, “The product will sell itself. As long as we build a good enough product, people will come. They’ll come, they’ll pay $20 a month, and then they’ll just love us so much that they’ll magically upgrade.”

Exactly. Which usually doesn’t work that way.

So that was, I think, number 1: a real commitment to doing it up front and knowing that if you can marry those 2 things, it’s magic.

I think the second was that they were in a great market that was getting pulled. Generally speaking, the coding market still today is just getting dragged by the industry because it’s such a good use case. Engineers are expensive. Having a copilot for them makes a lot of sense.

The technology is there to be able to take the structured nature of code and reason with it, then produce outputs that are great for engineers. So that was number 2. And then I think the third was probably just that they moved fast. They hired great, and they hired fast.

Yeah. Were you involved—Graham, Jeff?

I would give more credit to Liam on my team, who was intricately involved in building out that entire go-to-market. That team, I guess, is now at—

Cognition.

At Cognition.

So you’ve seen it firsthand.

Yeah.

No, like—

It’s no joke. This is why I put it as part of my Cognition thesis. Cognition, the company behind Devin, is very good at product and engineering, but they didn’t really have that much of a sales team.

Here’s the most cracked sales team I’ve seen in coding, at least in dev tools. You just bring them together—how hard can this be? This is a really good formula for success.

Yeah. I don't want to understate how serious Windsurf was.

Yeah.

About distribution, not just product.

Okay. The lesson that I take away from them is that they were as serious about building an incredible product as they were about building incredible sales and go-to-market.

Yeah.

And it's easier said than done.

Yeah. You can't be serious about everything and have everything be the number-one priority.

That's right. What the hell?

But they pulled it off, which is impressive. One anecdote I will share on the distribution side is that they're the first company I've seen where one whole floor of the office is dedicated to video production. One floor of their office down in the city is just a studio, and I've never seen that. I'm like, you're a pretty young company, you're mostly developer tools, but here's a whole studio set that you can do anything out of and make interesting videos.

It's because you really care about getting this across, even though you're just selling software. Totally.

Like, I'm sure Glean doesn't have it. I don't think I've seen a video from Glean that's not just a screen share.

Totally.

Okay. Give me one more thing on how you hire a sales team. We have founders listening who are building interesting products but don't really know how to go to market. Do you have to offer an arm and a leg to hire your first sales leader? Do you have to work only with Kleiner to do that? What's the actual principle that you advise founders to follow?

I'll give you some antipatterns. The first is: don't just go on their LinkedIn and look at all the fancy logos where they've worked, then immediately assume that because they were at Snowflake or Databricks, they must be good for your AI company. It just doesn't work that way.

In fact, in many cases, the inverse is true. If you had to sell the No. 3 product in a market, fight tooth and nail, and were still successful there, you're probably going to have a much higher proclivity to do well if you go to a great company. Whereas, if you joined Snowflake at $100 million in ARR and joined their enterprise team in the Bay Area, it's like, yeah, I get it, but that's not that impressive. No offense to anybody who joined Snowflake at that time. There are some diamonds in the rough.

Well, it's more like they're a fit for exactly that scenario if you're in that scenario. But you're not.

That's right. Especially for startups. The problem is that you actually have to interview them. You can't just see what they did on their LinkedIn profile and know whether they're good or not. You have to actually dig in. And it's not necessarily all of the things that they've actually done that make them good, because you're hiring for potential.

It's all sorts of intangible things that you have to feel, right? All the same things that we would want to feel with a founder. Do they have a chip on their shoulder? What are they motivated by? Is it money? Is it living in the shadow of their brother or sister? Is it that they grew up in a first-generation immigrant household? Whatever it is. So you go really deep on the background.

Really deep on understanding that there's going to be a million things that go wrong here. When they do, what is the driving force that's actually going to push you over the hump? Especially in sales, you get told no way more than you get told yes. After you get told no enough times in a row, some flame within needs to continue to burn to keep pushing you.

This is why every executive recruiter and everyone else has it so wrong in most cases, because they just go to the fanciest LinkedIn profile and say, “Oh, yeah, this person has all of these great logos. This is the person you should hire.”

I'll give you an anecdote inside the KP portfolio. Take our top 8 companies, companies like Rippling and Glean. Take 5 executive roles across those top 8 companies: 38 out of 40 of those roles are executives reporting to the CEO for the first time in their career.

Okay, so what does that tell you?

Generally, their experience is not the thing. It's the context that they've built, the trust that they have, and their ability to learn fast and grow with the company. It's not what they've done at their last 5 companies.

Does that make sense?

Yeah, it totally makes sense. It's very first-principles thinking, as Scott Wu would put it.

Yeah. I'd say that's probably one big failure mode.

I think the other one, especially in AI today, is that the bar for how technical you are is going up. It's just going up by default.

So salespeople have to be technical.

Much more technical. That's a tough one.

More technical than they used to be.

What do I mean by technical? You don't have to understand every intricacy of the transformer.

Exactly, the transformer.

But you should be able to go over to an engineer's desk and ask the right questions to get a depth of understanding that you can actually communicate and articulate effectively to a customer. I think that matters a lot. That bar has started to rise more and more for me in terms of whether you can actually describe the product, what you do, and how it fits into a broader ecosystem without relying on a sales engineer to do it for you.

Yeah. But in an interview, sometimes you just don't get that, because that's what sales training is for. People prepare battle cards, get time with the product and the founders, and then they get it right. It's hard to get that in the interview.

Yeah, but it's not hard to look at somebody's background and understand whether they were willing to do that.

Yeah. Yeah.

Did they actually want to do that?

What's really funny is that one of my core memories as an engineer learning the ropes in startups was our new head of sales coming in at Netlify, which is a KP company. I was very stressed. I was like, “Our competitors have all these things. We need to match them and exceed them.” And he's like, “Nope, that's engineering thinking. Give me anything, I'll sell it.” I was like, “Wow, that's a good sales guy.”

But I think, to some extent, salespeople who can sell regardless of the product—that's the old-school way, where they know how to do the steak dinners and golf and whatever else they do to make their number—versus now, I think, the rise of the more technical sales hire, who really has to care about the product, explain it, and get into the weeds with people. I think that's a shift that I'm seeing.

I'll add one more thing that really matters: have they worked at a company that is similar in size?

Yes.

For example, if you're a seed- or Series A-stage founder and you're evaluating sales leaders and AEs who have only been at companies that had $50 million or more in ARR when they joined, it's probably going to be really difficult for them.

The reason is that they've had a brand their entire life. They've had inbound leads that just come to them. They can generally lean on the credibility of the company. They've had a playbook that they just have to execute and run.

All of these things make the experience of being, say, the first sales leader or first AE very different. You're way more of an artist than a scientist. It's not a MEDDIC-type playbook in the very beginning. When you get to where Windsurf was, or is now, it's very systematized. It is a machine. There are boot camps and battle cards.

But in the early days, it's creative ways of getting something done. It just looks more like art than it does science. If you've never had to do that before, it's going to feel quite foreign to you.

Yeah, it really is. This is why people with more experience can come in and show us the ropes.

By the way, I probably would not be a good salesperson at Windsurf today. I'm not the person to execute this perfect playbook that was handed to me, where I'm qualifying criteria per letter of the playbook. That's not my thing.

Yeah, because people still talk about you internally. I don't know what you did. You just did motivational sessions or something.

Honestly, they did most of the heavy lifting. I probably went in there and did some random rah-rah stuff, which they need, and then introduced Varun to a bunch of customers. But the KP team—my team—did the majority of the heavy lifting, so I give Liam, Lauren, and Suzanne a bunch of credit for the work that they did there.

Yeah. Okay. So now we come around to your current thing. A few months ago, I think 2 months ago, you sat down and told me, “I'm working on a new thing.” It was super stealth and secret, but it's going to be the hottest new KP incubation since Glean, and I'm super interested in it.

All I know is it leans on basically everything you’ve done, everything we’ve talked about. But can you introduce Roadrunner and the thesis?

I would say you’re right. During my sales career, probably the number one thing that used to break my back was that the underlying software, like Salesforce CPQ and others, just to create a quote and get it approved, is horrific. You think if you’ve seen bad software, you haven’t until you’ve seen a 30-second loading screen to get from one page to another when you’re trying to close a deal with 2 days left in a quarter.

This is standard across the industry. This is just how it works. This is how it worked at every job that I was ever at. I used to get yelled at because I would be asking people to turn something around within a day or 2 because I needed to get a quote out the door. It happened when I was leading teams. They would always be getting yelled at because they were trying to move too fast for these systems to work.

The reason the underlying systems do not work is that pricing models went from a world where it’s like, “All right, Swyx, you want to get a Netflix account? You have 1 seat that maps to 1 person, and it’s $10.99 a month.” Very simple, right? Then you’re like, “Okay, actually, I want a family plan.” Now you can add 5 people, no more than that, and it’s $8.99 a month.

That’s how pricing models have generally worked in the B2B context. It’s like, “I want to sell 1,000 licenses of, you know, pick your product—LinkedIn, right?” That maps to 1,000 people at an organization. What has happened is that companies now have 30 products, more—50 and 100 in some cases. Those products scale by volume, and then there are discounts associated with them. Then you have to do renewals, early renewals, expansions, and do them across, say, 15 different product lines.

The complexity has just started to increase exponentially. Then you’re like, “Actually, I want the customer to pay as they go.” That may be how you guys are selling today. It’s like, “I just want to make sure they pay a minimum amount, and then anything above that, we’ll just bill them.”

Very custom function.

Exactly. That’s how Cursor and others are, too. It’s the amount of tokens that I consume—just bill me for that. These pricing models have gone bananas. By the way, this has barely even started.

The reason it’s barely even started is that with AI, all of these pricing models will, at a minimum, start to look like consumption-based pricing. That’s how you consume Anthropic, and that’s how you consume OpenAI.

Yeah.

The problem is about to get way worse. It was a problem that I was feeling because I was like, “The underlying data model is just breaking.” Twenty years ago, Salesforce CPQ was not designed for all of these permutations. It was a static world where 1 person is 1 Netflix license. The most you could do was a family account.

Yeah.

That happened. Then I joined KP, and Lauren on my team and I started a group of 35 tech CIOs—companies like Uber and Box and others—that meet twice a year. This was 4 years ago, so pre-LLMs, pre-anything.

I asked them, “What is the number one problem that you have in your company right now?” I was at a dinner with 5 CIOs, and they were like, “CPQ.” I was like, “No way.” They were like, “No, I’m not kidding you.” They said, “We are getting yelled at by our chief revenue officers and salespeople all the time because the underlying software that we’re delivering to them doesn’t work.”

Fast-forward 6 months later, we have a second dinner with a different group of CIOs in this network. I’m with 5 other CIOs, and I ask them, “What’s the number one problem you have in your company?” All of them said the exact same answer. I was like, “Whoa. That’s pretty rare.” Pain does not grow on trees like that.

As a firm, we got very excited because we were like, “It’s pretty rare and unique to have this many customers that have this bad of a uniform pain.” So we did a full market map. We were trying to invest in a company, but we didn’t find anything compelling. We just could not find any great companies.

Then GPT-3.5 came out, and I was like, “Oh, my God. I actually think you can abstract away a bunch of the complexity with these LLMs. It’s unstructured and structured text that you can reason with and do stuff with.” That’s why coding is such a great use case. That’s why Harvey is such a great use case, because you have all this case law, and then you can point the LLM at it and reason with it. Then you build a bunch of enterprise features, functionality, and workflows on top of that.

This is a very similar problem in nature. That was the light-bulb moment of, “Okay, I think we can actually build something better.” Then I started asking myself, “Why is nobody fixing this?”

Yeah. Why is nobody fixing this? That’s the question. Why has the incumbent, Salesforce, or anybody else, not fixed this? Why is this still an issue?

The reason is that all of these tools were basically built in a pre-LLM era. Their data models are broken because they did not foresee consumption, 1 million SKUs, and the sprawl that comes with them. In order for them to build a product that handles all of the complexity and permutations, they have to rebuild their entire data model and architecture from the ground up.

That’s the same thing that most incumbents have to do today, which is why there’s so much frenzy around early-stage startups in VC. In order for an incumbent to go do what Harvey is doing, you have to literally rebuild that company from the ground up. You have to build the entire architecture differently.

It reminds me of when I was in the public cloud. My career was in the public cloud before this, and in the very early days, everybody was moving from on-prem to AWS. Initially, everybody was like, “Great, we’ll just lift and shift our application and throw it into the public cloud.” Then all of a sudden, you realize, “Oh, no. S3 buckets can just disappear.”

You actually have to rebuild this stack cloud-native from the ground up. That’s the same thing that’s happening in AI today. That’s the classic innovator’s dilemma: What do you do?

Yeah. Do you rearchitect, or do you wait?

So that happened. Then I come to find out that, in this case, Salesforce, which is the gorilla in the room, has 95% market share. They have end-of-lifed their CPQ solution, and they’re making everybody move to a new product.

That product doesn’t exist yet. If it does, it’s incredibly flimsy. We’ve talked to some of the people who are trying it right now. We basically have a 2-year window where we have to beat them to the punch, and we love that.

The reason we love that is, boy, would I rather compete with some 100,000-person Salesforce, where I don’t even know what kind of engineers may or may not still be there, versus OpenAI. That’s who we want to outsprint.

Then I started asking myself, “Why hasn’t anybody done this yet?” The short answer is, 1, I don’t think the technology was there. 2, going back to your earlier question, Swyx, this is a very complicated go-to-market and distribution question.

It is upmarket. The problem is more upmarket because that’s where the complexity is. In order to do something elegantly upmarket, you need to know what you’re doing in the enterprise.

Yeah.

Right. It just so happens that you need early believers, like Glean had with Rubrik, who are willing to take a bet with you in a design partnership to co-develop it with you. It just so happens, going back to our earlier conversations, that episodes 1 through 80 were interviewing CROs, who are the people that have the pain.

You’ve been preparing this the whole time.

The other thing that I was responsible for at KP was these CIO networks, who are the ones responsible for delivering software to these people to alleviate their pain. I just so happened to know basically all of the key stakeholders in this problem. Do you get it?

At that point, I was like, “Oh, man. Do I really want to do this?” Life at KP is pretty good. I know I’ve seen what it looks like—the bite out of your life that it takes to build a company.

After talking with my partner at home and understanding whether this was a commitment that we were willing to make, and talking to my partners at Kleiner Perkins, folks like Mamoon and Ilya, I was like, “Hey, if we do this, I think I have to do it. I don’t think it’s going to make sense for us to hire somebody off the street right now, and I’ll build a co-founding team that is technically excellent and world-class.”

That was the thing—or, I should say, the million series of things—that tipped us over the edge.

Well, where are you today? What are you ready to share in terms of what the product is, the people you’re working with, and the problems you’ve solved?

Yeah.

So, I will work backwards from the list that you used of how you would evaluate companies, because it's the same thing as me.

Wait, money first? All right, let's see.

Team first.

Oh, okay.

Team was your first, wasn't it?

Yeah, but you said backwards.

No, no, no. I'll work top-down.

Okay.

Team first.

Yeah.

We're at 9 people today: 2 co-founders, AJ and Eugene. AJ went to Caltech at 15 and finished second in his class. The guy was in diapers when he was in school. It was absurd.

He met Eugene on his first day of school. They've been working together basically ever since. AJ went to Robinhood. Eugene went to Meta. They went to NASA together and built a bunch of the software for the Mars rover.

Then they started a company together called Athena, which was an LLM for college students to send in their applications. It would grade them, tell them how it was, all that. Then they realized that the TAM, or end market, of edtech wasn't that compelling.

Sometimes it depends. If you're capital-efficient, you can do this business.

It depends. They were not that inspired by it. They wanted to go build—

Salesforce CPQ.

How do you deliver better software for AEs? Yeah, CPQ. Okay. They were asking their founder friends, “What's the number 1 problem at your company?” and they kept getting this answer.

So anyway, then we met. It became very obvious that what I had was unfair distribution and an understanding of the problem coming from sales, while what they had was extraordinary technical chops.

That's team.

That's team, plus some killers across the board.

By the time this comes out, we'll have soft-launched. We're co-developing the solution with design partners. I was very inspired by what Glean did. Glean built for Rubrik, and we're building for 4 design partners with a shared Slack channel and weekly stand-ups.

We're taking the same bet that Glean made, which is that what these 4 design partners want is probably what the rest of the world wants.

Do you want diversity in those 4? You know what I mean?

What you really want is to make sure that our data model is infinitely flexible, so we don't run into the next permutation that we haven't seen. What you want are the hairiest—

Design partners that have—

Every SKU—hardware, software, SaaS, consumption. You want the mess to throw at your data model to make sure that nothing tips it over. Those are the types of design partners that we wanted.

It's helpful because I know the CIO, I know the CRO, and in most cases, I know the CEO. You don't have to deal with the normal big-company BS of legal and procurement and all these things. They can help shepherd you through the organization.

Yeah. Let's talk about the data model. Did you get it right from the start, or what were the biggest changes that you've made since you started?

The team probably spent its time only doing the data model. That's it—late nights.

And what does the data model mean?

For example, there are rules that every customer has. If you're an AE in the UK, you can only quote certain SKUs that have certain discounts on them. You can only have so many—there's a maximum discount that you can present to a customer, right?

If you're doing a deal through a channel partner and you're doing it out of Canada, there are all these rules that are connected to it. Imagine that all of these SKUs, all of these rules, and all of this stuff are an extraordinarily interconnected system that has to be accounted for.

Making sure that we threw as much of the real-life information, rules, and permutations as possible at the data model so it wouldn't tip over was incredibly important. That's where we spent a large portion of the time getting it right.

Yep. I also know that the best-laid plans run into reality and then get screwed up with the first contact, right?

For sure. Even if you get consumption right, for example, where you're like, “All right, they can do it,” how do you represent that in the UX, right? Cognition or Windsurf, like, magic. If you get that right—which is, I think, what we're about to get right—then you earn the right to go build a big company.

I like the way that you phrase that. I love earning the right to do bigger things. You know, what makes me uncomfortable with founders saying, “We'll build a compound startup when we're as early as we are” is that the ambition is very big, but have you earned it?

100%. Can I actually tell you one other thing that kind of annoys me? I think a lot of times in the Valley, founders will pretend that the mission they have is bigger than it is.

For example, if you're doing what we're doing, I think we're solving a really important problem for a certain set of people—high-value people.

Yes.

But we're not helping with a mental health crisis. We're not feeding people in other countries. We're not doing what the Chan Zuckerberg Initiative is doing.

I think it's really annoying when people pretend that what they're doing is this revolutionary thing. No, what you're doing is solving a really hard problem for a specific set of people. If you're able to solve that problem and those people are delighted, then you earn the right to go solve the next problem.

If you solve enough problems in perpetuity, then you earn the right to go build the big company. When you go build the big company, you get all of the cool things that come with that: people taking on bigger roles and responsibilities than they ever had, engineers owning new product initiatives soup to nuts, and ICs becoming managers who have no business becoming managers. Of course, there's all the financial stuff that comes with that.

In my mind, that's a mission I can get behind. I still think the problem that we're solving is interesting and cool, but I think it's really annoying when you're like, “Oh my God, this is the thing that I've been thinking about since I was 1 year old, and everybody else also has to feel like this.” I find that annoying.

Well, sometimes you have to pump yourself up for the fundraise, but there's always sort of 2 versions of the story. We're an AI engineering podcast. We do care about how AI is being utilized to transform and revolutionize things. I think you're going to find it in small little ways, but are there any surprises?

I think, if you dream the dream, where LLMs will be a superstar in this company is if you're one of our customers and all quotes go through Roadrunner. You can imagine a world where it just recommends, “Just do this deal.”

“Okay, you're doing a deal at Costco. Great. We just did a deal with Nordstrom that looks a lot like this deal. You should adjust these things and then deliver it this way.”

So it would proactively suggest?

Exactly. The system will basically have all the historical information about what you've done.

And once it has all of that, it will then tell you, “This is how you should bundle it up.”

By the way, today that's all human-in-the-loop. Today, if you're a new rep at Glean or a new rep at Cognition and you want to put one of these deals together, you're calling the deal desk and finance. You're calling the top AEs at the company and saying, “How do you even put this together?” Right?

Then, if you're using Salesforce CPQ, which most people are, you go into loading-screen hell. Then you build a bunch of custom software on top of that because the data model doesn't work, so you have to fit around it to actually make it work. It's a complete nightmare.

That's where the magic of Roadrunner will come.

So, if I can abstract a little bit, you're kind of automating the deal desk and not the AE. You're extending and augmenting the AE. You're improving ramp-up time or productivity for the AE. How would I describe it? Whose job are you taking away? [laughter]

Very specifically, AEs are quite expensive, and they spend a ridiculous amount of time doing administrative work trying to get these hacky systems to go.

Perfect.

The amount of bouncing around and ping-ponging that they have to do inside an organization just to get a quote created and approved is a nightmare. We should solve that.

And by the way, guess what? The deal desk and all these people should not be doing that either. There's way more strategic work that they could be doing.

Yeah. I mean, at Cognition, it's just a really active Slack channel where everyone's throwing stuff at each other all day long, and it's a mess.

Yeah, it's mayhem.

Yeah, it's complete mayhem.

Interesting. Okay. Is there anything else you wanted to cover on Roadrunner in general, like your vision? I think we covered a lot of it. Is there any part of the story that you want to get on the record?

I would just say that we're not demand-constrained. I know every customer, and they're all banging down my door right now. The only fight that I've ever had, or have, with my co-founders is that I'm like, “Hey, these 10 customers want to come and join us and work with us,” and they're like, “We do not have engineering bandwidth.”

Our roadmap is being dragged out of us. It’s very clear what we have to go do. There are no surprises in the things that we need to build. Obviously, the strategy is the same; the tactics will bob and weave. We are meaningfully bandwidth-constrained on amazing talent that wants to go through the grind of building an early-stage company.

Yeah. Well, we’ll get you that. Kleiner is very good at getting that. I don’t think there’s any doubt there.

Zooming out a little bit, I was promised earlier that you like running a lot. Tell me more about the general philosophy of high performance. What does that mean to you, from your personal life into your work?

Uh-huh.

I’ll try to be tactical rather than abstract.

Physically, to your prompt on running, I work out every day, no matter what. I sweat first thing every day as soon as I wake up. It’s pretty consistent. On Mondays, I bike up Hawk Hill, which is over the Golden Gate and up into the Headlands. It’s a nice way to start the week.

I’ll run twice a week, usually lift weights twice a week, and usually play a sport or something.

What sport?

Basketball has been the sport of choice recently. Although every time I pick up a ball, I’m pretty convinced I’m going to tear my ACL or something.

It depends how hard you push yourself. I guess you push me hard.

Yeah. Physically, I work out every day and have a salad for lunch every day. I’ve been doing it because it’s just easier. I’ve found that if I can do the same things over and over again, my life is easier. I don’t have to think about it.

For example, with working out, it’s way easier to know that I’m going to work out every day rather than have the cognitive load of figuring out what days I’m going to work out this week or when I’m going to do it. For lunch, I’m just going to have a salad and have salads for lunch.

Yes. There’s no choice. I don’t want to make the choice. The way I phrase it—I think Tim Urban says this—is that it’s much easier to do something 100% of the time than it is to do it 90% of the time.

100%.

Right? Think about if you were deciding, “I want to work out 4 days a week.” Then you have to figure out what you’re going to do.

Yeah. Did I skip yesterday? Does that mean I can skip today?

Exactly.

Even with the style of workout, I’m just doing whatever I feel like doing that day, besides Mondays. Even if I go and lift at the gym, it’s just full-body. Whatever I feel like is next, I’ll compound every exercise and run through it. I don’t have a set routine because I just want to reduce friction as much as I can to get those things done.

On a personal reflection, there’s a real reason I’m asking this question, but just as a side comment—and you can comment if you want—I feel like this is so important: your personal health and fitness and your peak productivity practice. I find it interesting that VCs don’t do that for their founders. It’s like, “Hey, I’m going to lock you in a room, basically. Only HF0 does this. I’m going to make you eat healthy, make you take care of everything, so you can go work on a company.”

Cognition has an engineering basement, and I’ve advocated pumping oxygen into there because that’s a very valuable thing to have.

The problem is that it kind of has to come from within. It has to be a habit that you’ve had and can then carry on to founding a company, because the amount of demand on your time is like a pie-eating contest, and the prize is just more pie. There are no limits to how much has to be done.

I think I just got lucky that I had some of these habits before, and then I was able to carry them on. Otherwise, even now, I feel very constrained in being able to do some of these things.

No, totally. The real reason I was going to ask is that you listen to a lot of podcasts while you’re doing all this. What are your favorite other podcasts?

Oh, man.

Do you listen to your own podcast?

She does on Rex[?]. Yeah, I do. I’m very critical. I send notes to my editor and my co-hosts.

I used to listen to everyone, and then my editor was like, “Dude, we’ve been doing this for more than 200 episodes. You don’t have to listen to every one and send notes. It just bogs everything down. Trust us.” So I’ve stopped. I’ve created some space.

Yeah, right. Anyway, I wasn’t asking about our podcast, but just other podcasts that you enjoy and recommend to others. I’m just giving people recs.

I think some of Patrick O’Shaughnessy’s Invest Like the Best episodes are pretty good. This whole thing with Colossus—

Yeah, it’s interesting.

I think some of Joe Rogan’s episodes and some of Tim Ferriss’s are good. Some of Shane Parrish’s are good.

I have guests that I like when they interview them, but I personally don’t follow any host religiously. I get interested in guests, and then I’ll go down the rabbit hole of what shows they’ve been on. I’ll listen to those, and then if I like the interviewer, I’ll think, “This is a new show. Maybe I’ll check out one or two more.”

For example, I’ve gone as deep as deep can go on Elon. I’ve probably listened to a bunch of them, and if he’s been on one, I’ve probably listened to it.

He’s not the easiest speaker to follow.

No, it is a bit jumbled.

Yeah.

So I’ll go deep on a guest. I’ll go deep on a guest.

Have you done speaker training?

No.

Coaching? We’re considering that because you basically do public speaking as part of your job, right? You should probably get coaching for it, just like anything else.

My reflection doing this with you now is that asking the questions is way easier than answering them. How’s it feel? I would say you have way more control over a conversation when you’re asking the questions.

Yeah, sure. I’m just loving this.

Exactly. You’re usually speaking maybe 20% of the time, and the guest speaks 80% of the time. You have a general sense of where you want to go, so you’ve got the plan here.

Yeah. Maybe if I were to do more on this side of the table versus yours, I think speaker training might make sense.

Speaking comes in all shapes and forms, including running a company. I view it as a very general use case. I run a conference, and I do the keynote every time, at every conference, so it really matters because I set the example for my speakers.

My favorite definition of sales is the ability to transfer enthusiasm from one person to another. When you’re recruiting, when you’re onstage at your conference speaking, when you’re an interviewer or interviewee, or when you go home for the holidays and spend time with your family, I think all of that comes down to how you transfer enthusiasm.

It has to be raw and organic. If I were to train or be trained, I would really try to get to the essence of how I can transfer my enthusiasm about whatever I’m talking about to those who are listening.

So, speaking of definitions, my favorite closing question: What is the definition of grit to you?

The namesake of the show came from Angela Duckworth’s book Grit, and I had the honor of flying out to Pennsylvania and interviewing her, which was amazing. The background is also that Penn pioneered positive psychology, and she came from that line of thinking.

Her definition is hard to define narrowly because she wrote a book about it, but it’s passion plus perseverance over a sustained period of time. The natural tendency when you think about grit is literally gritting your teeth: How do you endure? I think the operative word in her definition is passion.

The way I think about it is: How can I put myself in positions where the thing I’m doing, the thing I’m working on, the job I’m doing, the company I’m building, or the relationship I’m in are situations where I really care? If I really care, then I can transfer my enthusiasm to others. If I really care, it will feel like play to me when it feels like work to everybody else.

If I really care, as in the podcast example, I’ll just do it for longer than anybody else and outlast you. I think it’s all because I really care.

And so I think this idea of passion is probably the thing that I love most about her definition, which is: I just try to do things that I really care about. Therefore, it feels light for me, and I don't have to feel like I'm always gritting my teeth to do things that matter. It's probably a superlative form of grit that really captures that kind of flow-state grit, or passion grit, or whatever adjective you want to add to it. Totally. That would be nice.

But thanks for being on Latent Space. I feel like I've been experiencing the grit experience myself, and thanks for joining us.

It was hard for me not to…

Ask you too many questions. I know.

I tried.

You got one. Yeah. What was on your mind?

Do you have a dream guest?

Dream guest, I would say, is a supporter of ours who has promised to be on at some point. Andrej Karpathy.

Yeah. Yeah. He's been a mentor for a long time. He's a teacher. He's very authentic and super knowledgeable, and I think he's an inspiration for a lot of us who are trying to figure out, from trusted sources, the truth of what's possible with LLMs.

He's very autodidactic as well, which is something I strongly identify with. You don't really know something unless you've really taught it to yourself and built a version of it for yourself. He represents the simplicity and clarity that I want to see in the world, which I try to represent within Latent Space.

Really cool, man. I appreciate you doing this. It's the first time the tables have been turned on me. Crazy experience.

First of many.

Thanks, man.