Barry McCardel on Why Everyone Is Copying Palantir’s Playbook (And Getting It Wrong)

If there is not tens of millions, hundreds of millions of dollars worth of seats for us to sell because everyone's been laid off, the world's like ending. Like, I need to own a lot of guns. Right. The least of my problems is my fucking SaaS business. This is what it looks like to live in a republic. It's messy. You elect—

Right, so it requires on the part of other actors to sort of interpret what the will of democracy is beyond just like deference.

That's a republic. Every generation has felt broken in some way. I always think it's funny when people— we've never been more divided, bitch. We fought a civil war.

Palantir is one of the most mysterious companies in tech. If you're an outsider, it's a company created by Peter Thiel that's gathering intelligence for the deep state. If you're an insider, it's an overvalued enterprise software company that's heavily reliant on human consultants. Today I talked to Barry McCardel, CEO of Hex, which he says aspires to be the cursor for data. He was a longtime forward-deployed engineer at Palantir and saw the company from the inside. We talk about Palantir's subcontractor Anthropic's fight with the government and argue about how much tech companies should defer to the President of the United States in a democracy. But first, we also talk about what it was like working at Palantir, which he called mundane, while he simultaneously talked about projects in unnamed Middle Eastern countries. We dug into the forward engineer craze that Palantir pioneered, and these days every tech company wants to copy. Barry defends the model, but doesn't actually use it at his own company. Welcome to the Newcomer Podcast hosted by me, Eric Newcomer. I am the author of the Newcomer newsletter, which covers the business and people in the startups and venture capital industry. Check us out at newcomer.co. Thanks for listening. Here's my conversation with Barry McCardel. Well, literally, I heard today that we didn't get invited to Perplexity, some Perplexity event because they're mad that they were the most shorted company. It's not your fault.

You're just asking the question.

Yeah, you put me up to this stunt, apparently.

Yeah, I thought it was an interesting question. Who would you short? I think we're all— I think it's good that you're not able to short startups because I think that would get into a very pernicious and shitty game in Imagine if venture funds were like hedge funds where they've got long positions and they've got short positions and they're like hedging, right? That'd be fascinating.

The— all right. I'm here with Barry McCardel. Welcome to the Newcomer podcast. CEO of Hex. Spent a long time at Palantir. And as someone who's in my ear with fun ideas and on Twitter, assassin and spicy. So I expect you to live up to your Twitter persona on the podcast.

I'm sparse. Sometimes I tweet a lot and then I'll go quiet for like months. Yeah, you had a—

what you're The most viral one was on a subject we want to talk about. I don't want to butcher it.

It's the— It was, it's only for deployed if it's from the Palantirais region of France. Otherwise it's just sparkling sales engineering, I think was the—

Exactly. So we want to start off with Palantir, I think from two levels. One, sort of, I don't know, the YouTube show level, which is like, I don't know, people hear Palantir in the headlines and like what to make of it. And then we'll do our nerdy tech show question, which is I feel like every startup in the world has learned that you can actually be a consulting company if you call it a forward-deployed engineer. And not only will you be punished on multiples, you will be rewarded. It's like, oh, all of a sudden it's good. Like, throw humans at it. But let's start with like the spooky YouTube. You know, I mean, I just think a lot of people on the internet hearing about Palantir just have a like, ooh, this is, this is the deep state. Or like, having spent a lot of time, what, like 4 and a half years, 5 years.

Yeah. What?

Yeah. Is Palantir the big bad?

No, I— and I'll just— I'll disclaim, I haven't been there in a few years now.

You stay close to— and you know the whole network of people starting companies.

Yeah. And I'm still a lot of people there. I think the interesting way we understood ourselves then, and I suspect it's still very true now, and we said this in almost a slightly diminutive way, it was like, we're a data integration company. It turns out that all these companies in the world, these institutions, private companies and public institutions that you wanted to be able to operate better, really the problems all come back to some form of data integration. You actually have a lot of the data available in some way.

What you're saying is it's actually a really boring company with a CEO that says, "And sometimes we have to kill people." I mean, I'll be honest, there is a kind of boring core to it.

Like there were times where it was kind of like we're kind of just doing a lot of data plumbing and it obviously the brand is incredible. And I think the leadership there has done a great job developing and nurturing that brand. And look, look, I'm not saying that the results aren't there. It's a fabulous company. It's a company I'm very proud to have been at. I'm proud.

I almost feel like it's worth some like $300 billion.

I don't even know. I can't look at the stock price now because I sold most of my shares at less than this stock price. I have a very discordant relationship with it, but Yeah, like, I, I think at the core there's some really boring stuff that we just got really right. And the initial brilliance of Palantir in the early days was the ontology. It was this kind of insight of, you know, the slightly hagiographic version of it is like, you know, early team was going— the founders were going to these three-letter agencies and sort of realizing that the problems they had and, you know, the mythos was a little bit like the reason we missed 9/11 or one of the reasons you kind of missed events like that is because, You actually had all the information somewhere, but you weren't able to get the complete picture. And so by being able to integrate it and build the ontology on top of it, okay, now you can see the whole picture and you can sort of make it something that humans can reason about and understand and engage with. And that shape of dynamic is true in the government, uh, in the intelligence community, military, also very true in a lot of private companies. And so a lot of what I did when I joined, I joined when we had like a very nascent commercial business. So I started my career in consulting. I'd been like on the really nerdy data tech side of my consulting projects. My friend who worked at Palantir was like, you just come do this like for real. And I started, we didn't really have a commercial product at all. And most of what we were doing was just sort of going around and doing effectively these consulting projects, integrating data, kind of stitching things together, giving people finally like a complete picture of their organization and the problems they had.

And that by itself was really valuable because they would buy tech services but then not get the most out of them. Stitching them up with everything.

Yeah, you'd have, you'd have like the, that era, you know, 10 years ago, you'd have a lot of people building these like data lakes. It was like, oh, we'll get all our data in the data lake. It was like, yeah, you land all the data in this S3 bucket basically, or HDFS cluster, and it's like, now what? We have the data. It's like, yeah, you have it. Are you using, are you able to make better decisions because of it?

And Palantir would help them like visualize it.

And yeah, it wasn't even viz. You know, it's funny because people picture it like, oh, this analysis with the viz stuff. Our viz was always like pretty mid, is, is really the integration behind the scenes, which I know sounds kind of sexy and unsexy and vague, but like, it's literally like, okay, you've got these 5 different datasets. They're all sort of talking about the same thing. You want to get a complete picture between them. How do you integrate them? What are the relationships between these? How do I see an event or an incident that's happening here and go understand what are the other times something like that has happened? What were the resolutions? And is there a shared link? Like, that is a lot of the kind of thing. And it's, it's not like the actual, like, algebra behind the scenes is like, you know, it's just SQL queries. It's like fairly straightforward joins and stuff. And like, every once in a while we'd get wild and do some regressions, but it's like—

but really the reason it's forward deployed is you have to understand the customer enough to really understand.

Yeah, that's the business model around it then. And, and I think the forward deployed, um, thing, this, this kind of gets into the, the, the role and what people misunderstand. Like when I joined, you know, we had these forward deployed engineering teams and they still have them. At the time it was really like we didn't have a product. We were hacking a lot of stuff up. There was—

that was the whole critique of Palantir for years. It was secretly a consulting service.

Internally we'd like take issue with that. Like I remember these emails, some one of the guys in leadership would sign and being like, people don't understand us. They think we're just a services company. We'd kind of look around and be like, I don't know, it feels like we're doing services. I'd read this email from a windowless conference room in some town where I'm like, basically I'm like hacking together an Excel spreadsheet to dump into a Jupyter Notebook so I can export the results into our frontend app builder to make it look like we did software. It was like, I don't know, it feels pretty servicey. But what wound up really working was you go and effectively do services. Uh, and at the time we didn't have a core platform really to speak of in the commercial side that we were doing these services on top of. But over time we built that platform leverage. But you go and you basically do these services. You go in and say, hey, we're going to solve your hardest problem. You know, Karp, Dr. Karp would meet a CEO or some leader and say, let me send you a team. They'll come in, just tell them your hardest problem. They'll come in and help try to solve it. And you'd kind of have like 3 months, the canonical pilot length, to go and work super hard and fly around a lot and try to solve it. And what would wind up happening is you'd have to build new things in the field. And then like most of those things wouldn't work out. Like the pilot wouldn't convert, it wouldn't be valuable enough, whatever. But some of those things would go on to be multi-billion dollar business, you know, like just products and foundational parts of the platform. And the whole model of forward deployed engineering that really, really works. And the thing that people kind of miss is it's a radical deference to the field on what to build. Like most organizations, they're like, okay, we've got our roadmap and there's like all these slides.

We know.

"We know." "First is the customer." Centrally, and I think especially a lot of founders, like, I'm like this sometimes, like, "I know what to build. We gotta go build this thing. We just gotta go find the customers who want it." That's part of what you're selling the investors.

It's like, "I have the vision." "I have the vision." Right.

And the reality is like what really worked, at least in my time there, the way it worked was you just hired a lot of really smart people, kept the bar really high, gave them freakish, irresponsible amounts of ownership and autonomy. We were a bunch of like mid-20-somethings being put in charge of these like really high-stakes deployments. Budgets and it was just like—

In some ways, you know, McKinsey's good at that too, where it's like you speak for the firm, right? Or are you—

Kind of. I mean, the McKinsey model is much more pyramidal and like command and control.

It has to be approved at the top, but you get to speak. You set up a way that sort of your youngest people can sort of say the firm says blah, blah, blah.

I guess, but those people are parroting the firm's talking points, right? No one's hiring a 25-year-old. No one's hiring McKinsey to get a 25-year-old coming in and telling you what you should do about your business strategy. You're hiring McKinsey because they ostensibly have this centralized expertise and they've got a practice around your thing. You're a telco, you're hiring someone from their telco practice to come in and they've got best practices. And there's a lot of different ways to interpret what you're—

You're hiring because the partner has inside information that everybody else is slowing down, dresses it up in sort of process. And it's like, oh, don't— I know the answer we need to get to. Like, explain why we got to do this.

Well, there's a lot of different mental models of what consult— what job to be done consulting actually does. There's like the one that they'll advertise, which is, "We're the strategic advisors. We're here to provide our insight and business expertise to help our partners make better decisions." Like, okay, yeah. But like there's different versions of this. There's one that's like they work with kind of everyone and yeah, the teams don't talk to each other, but it all kind of gets distilled into best practices. And it's kind of like how you can go get, make sure if you're an executive, You're doing the right things. The other version of it, and the version I saw in my brief stint in management consulting before Palantir, was we were hired one time onto a project where we came to learn there was another consulting team from another firm working for another one of the VPs of the company on effectively the same problem. And these two VPs both had, they were like fighting about what to do. And they both hired consulting firms. And we were basically told, This is my theory from the VP. Like, I need you guys to build a deck, basically build a deck, like a slide deck to back that up. And then we came to learn there was another VP who had told their firm that, and it was like we were both like the proxy army for these like warring warlords, you know, like these factions. And it was very weird and it was kind of like, what are we doing here?

And like it says something about like intelligence. It's in the world of models, you know?

Yeah.

It's just like you're just accumulating your argument when really the argument is not about the reason.

So you can kind of draw— I actually thought about that more subsequently. And I'm like, there is something kind of pure about it, which is like, it's a war. It's a battle of ideas. And, you know, but I don't know. Anyway, the, the whole point is that Palantir, we were going in and, you know, we would get attached to these problems and then try to work backward to like, well, what's the product that could solve this? And sometimes it was kind of a hacky thing. Like we just threw together this dashboard that lets you see this joined dataset we have. So now you can correlate these two things. But what actually wound up happening in the mid-2010s is we had a lot of cases where we were coming into these commercial companies and they wanted some higher-order thing. They wanted like, "Oh, we want to see an analysis about this," or "We want a machine learning model that shows us the correlation between this," or "We want to see whatever." And then we would have to go build and the 4 deployed engineers would have to go build and kind of hack up a lot of infrastructure behind the scenes to be able to try to pull all that data together.

And then you turn that into a product.

Then we turn that into a product. So there's a team at a deployment in Europe that famously kind of started productizing this. They're like, well, we're writing these same shape of Spark job orchestration things over and over again. We should build a thing to help do that. And then that sort of started snowballing within that deployment and then spread. And one of the really cool things about the forward deployed model at Palantir that I saw was it was almost like a community-driven adoption. Like, you would see things go viral where you'd have one deployment go and invent something, and then another deployment would pick it up because it's on GitHub. And once we sort of had a core platform and it got— we got better at that, like, oh, like, your deployment built that thing, cool. Like, can I— you could just install it and it would work on your platform too, sick. And then, you know, it's open for PRs, so you can be contributing code back. And then these things would gain momentum, and someone who started a project working as an FDE on a thing, and they built this thing just to solve this customer. Maybe 6 months later, they're basically on— they're basically a dev team now that's kind of delaminated from the deployment. And now we're like the dev team.

You got too good at this. All right, now you're— yeah, you see this different thing.

You see this cycle. So there's a product some very close friends of mine built. They were over in the Middle East working with a customer there. Um, and they built a product, uh, to be able to do this sort of, um, drill-down data analysis to solve a problem there. And a very esoteric problem, uh, signals intelligence problem. And then that product was sweet. And like a lot of people, including my deployment, were like, that would be really useful. And so we kind of lobbied and got them to deploy it for us. And it was great. And then it spread and spread. And now it's like a super core part of the Foundry platform.

Does it have a name?

Uh, yeah, it's called Contour. Okay. Um, like every, like everything there, it went through multiple names. It's been named like 4, it was named 4 things. Um, and there's, there's so many examples like that and, and there's so many wonderful things about this. Like Contour is sick and it was never on anyone's roadmap. It's just like 3 guys went and invented it cuz they needed it and it caught on and spread and you know, was never on anyone. It was never on some central roadmap like in Q3 2016, we're gonna go build this thing. Now, the downside, the sort of invisible downside that's easy to ignore, is there was also a ton of shit we built that went nowhere. That was a complete waste of time. You had deployments where we were flying super expensive people first class every week somewhere, you know, just, just tons of time and opportunity cost and just piles of money.

Basically, like, customers were paying for it, right?

Customers were paying for it. Yeah. And they were kind of paying for the sort of service version of it in the short term. And there were customers I actually met someone recently who was a customer of ours in that era, and she kind of was like, "Yeah, we felt pretty burned by you guys." And I just think it's almost like this kind of necessary downside. I hate to be so glib about it, but like of having this high rate of innovation in the field. Like, you know, people say— People say fail fast, fail fast, move fast and break things. Like these ethos things in the Valley, it's really representative of what anyone who's really done this well really knows, which is The rate of iteration and learning is everything. And one of the big problems in enterprise software is a lot of enterprise software is built by big lumbering teams who are working from 10 to 3:45 in air-conditioned offices in the Valley with ample ping pong breaks and blah, blah, blah, blah, blah, working off of these long multi-quarter roadmaps. And they're very sparsely talking to customers. And what you're going to get from that is you're going to get consistency, you're going to get predictability, you're going to get all these things that companies value as they get bigger. You're not going to get that innovation.

Sometimes here, like, we didn't really figure it out. Other times it's like, this is so valuable, we're going to spread it to the rest of government.

Yeah, these structures wind up being almost designed to squash, like, high variance innovation. Like, what all process does, the whole point of a process is to reduce variance.

So you're an evangelist and defender for the forward deployed engineer, but Your own company doesn't do it. What is your heuristic for what company should use a forward deployed engineer and then why did you decide at Hex not to?

We have a lot of former FDEs at Hex. So what you see is a lot of startups now adopting the FDE model where they're like, we're, we're, we have forward deployed engineers. Well, I, and I, when I, when I critique this, it's not like to gatekeep the title, whatever, like It's really like, do you understand what worked? Why this works for Palantir? Because I think a lot of people are cargo culting. They're saying, oh, Palantir did this and look at their revenue multiple, look at their success. If we do forward deployed engineering, we'll be as successful. And they kind of go through the motions and then they're not getting the same results. And they're like, why? Well, you don't really understand the physics of it. And And for us, when we were starting the company, you know, we had obviously lived that, but we were like, we want to actually want to kind of go the opposite way. We want to build one consistent piece of software that we can deliver and scale. And, um, super out of fashion, right? Super out of fashion right now.

I know, but we're raising rounds. Sequoia is an investor, but somehow, but not on some of the narratives.

But, but, but no, but it's, it's, there's a synthesis here, right? Like, um, what we do is we just talk to customers. I'm in New York right now. I'm visiting customers. I'm bouncing all over town the next few days, just visiting customers, getting up close, and I'm doing my best not to just meet with them in a conference room, but like, I want to go by someone's desk. I want to be like, great, what are you up to? What are you working on? Can you show me Hex? Can you show me what you're working on? Like, go get up close. I'm bringing individual— I'm not bringing our sales— my salespeople on those visits. I'm bringing the engineers and designers. I'm like, I want to go see this stuff. And I try to get our whole team doing that and go get in front of customers, really understand what they're doing, really understand what they're asking for, really understand what they want and what they're using as an alternative, like go do a great job at that and talk to customers. And if we actually have a problem at Hex, it's like, I think some days it's like I worry our EPD team is our engineer product design teams are talking to customers too much. Like, it's like, guys, at some point we got to go write the code. You know, it's like we talk, we talk to customers a lot and some of it's just you're a smaller company and so you can't just have everybody spinning in different directions. But it's a cultural thing. It's like, I think it's really important. To have people really build empathy for customers. I mean, I don't— I think the worst engineering in the world is done via ticket. Like, if you wanted to have some tickets to, like, manage your work. Yeah.

I mean, it's a core Y Combinator, you know, principle of building companies is like, what did it build? Something people want. Know your customer. You know, that's—

Yeah. But a lot of most people don't do that.

They don't do it.

They said they had the roadmap.

But what startups do you think, or smaller companies, non-Palantir companies are doing forward deployed engineers in a smart way?

The only, I mean, the only one I know who's doing it the Palantir version, which is basically we don't really know what to build. We're just going to go embed with customers, get them to pay us some services thing to cover the time maybe, and have some exchange of value. And we'll just build the platform depending on what we learn is my friend Arjun at Distill, Distill with a Y. Um, he, he's the, I think there may be other companies now doing some version of this, but he's the one who I know is doing it the most pure.

Is he super charismatic?

I think he's very charismatic.

I'm not like insane, which is funny. You refer to Alex Karp earlier as Dr. Karp. He's like a philosophy PhD, right? Like, like, Alpapir uniquely has this sort of like, you know, we're like Wizard of Oz, like we're gonna do a magic thing. I'm like, can anyone else really sell that besides that company? You know, in the sort of mystical aspect.

He's just, he's Dr. Karp. It's like when you, it's like when you grow up and your neighbor is like Mr. Smith, you know, to you, and like later in life you run into them again, you still call him Mr. Smith.

It's just funny, you guys found your— I'm Jim.

I'm Jim now, son. It's all right.

He talks about it like, you know, it's a culture of, you know, people will tell him like he's totally wrong or whatever, but that's just a very deferential way to refer to him.

I don't know. Yeah, but it was— it is a very decent, you know, they're very open, very transparent, very best idea wins, not a lot of job titulature. I mean, I had the same job title all 5 years I was there. Had people reporting to me, I was doing IC stuff, I was doing big lead stuff. It didn't matter. Like, when you're in that environment, you kind of forget about job titles for a while. You forget about— there's different— you squeeze the balloon, the problems emerge elsewhere. Like, there's gonna be shadow hierarchies and stuff. Every cultural thing has trade-offs, but I thought the culture there really was designed— it's almost like it was very sharply designed around getting a very specific thing which is maximum pace of iteration and learning. And that meant there's all these negative tradeoffs. But culturally, I think we always like acknowledged them and accepted them. And there was a time I remember we had a period when I was there where a lot of people had left because they felt like there wasn't a growth path for them. And because we didn't have like a job title ladder to climb, like a lot of people are used to.

Was it hard to raise money without a clear title where people knew Where it's hard for me to raise money, or just, I would imagine going off and starting a company, you want to be able to say I was important here or something. Yeah, it's just sort of a vague—

it was actually interesting. I did another thing in between Palantir and Hex. I went to a really cool biotech company here. Yeah, it was called TrialSpark at the time, now Conformation. I remember interviewing and I was like, what do I, what do I say I do? Like, I'm a deployment strategist. What's that mean? Like, and I think people who were hiring from Palantir kind of knew, they were like, okay, yeah, we know kind of knew that that was a thing. Um, when I was raising money, it wasn't a big— I don't— that didn't really come up.

It was just the product, just sort of the vision.

It was just that we clearly had a sense of what we were doing and what we wanted to do.

But say what Hex does, just, uh, you know, we don't make this about the company, but just so people know where you're coming from.

So we build AI data analytics software. Like sometimes people call us Cursor for Data or, um, things like that. It's basically you can use natural language to go ask any questions.

I feel like you just listed a bunch of highly valued companies, like Cursor for Data, Anthropic for Data, OpenAI for Data.

Figma for data. It's like, yeah, because the first, the first arc of the company really was like, there's all these local fragmented data analysis things. We spend all this time bringing your data together and just putting dashboards on top. You know, we wanted to build this cloud-based front end for all of your data work. And, um, it was very similar to what Figma had done when we started the company. Figma was, you know, on its way north and was really taking off and sort of an inspiration of like, oh, you can take these local fragmented workflows, bring them to the cloud, make them collaborative, and in the process of doing that, bring more people in as participants. That was really the initial thesis of Hex. And then a couple of years into the journey, like GPT-3 came out and ChatGPT came out, and this was a thing we really got entranced by. So the second half of the company, or more than half now, the company has been figuring out how to really apply AI and now agents to work for data analysis, which is the holy grail. Like the holy grail, the reason people buy data tools ostensibly, invest billions and billions of dollars every year in data infrastructure is because you want to know things and you want to make better decisions. And we make that really easy and we do a very different set of things well than what Palantir did. We don't do like data integration and infrastructure stuff. We're very much like the app and collaboration layer.

Where do you sit relative to business intelligence?

We are— we think of ourselves as reinventing that. Like we are—

what people don't like that space or what's the— like you don't use that term.

I didn't— yeah, I didn't want to use that term because it had so much baggage. Like when people think for a long time, when people think BI, they're like, okay, well, I think of all the features that Tableau has built for 20 years. And I think about dashboards and I want you to do that. And we actually originally were very insistent on like, we actually think that the old way of thinking about BI is like super low value. We think dashboards is going to raise more questions than answers. Something I've said a lot. And like the most of the valuable data work is not happening in these dashboards. And instead, It's people are trying to do it in this melange of other tools, Jupyter Notebooks and SQL queries and spreadsheets and screenshots of a chart pasted into a PDF of a deck attached to an email.

You're more rooted in like the role, like you're hiring these smart data analysts and we want to sort of—

Originally, yeah, our initial sort of beachhead landing thing was like, we're going to come in and we're going to revolutionize your way your data team works.

Right.

Again, very similar to Figma coming in and we want to revolutionize that.

Now you're like, you don't need to hire them or— No, I do every AI I think every data embracing company has this trade-off between supporting the role versus saying, oh, we're helping make the role obsolete.

I think we're helping to reinvent and evolve it. And we have a great data team at Hex. It's almost like over-provisioned in the sense of we just have really wonderfully talented people on it, but also hired people that have been thought leaders and writers in our space for a while who are amazing collaborators for us to build the product, but also collaborators just thinking about like What is, what are these roles becoming? And it's interesting to see our own employees kind of disrupting their own job a little bit. And it's probably similar to what it feels like to be an engineer at Cursor or something like you're trying to reinvent the way your job works. And so, you know, we're, we see ourselves as fellow travelers with that data audience of like how we're helping reinvent the way their job works and what that looks like and what it looks like to fulfill the mission they've all always wanted for their organizations, which is you want data to be useful and help solve problems. And now with AI, the fastest growing part of our user base, of course, is all these other people in these companies that are now able to use data in a way they weren't able to before. And in that way, we are fulfilling what BI always wanted and purported to be and what people thought they were buying with their billions of dollars of BI spend, which was, oh, we want to make everyone data-driven. Well, you got a lot of dashboards. Are you feeling data-driven? Are you able— is everyone able to ask their questions?

Because if you can't interrogate the dashboard, you don't know its limitations.

Hey, they're really good.

Look, dashboards are fine.

Big discourse now, dashboards are dead. Everything has to be dead or broken or whatever. Dashboards are dead. I'm like, dashboards are fine. They're fine. But they've always been fine.

To the extent they're dead, it's because people want to ask a question directly of data in the moment.

Yes, dashboards, same as when we started the company. It's like, dashboards are great for seeing what happened. This KPI, blah, blah, blah, whatever. It's a collection of these charts. Great. Fine. But often what you look at is you look at that chart and you go, okay, well, I want to ask a next question. Why did that happen? I want to drill down on this. And are there other times something like that has happened? That's what we built Hex to really excel at. And now it— like the architecture and a lot of the things we built, we kind of feel like, man, right place, right time. Because the way we even structured the product is really conducive to these agent workflows, which is which are about depth a lot of times. The magic of these things is you can ask a first question, get an answer, and then you can ask follow-ups and it can go now and drill in and have multiple different tool calls and paths it's exploring. All these things we built even going back to the very beginning that we didn't know were going to be useful because we have magic robots that can talk to us now, it's all very useful now. Now useful for way more people, which is really, really cool.

Are you using Claude or what is the intelligence sort of behind this?

We use Claude and the GPT models and it's very interesting. We relationships with both companies. We spend time with their research teams.

Who's doing better right now?

They're both doing great. I think the GPT-5.4 models, which are the latest ones to come out, are really fantastic. And we have a quote, my co-founder Caitlin, who I think you've met, has a quote in their launch blog post summing up our findings. But they're really, really good and getting better. I think what we're seeing is these models, the main focus of the labs right now appear to be hill climbing the models on coding tasks, which is you can squint your eyes and look at Hacks and be like, "Oh, it's like coding. It's like you guys write SQL and Python and data viz." And the improvements on those models at coding tasks does make our performance for our domain better incrementally. But surprisingly, actually, we even saw with, I won't say which one, but one of the latest new model releases from one of the labs, it was actually a regression. For the evals we have. We have a really extensive set of really rigorous data analytics evals that are not the naive ones like, "How many widgets did we sell last quarter?" But really brutal, the real-world shit that people ask of data. It's interesting to see the models sometimes even regress because clearly they've been trying to hill climb on something else. One of the big problems is in data analysis, it's non-verifiable and there's actually different ways to answer a question.

Coding has this sort of code worked, it built this thing.

You can sort of—

Yeah. Whereas analysis, it's a little squishier.

And so this is something we work on. This is why we spend time at the labs, is trying to help figure out how can we make the models better for this? Should we be RLing our own models? I don't know. There's a lot of even conflicting opinions internally.

Are you spending money on your own models?

We do. We run some of our own inference for specialized tasks. We are using the GPT and Claude state-of-the-art models for our main agent backbone, but we see some interesting opportunities and it's not impossible that Anthropic and OpenAI even allow— they used to in the older era had a lot more fine-tuning tools and I think they may come back in some places too where they allow you to do these RL regimes on top of their base.

And where are you on agent mania? Are you going to be like agents as representative as humans in your product, you're going to be selling seats to agents, or you think that's sort of out of control?

Or I think it's a— I think it's probably overshot a little bit. Um, I mean, we've talked, I, you know, with the pricing thing is interesting. Um, you're selling seats. We sell seats.

Another not cool thing.

Yeah, we sell seats.

You're supposed to do usage pricing right now. What's going on?

We are, we, we're adding, you know, usage on top because one thing that's crazy is our weekly actives chart and our messages per day per week chart is straight vertical now. It's gone fully parabolic. The product is really working. The agents are really good. People are loving them and using them. Yay.

Though they're costing you more money.

The unfortunate part is our cost chart looks the same. And if you're a company today that has working agents, these things are expensive to run and you need to charge for them. So we're going to do it with our seats have a base amount of tokens and then, or credits, and then you can get above. It's pretty standard now for a lot of tools. It's like Lovable Replit, et cetera.

But why seats at all?

It's a good question. I think in some ways it allows a base of predictability for customers. We are a product that wants to get really wide in your organization. And if we can say, yep, it's however many, $50 or whatever for a seat for a month, and that comes with this number of credits, and the average user is going to, their usage is going to be covered by that, that's really nice for predictability. I mean, that is a big thing. People who have never sold or bought at-scale enterprise software knows, having it be fully variable is tough for a lot of companies. And generally, I do have these funny conversations sometimes about seats where even with smart VCs, which is funny, that are like, well, there's going to be pressure on the number of seats you're going to be able to sell because customers are hiring less people or whatever. And I'm like, we've grown a lot and we're doing really well, but we're still, the amount of revenue we make today is still very, very small to of, you know, the incumbents and where we want to go and all the TAM we think we can capture. And I'm like, if there is not tens of millions, hundreds of millions of dollars worth of seats for us to sell because everyone's been laid off, the world's like ending. Like, I need to own a lot of guns. The least of my problems is my fucking SaaS business. Like, there's going to be riots in the street. And so I think it's kind of funny, this conversation about this. Like, yes, on the margin, Is it possible that some of our customers will just be small or in absolute terms, the number of people? Yes. Are there still a lot of seats to sell for our product like ours? Yes. Do a lot of customers still want to buy seats because it makes sense for what they're doing? Yes. Can I layer a consumption model on top that helps me both cover costs and also capture upside from real power users who are getting excess value? Yes. And over time we may tweak those dials. I don't know, a year from now, maybe we turn those seats down to zero. Maybe we Maybe it goes back. Elena Verna at Lovable has this blog post that we talk about internally where she's like, I think the token cost is going to come down significantly and we're all going to be back to selling seats. Hmm. Who knows? So we like, we like having the options and, you know, we want to ultimately our job is to actually meet our customers where they're at. And I, people always joke customers don't want to pay anything. Well, I actually disagree. I think customers understand the value and want to pay for something that's going to provide good value for them. They want to support you. They want to be supported. I'm a buyer of a lot of software at Hex now, and I'm very happy to pay for stuff that works well. And if I feel like there's price to value and if I see the bill for what we're paying for something and I'm like, yeah, but we love it. I'm like, great, cool.

Where are you on this? Yeah, white collar work is about to get decimated in our country's economy. I don't know, going for the guns. Like, I literally was talking to a, you know, very respected person, you know, who was like, oh man, we're going to enter like an insane political moment. Like every, you know, Democratic cities, you know, if you know all the Goldman analysts or something are about to get fired, you know, then you're going to have this sort of weird class, you know, just like you have pockets of people are very resentful about the new economy. Like, I don't know how you see it.

I mean, downward mobility, downwardly mobile elites is right.

The most dangerous force in America.

Well, I mean, forget about America. Just study history, right? Like, that is a bad thing. Like, that, that is— that will lead to problems. And I, I don't know for sure what the future looks like. I would say I applaud, uh, Dr. Karp. I'll still say Dr. Karp, who I think has been a very bold voice in saying, like, we should all be thinking about how we're using AI to, like, increase employment and productivity. And, um, I, I agree with that. Now, I'm building a tool that you can kind of look at and be like, ooh, do I need as many data scientists or data analysts as I have? Sure. I also see a lot of the data scientists and data analysts we have using the shit out of it, doing a ton, being way more productive, being more valuable for their companies, and often now starting to work at these higher levels of strategic insight and maybe their roles are starting to evolve. So it's not at all clear to me. We build an AI product. We use a lot of AI tools internally. I'm trying to hire as fast as I can. So we'll see. The other thing I think about is like, where is the productivity? I mean, I can look within our own customers and see them doing more and all of this. I think a really interesting question is we're a year and a half now into coding agents being mainstream and we're billions of dollars of revenue into tokens paid to Claude and Codex and Cursor and other things that start with C. And where's the explosion in software? We're getting a lot of vibe-coded stuff. We're getting a lot of sort of slopware. But if you look at the cost, the software companies you buy from, it's like, where's the, where's the like cornucopia? Like, why am I not getting new features every day?

Right?

It's so— we've built a bottleneck lives elsewhere.

I don't know, my business lead has built us, you know, an internal like ticketing tool that he vibe-coded. We're, we're about to see, you know, when the metal meets the pavement or whatever, the, uh, whether, you know, we can accept payments and everything safely through it. But, um, but like why Like you did that, but I get that.

So you can't—

you can't be perfect.

Is that Riley? Yeah, yeah, that Riley can do that is cool.

It's huge.

I mean, it's pretty— it's pretty— but, but like, should he be doing that?

Yeah, there's a lot of things to do, but, uh, the, uh, I think it'll be good. You know, we'll keep track of everyone. Um, but to your point, I mean, do you feel pressure as the— at a startup, your investors want to hear this story of like, yeah, we're moving so much faster because of You know, Cursor or whatever, and you're saying, oh, but it's not like—

I do feel like we are. It is also interesting. We are living, you know, I've got— I've just came from our office. I've walked by any engineer's desk and they've got Cursor or Cloud Code or Codex now. They've gotten very popular up on their screen. They're using it. I think we're still figuring out how to use these things. Like we're setting up cloud dev environments to make it easier to have multiple different agents that can work in their own dev boxes. And it's like, okay, so now our engineers are just bouncing between 5 different agents working on different tasks at a time. Who's reviewing all this code? Who's initiating the thought on what we should build? I mean, I'm almost like really excited about the idea of just getting this torrent of PRs from all of our engineers. I'm also slightly worried because we're a product that's prided itself on design and consistency and fit and finish and feel. And our designers are busy now, like trying to do polish commits onto all these PRs and it's How do we even get all this out into the product in a way that our customers understand? It's gonna be changing fast, and that's exciting but also stressful. It's like, I think we're all still figuring out how to metabolize all of this. Something tells me—

there's just like a pace of— that humans can absorb.

I think it's literally like if you eat too much food, like your body can only metabolize so much of it, right? And I think we're still just figuring this out. And so I don't know that these things are going to happen overnight. And if your model of a Goldman analyst is that they're just doing this really rote, repeatable work that an agent can replace, sure. Is that all that a Goldman analyst does? It's the same, I think, for data people. If you're like, how do we understand those jobs? What are actually the bottlenecks in companies doing certain things? I don't know. We're still early in figuring this out. Empirically, it's interesting to me. I've observed almost coding agents have disrupted product management faster than engineering because the engineers are almost becoming PMs in a funny way where it's kind of unintuitive, but a lot of our engineers are now— their job is less like, okay, I take the product spec that was written in English and I translate it to code syntax. Like, that used to be the thing. It's like, you're an engineering candidate. Oh, you know how to take English and turn it into TypeScript? Great. Welcome. Right? That was kind of like at least the first order heuristic of what we're interviewing you for, like coding interviews. Well, now there's magic robots that can do that. And so instead—

So we're going to keep our smartest people, the engineers, and shift them to sort of the strategy rather than—

Yes.

I mean, that's not like a product manager. That's not I would never be an engineer. But I mean, that is sort of— It's like, okay, if they're not writing the code, they're still very smart, understand sort of the core product, and so they're gonna become the product managers.

Going back, by the way, the type of engineers we try to hire are those that like talking to customers and like thinking about what to build and enjoy our product space. It's a cool product space. So yeah, those people have always liked that, and now they just have kind of more time and brain space for that, and they can be thinking about what to build and how should this work and the architectures and all this, and it's really cool to see. So, um, We'll see how these roles change. What designers do— our designers have coded for a long time, but now they're coding a ton. They're mostly coding. It's like, well, are you— what do you do? Are you just an engineer that has really good taste and more tattoos? It's very interesting to see happen. Internally, I'm just trying to lead everyone through it, saying we're all learning this together. And if you don't disrupt your job, someone else will. And don't worry. How many people do we need to get to a huge exit at Hex? I don't know. More than we have today. So—

and more than our competitors. So we want to get good ones.

You guys are all safe. Let's all figure out how we can all use our time really well and then we'll, we'll go build this thing and have it be really valuable.

So I wanted to go back to Palantir and some of the— I mean, you know, obviously we just had this— Anthropic took the headline because they were the ones sort of fighting with the government over— yes, fighting with the Department of War or Defense, depending where you sit on the— sit on things.

I'm not going to take a position on that.

I think we— I've gone back to defense. I mean, defense was established by Congress, so it is sort of a branding exercise. You don't need to weigh in on this. But I think core to the conflict, the, you know, Anthropic had two concerns, right? One was this autonomous targeting and the other was this piece that we are going to do domestic surveillance. And I think embedded in the domestic surveillance is that fundamentally because of how much better their technology and technology generally has gotten, the government can just acquire publicly available data and then come up with this NSA dragnet that wouldn't have been possible, you know, when Congress was writing laws about domestic surveillance. And obviously Palantir, like, fits— fits into this. So what is your view of, like, what's possible in terms of exhibit A in a tech company that Lives in that gray area.

I was there 10 years ago after the 2016 election, and it had been not so long before that that we had like an, it was like an all-hands or some sync version we had of this that was like doing a win. Like the deployment team lead came up on stage to talk about our work with ICE, which we were really proud of because we were working with this team called HSI. It's like the really good guy, whatever else you feel about other parts of the government, stopping human trafficking and like drug smuggling, like go, go HSI. Yeah. And then like, you know, that election, the vibe shifted and all of tech shifted. 10 years ago is kind of a long time to remember back, but like, and all of a sudden this became really tough and we had internal concerns. We had protests outside the office. We had applicants not wanting to come, people not, people not applying or pulling out of interview processes because we were evil. And there was a big clamor internally to get Dr. Karp, Alex Karp, to say something publicly, say something, engage on this. And I was among the frustrated on that. I'll confess. Like, you know, I'm interviewing candidates who are like half the interview, they're just asking me questions about how I can sleep at night or something. I'm like, ugh.

It's also so secretive that it's hard to even necessarily know even inside the company.

And inside we knew. And I think we had an appreciation, and I still have a deep appreciation for, like, this was a company that recognized that to serve the government, to serve the government in these spaces, you have to be comfortable in the gray areas. And that doesn't mean moral compromise. It actually means moral clarity and having principles even when they're hard. And, you know, in retrospect, I think he was totally right. And they're much more public engaging in the forum now. But I think at the time, the right thing for us was to just stay quiet not engage, not inflame this and say, let's focus on our work and let's focus on doing the best we can and also embrace the fact that we live in a democracy.

But do you have a view on this Anthropic blowup specifically?

Well, yeah, that's my— yeah, to kind of segue, I love Anthropic. We're partners with them. They're customers of ours. I say this with love, but I do think It's like the heart and the head. Like, from a heart, you understand some of these concerns. Like, would I like— would most Americans vote for doing mass surveillance? You know, you can go gather up all this information and figure out where everyone's loyalties lie or something like— no, I think most Americans would not want that. But that's why we have a democracy, and you can vote, people can vote for that. And I think what we came to appreciate at Palantir, sort of in retrospect, it's like, well, okay, we're going to have red lines in what we're willing to do. Does the government have to call us every time they want to use Gotham, which is our product, every time there was a new Palantir to use this thing?

Like, isn't this an extreme deference to the executive branch specifically? I mean, it's not like Congress has passed a law, said, oh, we love Palantir. Like, you could have a belief that, like, our product is so significant.

They have—

I mean, they sort of yielded power, but they haven't, like, expressly Yeah, this is great.

I mean, I'm not like a constitutional scholar.

I'm just saying, but I do think, and I don't want to, you know, you don't, you're not, you're just the one here talking about it now, but like there, there, I do think there's this argument that it's like, oh, we're deferring to democracy. It's like you're deferring to like one branch that's taking on a lot of power and saying we represent democracy.

This is what it looks like to live in a republic. It's messy. You, you, you elect, right?

So it requires on the part of other actors to sort of interpret what the will of democracy is beyond just like deference to the executive branch.

That's a republic. You described a republic.

But the argument that Palantir is making and trying to suggest Anthropic embrace is sort of like yield to just the executive's interpretation of what the will of democracy is.

Yeah. Yes. Yes. And that will of that executive is bound that's Article 2 is bound by Article 1, the legislative branch and the laws they pass. And look, I'm not going to get into like debates on like current Supreme Court issues or all that, but like at least if you read the Constitution, it's pretty damn clear. Like the, the legislature passed— the legislature passes laws, including the power of the purse. They control what money is spent and the executive executes on that. And there's all these, you know, unitary executive and all these things. I'm not going to get into all that, but I will just tell you, even you see this play out now. There is a funding bill issue in Congress right now for Department of Homeland Security where there appears to be a legislative majority or a controlling legislative majority that is not comfortable with funding Homeland Security at the way that they have been.

Right.

And that is the democratic process working out. And you can look at the last— this current Congress and say, well, they've been overly deferential to the executive and whatever. That may well be, but you see this happening. And now people are looking forward to the midterms and people are worried about that and they're going to react to the public sentiment. And if the public doesn't want something, great. What a wonderful opportunity for a political entrepreneur in Congress to get up and propose a bill banning this type of mass surveillance. Just outlaw it. You can pass a law to do that. That's a— that is a thing. That's your job, actually. That's literally their job.

If this is a thing we don't want to have, we both want Congress to take a more active role in saying what they support and what they don't. And I agree that, I mean, ICE and Homeland Security is a good example where Congress is stalling on it. I mean, we also live in this liberal democracy where unless it is a— I understand in a war people get drafted and summoned into government service, but in normal times, I think the belief in America is that you don't have to do something for the government if you don't want. You know, you're not like—

Right, right.

Besides sort of making the whole thing work with taxes, you don't have You don't have to, like, go work for the government.

You don't have to, and Anthropic doesn't have to. No one does.

But the criticism of Anthropic is they sort of need to be summoned into it.

Maybe, but they could just— an option they have is say, we actually don't want to work with you guys. We're going to withdraw from the contract. Best of luck using other AI providers. Like, that's allowed. They're not— there's not the— But the Trump administration is— There's a War Powers Act. There's a War Powers Act. Where the government can mandate, like, I think it's called war powers.

I think it's like we saw some of this during COVID Like, that's one of the ones they floated, but they ended up saying, oh man, it's a supply chain risk. It's more dangerous, apparently.

But that's— but that's them. That's almost doing the opposite. They're almost cutting them out. I mean, I'm just saying— I'm just saying that like an option they have is just to not do it. Now, going back at Palantir, one of the things we talked about that I do believe to be true is if we're not engaging on this issue, other companies with less scruples, with less principled point of view, and by the way, less effective software, are going to. And we have a privacy and civil liberties team at Palantir that I loved. I thought they're— a lot of them are still there. They're incredible people, very thoughtful, that we worked with very closely on a lot of these things that— and, you know, you kind of look down the street at the other defense contractors and you go, they don't have that shit. Or they do. There are a lot of lawyers who— but it's, it's, it's Like, would we rather be the ones engaging on this and thinking critically about it and trying to find the synthesis of how you have robust defense with civil liberties, or would you rather cede the field? And for us, the calculus was we would rather be here, even if it's a little uncomfortable, because that's our role in this pluralistic, messy democracy we have.

I mean, I agree that there was sort of a misguided time where the tech industry said we shouldn't do— you know, Google didn't want to work with the government, or, you know, there was like, we're not going to work with the Department of Defense. We've moved to a point where it's patriotic to support the government.

Yeah, but I think there's nothing like multiple wars breaking out to kind of get Americans to, like, wake up and be like, oh, maybe we need, maybe we need good capabilities and weapons. Like, Anduril was like super unsexy, off-spec, not a sexy VC thing. And then all of a sudden a war breaks out in Ukraine. It's becoming like a drone war and everyone's like, Anduril is amazing. But I've always thought Anduril is amazing. I know all the founders and, you know, But it was funny to kind of see them going from these, like, black sheep to now these darlings on the other side of a war that kind of reminds you it's not the end of history. Like, the Ukraine war is a reminder that, like, no, this is— we live in a dangerous world.

But the argument that you're making that Palantir, you know, we're so— it's so thoughtful about, you know, sort of civil rights or whatever, or like understanding the effect of its technology. The argument right now is that Tech companies shouldn't be the ones deciding, and the government should have total, total control. Um, like, that's the line from what I said. Palantir seems to sort of say, well, we're not creating contracts that limit you.

Um, but they can just choose not to engage in a contract. There were— there were— there was work. I can't talk about it. There were things at Palantir that we just decided not to do. Countries we decided not to work with. Lucrative, lucrative deals. Where we sent teams out to go diligence something and came back and said, this is, this is not an area we are not comfortable here. We don't think we can actually have our software to be deployed in a way that we think is going to preserve the things we want. We're just not going to engage. I mean, that's an option.

But Anthropic, you know, had a contract they liked and then the Defense Department wanted to change it.

I mean, I've read, I've read their stuff. I read Emil Michael's interview with, um, Empire Wars. Empire Wars. I've, you know, and I don't want to— I'm— the last thing I want to do is get in any of that. I'll just say that.

You said you want to be spicy. It's just of the moment.

I'm not trying to get nuked.

Are you neutral on it, or are you— what is your—

you're supportive of the department? I was raised at Palantir, and part of the reason I was there and part of the reason I still am supportive of is I think we— the West needs the best capabilities. One of the things that defines Western society and the Western alliance is democracies and these republics we have. The United States, this wonderful republic we have, that in every generation has felt broken in some way. I always think it's funny when people say, "We've never been more divided." Bitch, we fought a civil war. This is just what this looks like. No one's ever going to be totally happy with the current state of it. That is kind of just how these things work. I don't say that to be like fatalistic or even like, oh, well, just whatever, government, use the software however you want. I just think it's like the voters take— there's a saying I like, which is people deserve to get what they voted for. Good and hard. And you know what? They're getting it good and hard right now.

I just think we're still a liberal society where people participate or don't. And like, we are— Anthropic is being bullied into participating. Like the government is using the tools and you control whether you're supportive of them or not. You know, it's my issue is not— yes, the Defense Department of War is in charge of making war and Anthropic is in charge of using their technology how they want. And my, my issue is them using every lever they can to coerce Anthropic to acting against their moral beliefs.

I mean, this administration has, has not been shy about employing coercive techniques. Look at law firms. I mean, there's a lot of examples.

And a lot of you want to extract away the current administration. Like whenever we're having this argument, they're like, oh, the executive, whoever they may be.

And it's like, it's executive with a track record. People had critiques of the last administration and the way that they were acted coercively toward private enterprises in some sectors as well.

This is by not inviting Tesla to—

like, there's other examples. People have stuff around crypto. I mean, I'm just— I'm just saying.

I'm just—

yeah, I'm just saying, like, you can point these This is not a novel new criticism of government. You are right. We live in a liberal democracy. One way you could look at this is Anthropic is getting bullied. Another, and something I will say I admire, is Anthropic is saying we're patriotic Americans. We want our government to have a great capability. We are proud of the product we make. We want it to be employed and we just have some concerns and limitations on where it can be employed. That is a perfectly valid place to be. And the government, if you take them at face value, is saying, we want to work with private enterprise, but we need to have the flexibility to go engage in the things we have to engage in to do the things we want to do. Emil Michael in that post was like, we do a lot of stuff with guns. If you're not comfortable with— I'm paraphrasing— if you're not comfortable with that, maybe you can't work with us. And I do think that's one of the things you kind of have to stare into. Again, going back to the time at Palantir, there's things we did and worked on that you kind of look at and you're like, ugh. Not because it's morally repugnant, but it's like, I would certainly never stand up and celebrate people dying. Even if they're our adversaries, there's something— people kind of look at that and it's like the reluctant warrior, right? It's like to engage in these areas is to have to engage on the the reality of this, the gray areas, the grisly, on-the-ground way that the sausage is kind of made. And I think you can try to abstract that away, but it's—

I just think Anthropy picked this fight at a very savvy time. I mean, we just went to war with Iran. We bombed a school. It was all in Claude's plan. They have no plan.

It was all in Claude's plan.

They were about to be wrapped into this whole thing. Like, I don't know. I just think that it's almost as if Claude planned it. Yeah, these people we're deferring to are— anyway, I don't know.

I'm not trying to get overly involved. All I'll say is I have a lot of optimism and hope this gets worked out, and I have a lot of optimism and hope for how AI can just improve the way our government works and improve the way our military works when we need to unsheathe it. And I think that It is our job as citizens to be engaged, to vote, to lobby our representatives, to make sure that if there are limitations we want to add around that, that those are done. And I think we're going to see that play out in the Democratic cycles ahead, as we always have. That's how this process works, and we will watch it work. Yeah, that's my take.

The, uh, we could— I want to move on to another topic. I mean, our last topic before I let you go. Just raising money from venture capitalists. I feel like I have a lot of VCs on, on the show. You know, they tell their story about how things work. As a founder, I'm curious, like, what you really look for in VCs or who you think is doing it well right now.

Um, one thing I try to do is when I talk to— when I do like rounds of VC catch-ups, either current investors or prospective ones, I try to have the same conversation each time, like pick a topic. It's very interesting kind of doing the side by side and not like it's not one dimensional, like one is better or worse. It's just like a lot of different dimensions. And one of the things I feel like I've come to really look for is like first principles understanding of things. So like we were talking about pricing. You talk about pricing, like the kind of like thin VC version is like kind of regurgitating a blog post and sort of, well, these other people are doing this and these guys are doing this and here's what margin they have and all this. I'm always trying to figure out like the first principles version of it, of like if we really strip it back and understand the technology and understand our customers, what do they want? And you can kind of tell when a VC has been like really involved and very thoughtful with one of their portfolio companies where they can say, they're like, oh, well, at this company, the way we looked at it is this, and what we found is this, and we have this one customer told us this, and that made us think about this. They can really unpack it and it's very different from the blog post version synthesis. So that's something I feel like I really, really look for. Someone who's thinking about things clearly and with fresh eyes. There's a criticism of VCs like they're herd animals or whatever and there is a lot of that. But I'm always interested when I talk to someone who's got a really novel insight or opinion or contrarian take, because at least I know I'm getting like fresh thinking. It is.

I feel like I used to roll my eyes at first principles thinking. I think now that I have a business, I mean, you know, media is its own sort of animal, but there are things where it's like if you apply generic principles of like well-run media companies, some are lessons I should definitely learn and some you'd say, oh, you really don't understand some of the core things. That might make my business work. And you really do need somebody who has like empathy for like what is working about your business and what is sort of like the core. Yeah, sort of non-negotiable. And if you have someone just sort of sloppily applying like, totally just like heuristics without understanding like what's really working, it can— it could send you on a total detour.

I do a lot of angel investing. I love it. It's like a great— I'm busy with my day job, but it's actually like a great kind of like mental stimulus. It makes me think more about Hex. And I try to be really careful. Like, if you're a founder coming to me asking for advice, you have to sit through my 5-minute disclaimer.

This is why you shouldn't trust anyone else. You are in it. You're not one.

This is, you know, like, I'm probably wrong. Take it with a grain of salt because I just— I like— I— and what I try to do as best I can is tell people, I will tell you what we did and why I think it worked or didn't work. I will try to unpack from first principles the facts on the ground, what we've seen, the data to back it up. and then you can draw a conclusion on whether those ground facts, how well those map to you. Because that, I think, is the thing you're trying to do. And I find this experience myself. I'll go talk to founders and get advice. Talk to 3 really smart founders and get 5 different opinions. And it's like, well, they're all really smart. Dylan at Figma said this, and Kezar at Applied Intuition said this. And I'm like, oh God, they're opposite things. What do you do? And it's like, well, Did you really understand the grounding thing and the path dependency? And the other part of this job is it's a lot of luck and it's a lot of circumstance. You ask, "How did you find that great executive?" "Oh, well, I did the search like this." But also, someone happened to be available on the market and it's always the survivorship bias thing too of Well, this worked for this person, therefore it'll work for me. Or the failureship bias thing that you also hear from a lot of VCs, which is, well, this other company tried to do that and it didn't work, therefore it won't work for you. And it's like, well, what was the first principle of why it did or didn't work? And can you really understand that? That's a thing I'm always trying to grasp for as best I can and usually failing, but sometimes getting it right.

Barry, thank you for coming on the show.

Thank you. This was fun. Great.

That's our episode, and thanks for sticking around to the end. Please, if you've made it this far, you've gotta like, comment, subscribe. Excited to grow the channel, and of course, you can find our writing and reporting on startups and venture capital at newcomer.co. We also host events. You can check out what we're doing at newcomer.events. Share other podcasts. Help us get distribution. Support the channel. Comment, support, like, comment, subscribe. You know, you know the deal.

Thanks, thanks for being on the journey with us.

All right, see you next week.