How Model Performance is Plateauing

heard. Yeah, I know. If Pat’s listening, he definitely needs to give some credit to Sarah here. Yeah, Sarah actually was so our our first investor was OpenAI, and then our first two angel investors were Sarah Guo and Elad Gil.

Yeah, so I played sports when I was in high school and then didn’t as much when I was in college. You know, when you start a startup, things get pretty stressful. And I had a mentor who gave me advice of basically like, hey, stop lifting so many weights and start trying to run a mile. And I remember when I started running a mile, I think I was at like eight minutes or something. It was really, really bad. I was like pretty out of shape. And I had basically a goal to get up every single morning and just reduce my mile time as fast as possible.

And the way that I did it is I’m gonna run one mile no matter what and then just kind of see if I can reduce the back end of the mile again down, down, down, down, down until I can get to as fast as I possibly can. And the outcome of that, which I think really helped was, and something I’m actually trying to do more and more in my life, is every morning when I wake up, I get up pretty early and I just try to destroy myself and run as fast as I possibly can.

It just reduces my stress for the rest of the day. And I’ve found that over time, a lot of company building is just making very good decisions. And if you start your day off with basically something that is very challenging in a physical way, you kind of have this stress relief through the rest of your day. Your body has absorbed that stress, and I very much believe that in everything else too. I try to do a stressful thing every week because I think a lot of it is stress tolerance over time.

Yeah, yeah, What would yours be? I think it’s getting up early. We’re in 60 countries now, and so no matter what, I try to keep basically East Coast time. So when I’m in San Francisco, I’ll get up at like 4AM or 04:30AM, and what that allows you to do is you can kind of like focus before the stream of Slacks and the stream of emails come in, and I think that like couple hours in the morning, especially when I can go to the gym and I can kind of think about like product and those things, that’s like changed the trajectory of how I operate the company more than anything else, and I do it when I travel too.

One of the ones I have that I think was a good habit in the beginning of the company is now, and now it’s starting to get bad, is I zero out Slack like every like fifteen minutes, and so I’m in almost every single Slack channel and I read every single thing. It was really good in the beginning of the company because if you do that constantly, you’re basically every day all you have to do is catch up compared to what happened yesterday. And so it’s really easy to make decisions.

The problem with that is I’ve probably done that for too long. As you scale, you actually have to focus more and more on what is the p zero. And I think I’ve done a little bit too much of like still being in every single Slack channel and checking every little thing. There’s a time I was talking to someone from Sequoia and they had been living with other founders during COVID. And said that what they and these are very two very famous founders. He was basically saying that a lot of their routine was they would just jump into random meetings at the company.

And what they were doing is they were basically checking just to see how does this department do this sales call? How does this part of product create their PRDs and analyze their PRDs? How does this part of the company come up with OKRs or metrics, right? And I remember I left that dinner and my co founder and I were like, Wow, those guys aren’t working. That’s so lazy and all this stuff, right? And now when I think about those guys, I say, Wow, they are incredible.

I get why they’re some of the best founders on Earth. Because they’ve created a machine where they have so many folks at the company that are doing a really good job that they spend the majority of their time actually moving the machine. And you go back to being, I can focus entirely on product. I can focus entirely on what the most important things at the company are. And, like, that’s really what I’m trying to do this year is transition from kind of all those heroics to can you build a really well run machine.

A 100%. And I think it’s also a testament to something else, which is these companies are growing so much faster than they used to. It’s very important to go through a couple stages, basically. Like, I see stage one as product market fit. Stage two is company market fit. In other words, have you created the structures of your company that are the same as traditional, you know, B2B SaaS or whatever, consumer SaaS, whatever you’re doing, and what’s different. And there are differences. And any VC who says that there are no differences at this point hopefully has changed their mind.

There are differences. And it’s different based off of what vertical and which company you’re trying to build. So you have your product market fit, then you have company market fit, and then do you know what you wanna go right back to? Reinventing product market fit again. Right? And so it feels like there’s a cycle of doing that. And I think for us and like me personally, some of our first like couple years was product market fit. Last year was company market fit. I’m back to product market fit again, where it’s like what I spend a lot of my time on is like what is the direction of our company and our product for the next six months, year, etcetera.

And now you can start thinking like that.

Yeah, yeah, So we It was funny. I think it was our off-site in 2024. And I remember we kicked off 2025 and the first thing I did is go up and I said, Hey, we had a good year, but I’m pretty sure Anthropic’s at like 3,000,000,000 in revenue right now. Right? My point is the entire market is massive right now. And I don’t mean just like legal AI market. The AI market is exploding. And so I think you have to, when you are a leader of a company like this, one of the biggest jobs that you have is to make sure that your team doesn’t feel like they’ve already won.

Because the reality is the market pull is massive. And so sometimes your success isn’t just your execution, it’s the market pull. And so you have to benchmark yourself against other folks in the market and not just legal AI companies, but actually just AI adoption. Right? I also think that we’re just on an insanely compressed timeline. Like, think the winners and losers are gonna be decided in the next couple of years in a lot of these spaces. And so you really have to at all times make sure that the company doesn’t go, Wow, I did a really good job, chest bump, head pat, like, we’re done, right?

You have to instill this, Yeah, well, Anthropic, you know, 10x, and they started at x, y, z billions of revenue this year.

Our goals are much higher than that, and I

uncomfortable high. Yeah. I I think maybe the, like, Series C felt very high. What was that? It was 1.5 is what we were valued at, and our our revenue was definitely lower. Lower. I think maybe that was one of the ones that that felt super high. The other ones haven’t. Another thing too that I’ve I’m very picky with investors. So I don’t spend tons of time fundraising. What I actually do is there’s a couple investors that I’ve become decently close with, and when it’s time to fundraise, usually we get preempted or I say, hey, I think it’s like time to fundraise.

And I reach out to literally like one or two people. Yeah. And so I plan my fundraises pretty far in advance. So I’ll basically be like, like, I know who I want to lead the next fundraise. My point here is we, almost every single time, we probably could have gotten much higher valuations than we took. And instead, we chose the best investors and the ones that I trusted and personally wanted to work with.

I think the main thing about fundraising is you should always think of it as start it six months ahead of time and you will do much less work than if you actually go out and do the process. So what I mean by this is, and I learned this from some other founders who are really good at this, is what they do is they try to basically get folks to come in for a couple million dollars, like just like one or two, and you give them information rights.

So you say, Here, like, we’re gonna tell you how we’re doing. And what you do with that is now they can check on the business, right? And the most important thing, at least, that I have found, I think VCs care about this the most, and hopefully they do, because this would be a good thing if they do. I’ll tell you. Is trusting that when a founder says something’s going to happen, it’s going to happen. And if you do that over enough times, hopefully VCs really trust you.

And so going back to that strategy, if you let them invest a little bit in the beginning, and then you say, in three months we’re going to do x, y, z. And then you say, in six months we’re going to do x, y, z. And then in nine months we’re going to do this. And at the end of the year this is what’s going to happen. And if those things come true, they start to really trust you and they start to believe you. And then when you go out to do that fundraising process, it can happen in twelve hours.

You don’t need to make tons of materials. You don’t need to go out and do this massive, massive process. The problem with this is you are then not optimizing price. So that is assuming that you are not trying to optimize price. What you are trying to optimize is partner. It’s much more targeted. You’re basically targeting a group of people and you’re saying, these are the people I wanna work with and I wanna gain their trust and then they’ll invest in me versus I’m going into the market, I’m doing this massive competitive process and I wanna maximize price.

I think it really depends on who you get. I’ll give you an example of something that I haven’t trusted VCs as much with, and I think I’ve been right in some instances, wrong in others, is hiring. The areas I’ve been wrong the most in is when to hire a more senior exec. The VCs have been right. Like, my partners are right. They’ve been right. I took too long to hire senior execs in some instances and it caused us problems. It created competitors when there shouldn’t have been competitors, things like that.

The thing that I think they’ve been wrong about is who to hire. And I think sometimes the problem that VCs have is they’re managed up. They don’t actually see inside a lot of these businesses. They see the board meetings. And so sometimes the person who like presents really well at all the board meetings or something like that, they think of as that’s a really good executive. And then that person gets a reputation for being a really good, you know, executive. I’m not from the tech world.

I don’t know any of these backgrounds. And so I’ll sometimes get intro’d to someone from a VC and they have an incredible background. And I’ll be like, that person didn’t seem very good. And it’s just like my gut. And I think I’ve been right in some of those instances, and I’ve bet on people that sometimes they said that I shouldn’t have bet on, and they’ve turned out right. So when to hire execs? I’ve probably been wrong the majority of the time. Who to hire, I think I’ve actually been right a decent amount of times.

That happens too.

Yeah. Mean, so I’ll give you one example in our vertical. The vast majority of our customers don’t know who Sequoia, a16z, or any of those people even are. And so I think that, you know, maybe there’s a couple ways that people think about king making. One, they think of key making as it provides you with like more capital. More capital does not mean you run a better business. You could have as much capital in the world as you want. If you make the wrong product decisions, you’re just gonna invest in all the wrong places and it doesn’t matter.

It’s the same as VC. You have a 100,000,000,000 and if you put it all into the wrong things, that still goes to zero. So I don’t think capital makes folks win. The area where king making, or people think king making matters, is customers, where they basically say, hey, this has branded trust and so that is good. I think there is a little bit of legitimacy there, but it’s not like only the top three VCs give you that brand. Sure. Right? A vast majority of VCs give you that brand.

And what’s actually interesting for us is someone like EQT actually gives you that more than Silicon Valley because they’re private equity and, you know, a lot more lawyers know who that is, etcetera. So I don’t really believe in those two. The third one might be the one area that is helpful, and that’s just recruiting. Humans are very bad at judging how good other humans are. We’re really bad at it. Like, we’re really bad at it. And I can tell you a very clear reason for why we’re bad at it.

We still pay so much attention to someone’s resume. We care so much where they went to school, and this happens so much in technology. It happens a lot in tech. The only other area where I know it happens an incredible amount is legal. Legal and tech are probably two of the main ones where prestige matters, where you went to school matters, how your grades were, who you worked under, things like that, right? It does help you in a sense of if you get one of those brands, people assume that there’s maybe a higher chance of the company being successful.

The reality is that might be the wrong person to hire in the first place because the people that think or go to a company because of the investors usually don’t care that much about the mission of the company. And so my point with all of these things is there might be like some short term games of perception mattering. In the long run, it doesn’t matter at all because if you make all of the wrong decisions, nothing matters. It might help you recruiting,

Yeah. Do you care? How many times a day do you think something goes wrong at Harvey?

Constantly. Like twenty four seven. How many times a day do you think we feel like there’s like an existential threat? The big model providers are gonna release something and maybe we haven’t released something. All the time. Yeah. Like startups are very difficult places to work. And so you think of like from the outside, oh wow, they’re like growing revenue so much. They’re the category leader. They have all these investors, etcetera, etcetera. Right? GRR is high, all these things. Right? But internally at all of these companies, it’s chaos and it goes up and morale goes up and down, you face really difficult things, and then you have to figure out how to get through them.

Being a missionary really does matter because the reality is once you’re on the inside, the brand of the company and the success of the company matters less than when you’re on the outside. It matters a lot on the outside because people looking in are like, oh my god, that’s the most successful, it’s super well run and all these things. Once you’re inside, your day to day could be crazy, and you could be thinking you’re not doing very well. So I actually think it matters a lot.

People just don’t realize that because they aren’t inside of these companies,

I think just moving fast enough on product. Like that is always, I think, the biggest existential threat for all the application layer companies. And it’s not necessarily that, you know, Anthropic or OpenAI are, you know, tomorrow gonna put 50% of their resources after the legal vertical or tax vertical or anything like that. But they’re just improving their product and models. And the value of your product is going to go down unless there is a massive delta between what your product does and what you could get from an Enterprise GPT license.

Right? And so it’s just a constant existential threat of how do you make sure you get to escape velocity on product so you have enough of a product moat for them to not run you over. And I think about that daily. When I’m thinking about competitors, the main thing I think about is I’m more bullish on these labs than most people, I think. I mean, a lot of people are. I’m very bullish. They have incredible talent. And I think more about what are the frontier problems that our customers have that they’re gonna solve later.

Not entirely, but there were definitely so we route basically based off of the use case. We route to the best combination of models. And definitely our traffic went up to Opus 4.5 significantly.

Well, we aren’t routing

majority yet, but even if we were, there’s no conflict. Basically, like, they want us to win and they want, you know, us to use the best model. Right? There’s nothing in our agreement or our relationship or anything like that that you have to use OpenAI models. And if anything, I think that for them, getting the feedback from application layer companies on this is where your models aren’t doing as well, this is where your models perform super well, and this is where you need to improve is super valuable to them.

I think that we’re seeing a plateau in performance for consumer use cases. And the reason why I think this is like a misnomer or something that people actually shouldn’t pay attention to is we don’t need them to be better for consumer use cases. Like, I feel like this is something that folks don’t quite understand is like a lot of the consumer use cases four was like, we’re done. You don’t need better reasoning to solve these problems. What you need is like different contexts. You need it to connect to your calendar.

You need to connect to all of the different apps you use and things like that. That’s what an increase in performance is for them, right? There might be a plateau in some of the consumer facing side of things. On the enterprise, I think things are gonna keep going and especially cogen. I think we are not going to see a plateau in cogen. I think that is gonna get much better really, really fast.

I think the slope will only increase. I think that it will get better and better and better, and I think that will unlock a lot of productivity just across the entire world.

It’s a combination.

Yeah, I know. And we talked about this. I’d buy them both at double.

Maybe not quite 1.6. There’s a couple of things I need to see from them before I would do 1.6.

to see from them before? I think the main thing with OpenAI is they have so much consumer brand. It is so powerful. And especially outside of x and outside of kind of like our worlds, worlds, it is so unbelievably powerful. And I don’t come from the tech world. And so all of my friends, all of my relationships before this were outside of tech. That brand power is incredibly powerful. And I think like more focus on consumer and just tripling down on that, that I think is where I’d be the most bullish.

I think that on the enterprise side, there will be multiple winners. Like enterprises don’t allow there to be one winner, right? So no matter what, there’s gonna be OpenAI is gonna get some of the enterprise market, Anthropic is gonna get some of the enterprise market. But the consumer market, I do think OpenAI has an opportunity to take a lot of this. And obviously the main competitor there is gonna be Google. But I think they both can be just astronomically huge companies. The other thing that I feel like folks don’t realize is everything could pause.

Like both of those companies could stop developing things right now and the amount of saturation of AI that would just happen to the economy would still skyrocket. Like, we’re so far from just the ability of the models right now being integrated into daily life. People do not know how to use these systems. Like, don’t. They don’t. And businesses definitely don’t. And there’s so much of, like, the capability overhang is so high. Like, I think it’s higher than anyone is even talking about. It’s astronomical. And so I think if both companies literally just stopped shipping things, their revenues would still explode because there are going to be so many companies building on top of their models.

There are gonna be so many different parts of the economy that adopt these things that I think we’re still in early days.

I think like three to five years until we see like massive, massive productivity gains in enterprise. I think the capabilities are there already. The capabilities were there two years ago. A lot of this is like, if you think about just like the average enterprise workflow, there’s like 17 different systems they’re pulling data from to get that workflow done. Like literally, 17 might be on the low end. Sometimes it’s like 50. And you have a 100 tabs open, you’re opening in all these different apps and they kind of connect to each other.

They don’t really connect to each other. Right? And so the long tail on actually getting these systems and agents to do a task from start to finish is so difficult. And the problem that you’re gonna end up having is you have these building vertical agents like us and Sierra, etcetera, but a lot of even the verticals connect to all of the other parts of the enterprise. So like one thing that’s happening that’s interesting for us is, you know, a lot of our revenue is starting to come from Global 2,000 or Fortune 500 companies.

And we actually haven’t built many features for, like, tax compliance and procurement. Right? Starting to happen is those departments are adopting Harvey, even though we haven’t built features specifically for those departments. The reason why is the legal department actually interacts and just like legal documents are such a core part of a business that they interact with all of these different parts of the business. Right? And so we released basically a feature that’s multiplayer. It’s called shared spaces. At first, like a lot of the impetus for doing it was you want a large corporate like a Walmart or whatever to work with their law firms in the same platform.

And that’s happening. But actually what’s starting to happen is these corporates are using the legal team, is working with the Compliance department, is working with HR, is working with everything else, all in Harvey at the same time.

No. We were working on multiplayer a long time ago. I think one of the things that’s interesting about our company is we started with, like, the hardest customers, and we did the same thing actually on the in house side too. So, like, we had bank customers a while ago, and the security and permissioning systems that you need to build for a bank are so much more in-depth and, like, the enterprise readiness than for a lot of the other folks. And the biggest problem with multiplayer and the way that we’re doing it is we’re allowing the in house side to kick it off or the law firm side.

And to do that, the security and permissioning that you needed in place for both is astronomically high. So we were working on this like a very, very long time. We were going on it for like six months to almost a year. We just did all the permissioning and all of that stuff first before you do kind of the UI and on top.

I think that if you are number two in the market, one of the things that can get you a lot of attention is just attaching yourself to number one in any way, shape, or form. Right? You kind of get free press from doing that type of thing. It’s a good way to basically jump on to the distribution that the other front has. Do

I mean, I definitely respect them. And I think one of the things that they did really well is I think they did a great job in Europe. And, you know, this is back in 2023. We’re we’re not that I think they’re, like, six months after us or something. It’s not that big of a gap. And I think, like, one of the things that I would have done differently in the beginning is just invest like more in Europe in like 2023. And actually, we a lot of our first customers were in Europe.

But having folks on the ground here is just really, really important and respecting kind of like the different cultures and how to productionize that and all of those things. Why did you nod out of interest in terms of coming to Europe earlier? It was just bandwidth. Like, when we signed Aino Sherman, which was our first customer, we had four people. So we did a 4,000 person enterprise grade roll out with four people.

The engineer who had joined, was our first engineer, Gabe was basically coding everything before that. He had been there for, I think, a month. And then we onboarded a 4,000 person, very large scale enterprise security. It was incredibly important team. And so it was just bandwidth. Like, when you’re scaling that quickly, it’s harder.

Not anymore. So

But but actually I think this is important. This is this is something that’s interesting. When I look at a lot of AI application layer companies, if you go through their LinkedIns and you look at the engineers that they’re hiring, it’s like 90% front end engineers, which is interesting to me. And a lot of that reason is I think like vibe coding works much better with front end than it does for infra. And a lot of, I think, what’s happening to AI companies or is going to happen to them, and it happened to us in early twenty twenty four, is you do a bunch of front end and you make really pretty UIs and really nice demos.

Right? And then you use that to land all the customers. And now you have a lot of actual active customers, and you haven’t invested in the architecture and the infrastructure of hundreds of thousands, millions of customers using your product. And we kind of made that mistake in 2023. And in the beginning of 2024, something that kind of slowed down our shipping velocity is we added just tens of thousands of users in like the Q4 before that. Right? And we didn’t quite have the infrastructure, like, to support that.

And now if you look at our team, it’s almost like 40% of our entire EPD org is very senior from like a Databricks or something like that, infrastructure engineers. And it’s a long term bet that as you get these agentic systems that are processing tens of millions, for us, I think last year we almost did like half a billion documents or something like that, that you need the infrastructure to actually support that. So it’s not just about how do you win the demo and how do you win the deal, but how do you actually create an enterprise, very scalable infrastructure on your product.

And I see a lot of AI application layer companies not doing this.

What I would tell them actually is your GRR matters. And I think like one thing that a lot of investors in the AI space have been not paying attention to is GRR. They’ve been basically just looking at net new ARR and kind of being like, Ah, churn is fine because they’re growing so fast that maybe they’ll pivot or they have some customers. I think that’s a huge mistake. You’re gonna see a lot of companies in a lot of verticals that go really, really fast to signing a bunch of customers because maybe there’s only one in the vertical and now there’s a second player, etc.

But then they have to actually support all of those customers. And if you don’t have the infrastructure in place and you make a bunch of promises upfront and then all of that falls down, you’ll start losing customers really, really fast. And I see a lot of AI companies not focusing on this. And I think that’s gonna be like a huge reckoning for folks once they get past 100,000,000 ARR.

Agree. And I think that what that requires then is going back to what we were saying, which is how you go product market fit and then how do you get company market fit? And that’s actually like how you structure your company. And I think part of that is different than how it used to be in the past, where if you have long implementation cycles or one thing that is gonna end up happening, I think, is a lot of these verticals are gonna land at a big Fortune one or Fortune two or whatever, right?

And their product is gonna expand massively. This is a really interesting thing about Microsoft. I don’t know how much you know about Microsoft sales force, but they started in the beginning and the vast majority of their sellers, it was presales. Right? So spear fishermen. Like, have tons of that. Right? Old school spear fishermen. And they actually eventually migrated to a lot of what they have is post sales. Like a lot of their investment is actually in post sales. And the reason why is because their customers just their NDR goes up, up, up, up.

They keep buying more things, they buy more compute, etcetera. And I think a lot of these enterprise companies should start thinking about their company that way. In the sense of like, sure, there’s some of a land grab right now, but really what’s gonna matter is if you are bullish on AI, you should be bullish on your product, the value of your product. Like we are in day one of product development. Like it’s gonna change astronomically. And so what’s more important than landing new customers and getting really high ARR is can you retain those customers because that customer that pays you a million today, there’s a real world in which they pay you a 100,000,000 at some point.

I think Databricks is a company that’s done an incredible job with this.

Yeah. Mean, there’s a third one that now develops, which is these AI products are so powerful and I think over time the ROI is so high that your ROI can become so massive that it’s less of a hostage and it’s maybe closer to how Palantir thinks about things. In other words, the more value that you create for the customer, the higher you get paid. And I think that more and more companies are gonna start aligning to that. And the reason why is let me look. I’ll do the law firm side and I’ll do the in house side.

For law firms, they bill by the hour. And so a lot of people are like, how would could you ever sell to them? There’s no way that this is gonna work. Right? Two things might happen. One, they might switch to fixed fees. Well, now we’re good to go and efficiency is really good. I don’t think it’s gonna happen that The second thing that actually is happening, and we have so many law firm customers that have gained new business by building something custom in Harvey and saying, we’ll do this M and A with this custom solution that we did in Harvey, and they win that deal over another law firm.

That’s not a hostage. That’s this product that I’m paying maybe like a million dollars for a year just earned me a deal that’s 20,000,000. What is the ROI on that? Incredible. And on the in house side, it’s even clearer. It’s just if you save time, you’re saving tons of money. So I think that hostage thing can actually change to more like how Palantir thinks about it, which is I think the value of B2B SaaS is about to become astronomical. And if you can figure out how to align your product to that ROI, it’s not a hostage.

You’re just completely aligned with your customer.

Yeah, I think that that’s not, at least in our vertical, what I’ve seen. I think, like, there are a lot of areas where we are gonna start moving to consumption based pricing. For at least for us, for our customer base, that would be completely fine with them.

Already seeing it happen.

Yeah. So there are a couple companies that have basically said that the Harvey budget comes out of their spend on professional services, not out of their tech budget. The budget for professional services is in the billions a year versus the tech budget for that G and A group is astronomically smaller.

and I think that’s what’s really interesting about our business is a lot of the work that we’re doing for like a corporate is not the work that our law firm customers are doing. It’s like alternative legal service providers. It’s this like lower end work. What percent of revenue

Right now oh, so you mean corporate versus law firm? Yeah. I think it’s around 40 percent of our revenue is in house corporate, and 60% is law firm. Something like that. Is that what you thought it would be? Something like that. I mean, if you just look at the breakdown of how many lawyers exist on earth and how many of them are at companies versus in house, that’s pretty much the same. In five

I think it’ll be similar.

it’ll be the same.

I don’t think so. I think we’ll just get more work. So interesting, I had a conversation with Are you gonna make my girlfriend unemployed anymore? No. I had a conversation with a pretty large private equity shop recently, and they were talking about their year is gonna be incredible. They think it’s gonna be, like, a big M and A year. It’s gonna be great. And they were talking about, like, how they think about legal fees. And the way that they thought about legal fees is, the reality is, like, it’s gonna be a big year, and whenever we have a big year, we pay more in legal fees.

That’s just like how it happens. Right? But there are certain things I don’t wanna pay for anymore. Like, there are certain parts of the deal, etcetera, that I just marking up NDAs, whatever it is. I don’t wanna pay for that anymore. But there’s all these new things that I’m paying law firms for. AI risk, like should you buy this company? Is there a, you know, a problem in x y z country with, you know, an act or something like that that’s gonna change it? There’s so many new pieces of work for professional services that my gut is that’s not what’s gonna happen.

In fact, think what’s gonna happen is the professional services market is going to actually keep growing at the same as GDP. One way to think about this is most professional services is cyclical. So So if you have a really good year, professional services have a really good year. That’s almost always how it works. Other than bankruptcy and litigation is somewhat countercyclical, it depends on the area. And so I think people think about this and they’re like, Oh, wow. AI is gonna impact legal and it’s gonna just destroy all these jobs.

The thing they aren’t thinking about is all of their customers are using AI to create more products. What happens when you create more products? You need more product to legal advice. What happens when you’re expanding into other countries faster? What do you need? Regulatory advice, right? And so I think people are thinking about AI in all of these industries as like a vacuum. And the reality is you should think about AI as like the entire economy what’s gonna happen. And probably what’s gonna happen is the economy is gonna explode, these companies are gonna have crazy expectations for what they can do, and the professional service providers are gonna have to respond to that.

I don’t think it’ll be this year. I think there will be bumps. Like, I definitely think we will have more moments like the deep seek moment where everyone freaks out, right? And I think we’re close enough to an edge of If enough people say that there’s gonna be a bust, it’s pretty easy for one thing to happen and for everyone to freak out and there to be a bust and it’s a self fulfilling prophecy. Those are usually pretty short. I think that we will have a bunch of short ones, but I think long term AI is going to completely reshape every part the economy.

Like, I very strongly believe that.

Oh, I think that it’s similar in Europe to where it is in a lot of places, which is you don’t wanna go into a country or a domain or anything like that and act like you know how to do something. Right? Like, you really need to partner with an industry or you need to partner with a geography. Right? And when I say, you know, we should have invested more in that, it’s more like we didn’t invest in it that much in 2023 and 2024. We invested tons last year and we’re investing even more this year, and the difference is pretty massive.

The difference in kind of the quality of our team last year and the partnerships and things like that and how our product is localized for each geo is just a huge difference. But you can’t do this from sitting in San Francisco and, like, kinda thinking about how to do it. You gotta travel.

It’s not a difference in talent. It just takes a long time to hire people. And so you have to just think about it with a way, way longer time horizon.

just really hard to hire people. And so that was just kind of interesting to me or something that I wasn’t used to, whereas in The States, you can hire someone and they start quite literally the next day sometimes. Literally. Or if they have to give two weeks notice, they start exactly two weeks later. That allows you to be a little bit more just, I quickly need to hire this retroactively. I can fix a problem. In Europe, you have to plan out more. And so we’ve done a lot of really big office openings, like we just announced Paris and Dublin and a bunch of other ones.

But you have to just think about this stuff at like a longer time horizon. You can’t do it instantaneously.

That’s not what I found. But I will say we interact I interact mostly with lawyers. And, like, lawyers have billable hour targets. And at the end of the day too, they are either at international firms or they’re competing against international firms. So I have not found that at all. Right? I mean, there are so many incredible hardworking lawyers

UK lawyers work insanely hard. Yeah. And so it might be that I don’t notice it as much because the domain that we’re in, they work the same across the globe. Lawyers are just incredibly disciplined, hardworking people.

That’s a good one. You definitely need to be obsessed.

So obsession is definitely very important, but the one that I look for right now a lot is ownership. And there’s a bunch of different ways that you can assess this. But you do, over time, you start to be able to just read if someone actually can take ownership over something or not. And the reason this becomes really important is as you scale as a company, it becomes really hard to figure out where a problem is stemming from. Becomes It hard. This is how this ends up going, is it used to be I knew every single thing that was going on at a company, and I can just be like, hey, that’s where the problem is.

I’m gonna unblock that. Right? Now we’re getting to the point where I know most of what’s going on at the company, but sometimes something is, like, so low down that I don’t know what the problem And if I ask five people, they’ll all do this. It’s like the Spider Man.

know. And I have found that there are a lot of people in tech that have done a really good job managing up. They have ridden a wave of their team’s success without being successful themselves. So the thing that I look for a lot is can people admit their mistakes? It is so obvious when someone is actually admitting mistakes versus they’re saying that one of the biggest things they’ve done wrong is actually like something that’s really great. We’ll go back to something you said earlier. Like when you said some of the things are bad habits that you’re trying to Well, I mean, yeah, yeah.

And you said, like, check Right.

Being slack too Right. And and, like, that I think is a good example of I was

Yeah. And and, like, I would actually if I had interviewed myself and I saw that as a like, the way that I answered, the way that I would push on this is I would say, why do you do that? Like, why do you do that? And my genuine answer to be to that would be I have trust issues. Like, I have trust issues. It is hard for me to trust that somebody else is going to handle that problem. And now all of a sudden, it actually is ownership.

That is an actual problem of being a leader. You cannot scale a really good company and get to tens of billions of revenue if you have constant trust issues and you can’t trust other leadership.

Great.

Yeah. I think my trust issues, mean, I think part stuff is, it’s hard to tell whether it’s like nature or nurture, right? But I think that I definitely had some problems with authority when I was younger. And I had a not the normal Silicon Valley kind of background and upbringing. And because of that, I think that I had to I really went out on my own at like a pretty young age and was pretty independent. One thing that you have to learn when you’re leading a company is you are a leader and you are a partner to the rest of your team.

It is not just you. Like, I am not Harvey. Like, Harvey is not me. Like, it is a group of people that are building this company. And I think that sometimes what founders can end up doing is they can start basically saying, I want to be the number one I think of this as like a sports team. There are people I know that they don’t care about winning the championship. They want to be the person who scored the most points. And they’re okay with losing the championship as long as they’re the one that scores the most points.

Those are the exact type of people that I do not want to work with. I want to work with people that do care about how many points they’ve scored, but they care about that because they helped win the game. Right? And I think that’s a huge problem in tech, is where you have too many people that it’s me, me, me, me, me, not company, company, company.

I mean, I think The

But but but, like, look at

This also I I forgot who tweeted this. I don’t I don’t remember who it was, but this was, like, end of twenty twenty two, and it was, like, right after the ChatGPT launch. And someone I forgot which VC did this, so sorry for not giving credit to them. But they basically tweeted, My prediction is what’s gonna happen is a lot of VCs, because they don’t understand the AI ecosystem, they’re gonna revert back to looking at resumes because they don’t understand. And this is what people do.

When there are situations of chaos and folks don’t know what’s going on, the safest thing is to go look at other social signals to make decisions instead of using your own gut. And so I think that that’s happening a lot in AI and hopefully, you know, as the markets mature, this stops happening. But there’s a lot of, I don’t really understand this. And so what I’m gonna do is look at the resume or look at a logo or look at that because that seems like a safe bet.

How many truly great researchers do you think there are? Hundreds, and that’s it.

I don’t think you can. That I think is actually a pretty big disconnect. The researcher community knows. So like this is a huge disconnect between VCs, I think, and the researcher community. So what would you advise

What I would do is you can use the researchers to pinpoint who is the best researcher. So in other words, if you ask a bunch of the researchers, not other VCs, ask a bunch of the researchers, who do they respect the most? They have such a tight knit community and it’s all merit based. Like, I’ll tell you a group of people who do not manage up. Researchers. That is not what they do at all, for better or for worse. And because of that, if you ask that community who are the best folks, they will triangulate for you and they will help you find.

And usually it’s not the loudest. They aren’t necessarily the most famous, etcetera. Like, it’s not like that.

Yeah. I think there’s a combination of things. I think the reality is what researchers care the most about is working on really hard and interesting problems. And I don’t say that as like bullshit. Like, I think that there sometimes, like to your point of like, people just wanna join hot companies, a lot of tech, I think, is like people just wanna join hot companies. Researchers really genuinely do care about pushing research forward. And I think what ends up happening sometimes is the leaders at some of these companies kind of change the direction of the company and the researchers are no longer interested in that direction of the company.

And that happens a lot at these big labs. Like, they’re making multiple bets at multiple places and what ends up happening is the people say, Ah, this isn’t what I signed up for. It’s kind of like a bait and switch, and they go to something else. And by way, this has nothing to do with thinking machines. I’m just saying this is what’s happening, I think, at all of these labs is, you know, you go to work at Meta, go to work at OpenAI, you go to work at Anthropic, and you think you’re gonna work on something, and then you end up not.

And, you know, they’re in insanely high demand. They’re able to pick what projects they wanna go work on.

Yeah. It’s incredibly it’s very, very high. And one thing maybe to bring up is I think that over time we have this interesting thing where it’s like a lot of the AI companies, like application layer companies, including ourselves, like in the beginning were like rappers. Right? And what’s gonna happen over time is these companies are gonna do two things. One, a lot of just the core software that they build, it’s not even gonna be AI related. It’s just core software building. And that’s actually gonna be very differentiated.

The second thing that’s gonna happen is none of these companies have access to proprietary data until very recently. And so a lot of the stuff that we’re looking at is how do you actually create custom solutions for a large enterprise? And that is AI problems again. Right? So now we’re all the way back to AI actually mattering and I think that what we’re gonna see is that kind of main problem of a lot of these companies haven’t hired any AI talent. And a lot of what they’ve done is kind of, you know, the model is basically the entire product.

And as these companies scale and they have to create more differentiated products, I think having AI talent is gonna matter again. And we’re just getting, just now, getting to the size of the company where we can hire the people that we need to kind of do a lot of that frontier work.

Maybe two pieces of advice. One is listen more than you speak. It’s very, very dumb, but it’s true. I think a lot of people in deals, they think that movement is action. So they think that movement is progressing the deal forward and they think that if they talk the most, they’re in control of the deal. Not true. In the same way that in conversations. Just because someone isn’t participating in that conversation doesn’t mean that they aren’t listening. It doesn’t mean that they have maybe the upper hand or something like that.

Right? And so I think listening is really important and I see a lot of folks think of deal making as like chest forward and if I’m the loudest and I’m saying the most, the reality is all deal making is just people reading. That’s it. And it’s people reading at scale. So it’s people reading like a one on one conversation, and then it’s reading groups of people, and then it’s reading entire verticals of people, etcetera, and it’s figuring out what they want. The second piece is know when to not negotiate.

This, I think, is actually really, really important, and the best deal makers I know are very good at this, which is there are certain deals where you want one thing from the deal and nothing else matters. This only works when you understand the value of something more than everyone else does. And if you understand the value of something more than everyone else does, throw all your all of your principal deal making and you’re supposed to negotiate x and then y and then it’ll be 50% in between and all of that.

That’s all bullshit. Throw that aside and get the thing that you know is more valuable than anybody else does done. The best deal makers I know, like the best, they do they know that very well.

There’s just certain deals that we’ve struck where I wanted one thing in the deal, And maybe the financial part of the deal or something else about the deal, like my CFO or my VCs are like, Oh my God, don’t agree to that. Like, do X, Y, Z. And I knew that if we signed it and we got particular thing that I wanted from that deal, it would help our company to such a degree that it would help close another deal or it would help do something else.

Like a lot of what you’re doing, I think like the very good deal makers, and I think Sam Altman is incredible at this, is you’re holding multiple ropes. You can kind of think of it as like, you have like 17 ropes in this hand, you have 17 ropes in this hand, you’re grabbing all these ropes. And at some point, you’re gonna get like pulled apart because it’s just too much pressure and you’re gonna lose. Right? And what you do is you get good at tying off one of the ropes and then that pressure is gone and you have one tied.

And then you tie another one and then you tie another one and tie another one. And tying those ropes allows you to pull more ropes. And like that’s a lot of how I think about deal making. I think another company that has done an incredible job of this is Microsoft. Incredible job. They’ve created this partnership ecosystem. Right? A lot of people have given them a lot of flack of like, why do you let people do x, y, z partnership? Why do you let people build on this?

Like, they’re just gonna take it. And it’s very clear that Microsoft has actually won in a lot of areas because they’ve decided to partner with kind of everyone. Instead of saying, Ah, no, brass tacks. We’re gonna be very tough on this.

With hiring, this is a huge mistake that people make, massive mistake. If you want to hire somebody, hire them whatever they want to be hired and put them in the position that they want. If they’re best in class If you can’t tell if they’re best in class, it’s a separate problem, but don’t go back and forth.

Well, you don’t be That’s in that not gonna be a legendary category defining company. 100%. Won’t be.

It’s

I think one of the things that I’ve done wrong in the past is I think that I have thought that someone couldn’t scale because they have bad communication skills and I’ve misread that. And then I didn’t realize how easy it would be for them to learn how to do it. I think that sometimes because I feel like I do a lot of in person interviews and a lot of those things, that sometimes I also end up making the mistake of reading too much on the surface and not going deep enough.

I think I’ve gotten better at it, but that’s one of the things that I’ve done with hiring is I’ve fallen for the resume trap too. Like, I’ve definitely fallen for it, and I think that’s

quick

So it wasn’t a cold call. We cold emailed them. And we cold emailed Sam and Jason Kwan. What we had basically done was we went on rlegaladvice, which is basically like a subreddit for asking legal questions, and we grabbed a bunch of those questions, ran a chain of thought product that we had basically built on top of it, and gave it to a bunch of landlord attorneys. And then we basically said just like, look at these questions and tell me if they’re you would send the answer, we didn’t say anything about AI, to the consumer who asked the question.

And of 86 out of 100 questions, three out of three said, this is a perfect answer. I’m done. And we cobbled all that together and we just sent a cold email to Sam Altman and Jason Kwan. The idea was basically, hey, did you guys know that at this point it was GPT-three and just the API was public, I think the end of twenty twenty one or beginning of twenty twenty two, had an API. Did you know it was this good at legal? And that was it.

That was basically the subject line of the email, was like, Did you know it was this good at legal? And we met them pretty recently after that.

It went well. We had a call with Jason first and just kind of talked about what is our strategy as a company and what would we build? And then we had a final pitch to kind of the rest of the OpenAI C suite, actually the morning of 07/04/2022. Yeah. It was, like, 11AM on July 4, and we did a pitch to the rest of the company.

Yes. I had a weird thing where, like, I didn’t know who any of these guys were. And so, like, when we were doing our series so we It’s

them. Like, now I do. But back then, I didn’t know anything. Mean, I didn’t even have friends in tech. And so when we did we did our seed and we didn’t go to anybody else. It was just OpenAI. And then for our Series A, we got How like

I don’t remember what the post actually was, but the pre investment was like 4,000,000 or something like that. Anyway, what happened at this series A was we went and met with basically I think it was like 10 VCs or something like that in like forty eight hours. Right? And I quite literally did not know who the VCs were. Like, I didn’t know. Like, my co founder was from tech and he, like, knew and he basically gave me advice and things like that. So I actually think I had, like, a weirdly unique view of the VCs because my entire judgment of them was just how that first meeting went.

Like, I didn’t know that Sequoia was in this tier and this one was in this tier and, you know, whatever. I didn’t know any of that. I didn’t even

I think like half or something like that. Yeah.

The worst term sheet or the worst meeting?

Oh, I’m not gonna say that publicly. It’s worth a go. There was one. Oh, there was one Why was it bad? Just tell me that. Yeah. So there was one where the person was quite literally on their phone. It was a Zoom, and they were on the phone the entire time during the pitch. Did not even make eye contact. Literally.

a friend.

So I said earlier that a lot of company building has changed. I actually think a lot of it remains the same. And so there’s a lot of kind of just like core first principles of scaling a company that I am much more focused on that I wasn’t focused on in the beginning. I’ll give you the dumbest example ever. For the first like two years of the company, when I was doing like revenue projections, I never, and this is embarrassing, really embarrassing, I never was like, If I want to hit this amount of net new ARR, you need to hire this many AEs at this quota and this is how long it takes to ramp them.

So you need to hire them at this point before you do that. I’m dead serious. Like, never even thought about that. Right? And it’s these really core, like, laws of physics about companies that remains the same. Like, there’s no different in AI. I think like in the past, maybe what have I learned in the past twelve months, maybe it’s more what have I learned in the past eighteen, like the second half of the company basically, is how much of a lot of the company building is actually the same.

And I probably should have listened to people a little bit more about that.

Yeah.

Yeah. I think that there’s obviously a difference in scale, right? A16z is just a lot bigger. I would say that a16z is also louder, and so they’re a little bit more in these other regions and other areas and things like that, whereas Sequoia is, you know, it’s just a different style. Right? It’s closer to there aren’t as many partners and things like that.

Definitely. Yeah. Definitely. I mean, I’m always nervous. And I think, like, one thing too, I think I said this earlier, but I think Keith Rabois says this, who I’ve actually never met in There’s a couple people. I have a list actually at the top of my doc. I have an operating doc, and I have a list of people and they all have two words next to them, and it’s the thing that I’ve learned from them or something like that. One of the things I think he was talking about at some point is how important it is that you should be constantly stressed, like constantly stressed, and do things that make you stressed every day.

I strongly agree with that. Like, really agree with that. Like, I think the times that I’ve stagnant or the company has stagnated is every day I don’t have something that’s really stressful. The weeks that I do the best work or feel like I did the best work is every single night before I go to bed, I’m like, oh, shit. Tomorrow is gonna be there’s so many things, everything needs to go right, I’m really stressed. It’s compounding, like massively compounding for you as an individual to just put yourself through that stress.

That’s he

but it’s interesting. Like, I think Pat also has this too and all of my VCs do. And Pat in particular, he really thinks of like just constant, like relentless application of force is what he says. And that’s actually what I have next to his name. That’s incredibly important. And I think that if you lose that as a company, the company is pretty much over. I think that you as a founder need to constantly, constantly be applying force, and it starts with applying force to yourself. And if you aren’t applying force to yourself, you don’t have that ownership mentality, and you’ll start to get weak.

And I think that will trickle down to the rest of the company.

I have Brian Halligan’s on there and his is no. It’s just the word no. And one of the things, he’s been incredibly influential and helpful to me. And one of the things I had a problem with is just like saying no to things. And it’s a huge, huge problem as you scale as a founder of figuring out an ability to actually block off time for yourself and to say, No, this is my priority. I’m saying no to everything else. Right?

And it’s the same with this is also probably one of the main things that hopefully I’ve improved on product is I used to every quarter just be like, And there’s a P zero and then also there’s a P zero zero and then there’s this and this and this. Right? And I’ve to be a lot more disciplined and try to be disciplined with my team of like, every time that we do product planning, something should hurt. It should feel like a breakup. Have to there has to be a couple really good ideas that you say no to.

It’s the same as across the entire company, and Brian Halligan has been really helpful with kind of teaching me how to do that for myself and how to do it with the rest of the company.

Yeah, yeah, yeah. No, I think that’s right. It should feel painful. Painful. It really should. And I didn’t do that in the beginning. I was like, oh, we’ll just do everything, whatever. And I think having said that, I do think you can do more, and I think you can be very ambitious, but you do have to strike a balance between the two.

I’ve always respected Founders Fund. Like, I’ve always really, really respected them. I’ve respected the hell out of their companies too and a lot of their founders. I just I I respect them a lot, and we we haven’t ever worked with them.

I mean, we obviously have revenue targets and things like that, but what I wanna switch from is a productivity software that is like a nice to have to closer to like an operating system that is like pretty much crucial to the industry. And so it’s more how do we we built a lot of different, like, features across the board. Right? Like, we very much, like in in Parker Conrad’s voice, like, built a compound startup. Like, very much. We haven’t tied it all together yet. We’ve done a really good job of creating all the pieces.

Hasn’t tied together. Maybe an interesting stat on this is our DAU over MAU for people that use four plus product lines is 74%. That’s like Slack level, right? Slack was like 80 or something like that. It’s incredibly high. The percentage of people that have used four plus products is very low, and it’s doubling every quarter. But what I really want to care about next year is can we make this like infrastructure? Can we make it so it is like this is a core piece of a lawyer’s work and day and they live in it?

It’s a 100% that, and I think we’ve done a good job of creating all the features. But now I wanna combine all of them together so we can get to that 75% DAU over MAU. That is how you really show that you’ve created something that’s integral to the industry.

Well, thank you so much for having me, man.