Harvey AI hits $100M revenue with 350 employees and 500 customers as legal AI reshapes law firm economics

about legal AI and the impact of open source. I want to know if anything changed this week for his business. We will bring in Gabe from Harvey AI, the legal Hello. Welcome. How are you doing? What's going on? Good. How are you guys? We're great. Great background. Great background. About to say is that mahogany?

That's nice. The official wood of business and law. That's mahogany. Some books. It's got to look professional. Fantastic. Um, what what's the latest uh g give us the update. I have a bunch of questions that'll go all over the place, but give us the update in your world. Yeah, so last week hit three-year anniversary.

Um, have kind of scaled up to 350 people, 500 customers, just hit 100 mil in revenue. There we go. Love hitting love hitting the gong for revenue. You know, we hit the gong a lot for hundred million dollar rounds, but it feels better when it's real customer.

We we have this silly soundboard where we play the over overnight success sound, but uh overnight success. Yeah, that one. Um and usually it's because some founders been grinding for like 10 20 years and then they finally hit breakout success. But like Harvey is kind of an overnight success.

What actually set you up for this? Obviously, uh you were in a unique position to actually go after this opportunity. this wasn't just something that you just lucked into obviously. So give me a little bit of the prehistory.

Yeah, I I actually I started doing AI research in like 2012 and tried to do some other startups and so I think there was a lot of this experience of trying to build these companies without this technology. Um, and I think I was at Meta on the large language model team right before Harvey.

And my roommate, who is now my co-founder, Winston, was an associate at a law firm. And I was showing him kind of the GP3 models and he was showing me his kind of the legal work he was doing. And I think it was pretty obvious these models were going to get much better. Uh, and that seemed like a great application.

And then I do think there was like some factor of luck of you know raising from open AAI how good GPD4 was and then kind of really focusing on big law. Yeah.

So you add all that in then we can do the overnight success because it's like easy if you want to just grow really quickly just spend a decade researching a fundamental technology that will change the world. Uh anyway um what uh yeah uh so so much to go into here. Yeah.

I mean I one one thing I want to you know you at this point I'm sure are talking to or have closed all you know pretty much all the top law firms in the United States. Uh I'm sure the reaction internally at all the different firms. I'm sure there's some excitement.

I'm sure there's uh there there's clearly excitement otherwise they wouldn't be uh you know signing up uh and becoming customers. But I posted something uh last week.

I was in New York and I spoke with uh I was speaking with a lawyer randomly and he was telling me about their AI adoption uh internally and he said and uh this was his point of view. He said adopting too much AI will hurt our bottom line. Uh and I po I posted that note uh yesterday and it sparked a big conversation.

Um, you know, the obvious response is that you don't pay great law firms and lawyers for the execution. You pay them for the overall product and the sort of like guidance and advice and strategy and and all these other things.

Uh, it's not just like the raw like generation of documents or reviewing contracts and things like that.

So, how how do you know how are um give us kind of the overview of how different firms are reacting and sort of leveraging AI and then I want to hear how you think law firms uh might need to like reinvent themselves and like kind of even adopt new business models and approaches and things like that over time. Yeah.

So, it's definitely like a wide range both law firms but also kind of individual partners within law firms. And so we have, you know, firms that are still in the let's test this out. Let's see what to do with this.

And then firms that are we have already co-built software products or models with them and they are we are revenue sharing and selling that to their customers. So there is kind of the full range and yeah, I think it'll be interesting to see how it plays out.

I think my guess is long-term where this is going is I think your point is exactly right.

Like I actually think if you look at partners billables, they're actually undervalued where it's like if you want to do a massive public merger, that partner is worth much more than 3,000 an hour, but it kind of gets subsidized by all these associate hours.

And so I think there will be some changing of, you know, how you charge for these. Um, but so far I think we're still kind of early days in that conversation and it's much more how do you, you know, get adoption? How do you start upskilling lawyers to use this technology? Um but do have a lot of fun. Yeah.

The interesting thing I think you know we obviously have a bunch of founders uh that uh watch uh the show and I think it's interesting to look at the invoices you get from your law firm and and just like look line by line where was I getting like real value and legal advice and strategy and all these things that are extremely high leverage and where was I paying to like recreate a variation of this document or fix a punctuation error and things like that.

And obviously you can't see exactly but I it is easy to imagine in the future where in the same way that you have a 100 100x engineer today that's using uh you know cognition devon or or and windurf or claude code or or whatever variety of cursor etc where they can be ultra high leverage because they effectively have an army of people you know working under them uh agentically.

I think you'll see that in in uh it's hard not to imagine that happening in in law where you have this like incredible individual player who's able to like produce potentially an order of magnitude more work uh in the very near future. Yeah.

And one thing we're also starting to think about is like it's always interesting to ask why do you have this billable model structure in the first place? And I think one big problem is pricing legal work.

And so like the reason you have this model is like these projects are so complex that it's actually impossible to do fixed fee without you know this technology where it's just how much does it cost to do this merger. It's like you're going to have all these things that come up.

But I think as these models get better the same way it's hard to predict like how long will it take to fix this bug. But now with these models you can kind of look at a codebase and say oh actually I can tell you you know some good error bounds.

I think there's things like that you can start exploring that make more of these new business models interesting. Have you seen Have you seen lawyers in big law spinning out and and building firms from the ground up to try to disrupt themselves and create new models?

I mean, we've seen uh we've had the founder of Crosby on which is like they're building a firm themselves and like focusing on like a few specific types of of contracts initially.

And I could imagine it's kind of hard if you have a a law firm with thousands and thousands of lawyers, it's hard to like totally change the business model where if three people spin off and they come to you at Harvey and say like we want to leverage this ground up, build a firm around this.

Are are you seeing that at all? Yeah, one of the fastest growing I think they might be Amla 200 firms now is actually partner only and so I think this model is I think that the challenge is it just depends on the type of legal work you're doing.

But this is also traditionally like not traditional in the past couple years what uh firms like PWC have also been doing of how do you bundle this legal work and provide alternative ways.

So I think some of this work is is getting unbundled and with different models you can kind of tackle different parts of the legal work. How much how much do you guys care about models getting more intelligent versus just delivering on the current capabilities?

Yeah, I think for us the big challenge we definitely still have is like the models aren't good enough to do the very complex legal work. And so I think for us, you know, Sam has the quote of like try to build some company that like as the models get better, your company gets better.

And I think we are very much in that position where it's like I think as amazing as these models are and I think lawyers get a lot of value, we still have so many use cases where it's just like the models can't draft these very complex like merger agreements and things like that.

they can't look at these massive re case law research corpuses or discovery corpuses or data rooms and like really understand them.

And so I think there's a bunch of room both for the foundation models to get better but also we're starting to think about like can you you know custom build these reasoning models given this kind of specific data you have. So I think there's huge room for improvement for the models.

What about uh what about the back office? There's a lot of costs uh that go into law firms that are that are non-legal work, operations, billing, you know, all all that. Uh is that something is that a focus at all for you?

Is that something you think about or is there just enough work uh to be done uh on the client side? Would you do like a legal billing product too? Yeah, I mean I think that's something we've thought about of you know partnering with the folks doing that in the future.

We've had law firms ask can you you know help us automate parts of how we do billing. We have a bunch of firms that just buy Harvey seats to do some of this back office work again. And then there's clients that have asked about, you know, can you help us build spend management, things like that.

But I think right now the focus is still kind of building the product for for lawyers for law firms. Uh we have a question from the chat. Is the company named after Harvey Spectre, the fictional lawyer from Suits, or Harvey Dent, the Batman villain who I think also had a law degree? Uh Harvey Spectre. Harvey Spectre.

There we go. Makes sense. It's a great it's a great character. Um, I want to talk about costs. Obviously, tons of revenue coming in the door. Also, we're seeing this weird phenomenon. Maybe it's not that weird, but like with test time inference reasoning, you want to put, you know, GBT, what was the ARGI?

They were putting $2,000 of inference behind every task just to crack that puzzle. If I'm, you know, coming to you and I want, you know, to really make sure that the AI is checking its work fully, I'm probably want to put a ton of tokens behind that. Um, what does the cost structure look like?

How does that evolve over time? Um, what is what are the relevant changes that could happen and changes in the landscape? We saw the open source model come out from OpenAI. At the same time, even if you're running even if you switch to open source today, you still have to pay for energy and GPUs and stuff.

So what yeah how does the cost side of the business evolve over the next couple years? Yeah, I think our bet kind of from early on has been focused much more on like capabilities. These models are going to get cheaper as you scale get economies of scale you can kind of solve these problems.

I would say right now our cost structure is similar to um you know if you're using like a chat GBT where you know that gets amvertised over the task you have some tasks that are low compute some that are a lot.

I think legal is actually one of the very unique domains where you have these tasks where I want to draft a motion and typically this is going to cost $250,000. I want to do document review and this can cost a million dollars.

And so you have these specific problems where you can actually just run models over every single document, throw just a ton of compute at this. And so we're starting to kind of explore like what do those different models look like if you're doing tasks like that.

Um but yeah, lessons from Atrium, lessons from Clear Spire, lessons from binding a actual law firm to a technology company. Seems like it hasn't doing that certainly hasn't hurt you, but uh did you know that I'm sure I'm sure you had a lot of investors that don't why don't you sell the work instead of the software?

Exactly. We we actually we actually had the opposite.

Like when Winston and I started the company, we talked to the founders of Atrium and like 30 of the folks that worked at Atrium and actually Jason Quan who at the time was the GC of OpenAI was at Y Combinator when they did Atrium and Sam was as well and so we kind of like mentioned that and they were like you guys should focus on building tech business and for us it seems like the scalable play here is like we don't want to build a professional service firm ourselves we want to enable every law firm every professional service firm to kind of become the next generation um of firm.

Yeah, it makes so much sense. Microsoft Excel, their team didn't build an investment bank and yet it's used by every investment bank and it's and that will continue for a long time. Anyway, I think we have to jump to the next guest. Jordan, thank you so much for coming on. This was really fun.

Congratulations on all the growth. Huge. We'll talk to you soon. Bye.