The Exchanges

Q How do you figure out which cities to go into first? Is it like you have your customers and then you build around them?

A Yeah, so it's actually, um, yeah, so we did it first based off of just like reacting to like big customers, so like we'll sign like Deutsche Telekom and it's like, oh wow, we need an office in Germany, right? Things like that. Um, there's actually something really interesting That we had a problem of in the beginning, which is because we process sensitive data, a lot of the countries we actually needed like an Azure instance in each one, right? So like in Australia, you can't process financial data outside of the country. And so we had, we would set up these offices and then we'd set up like Azure instances too. Um, and it was almost like we set up an Azure instance and then that would be like a pretty good indicator that we'd have to set up an office pretty soon afterwards, just because like customer demand.

AI assessment note: “we did it first based off of just like reacting to like big customers”

Q And you've raised over a billion dollars now, so have you used most of that capital? Where is, are you using it on tokens? Like, what is the office?

A Yeah, now we're using it on tokens. Now we're using it on tokens. Um, no, we actually, we haven't used a lot of that money. Um, I think, like, the, the interesting thing that we always wanted to do was actually do a lot of post-training on the models, and there's a couple problems in legal that makes us, like, really hard. One is the data isn't available. So like, if you went online, you're like, I want to go find a bunch of documents that are related to like a random fund formation by Blackstone. They don't, they don't exist. Like you'd have to go to Blackstone for those documents, right? Um, but the thing that happened with like the last generation of coding models is you can actually take sets of documents and create synthetic docs that are so good that the lawyers can't tell the difference between whether they're created, you know, by an actual lawyer or they're created by the coding models. And so with that, we've now actually created basically like a pipeline for creating synthetic data sets across like every single legal use case. And because we have that now, now you can start actually plus training models, and that's gonna be expensive.

AI assessment note: “no, we actually, we haven't used a lot of that money.”

Q of how fast you're growing the company, I don't know, maybe you still have a billion dollars in the bank. How are you thinking about buying, or sorry, building out The company itself or buying. There's like a, now there's like an onslaught of M&A within all these AI companies and buying up smaller startups for talent and all this kind of increased competition. How are you thinking about that?

A Yeah, devalue, um, like higher value on team and lower value on what they've built in terms of like, I do not believe that it is a good idea right now to go around and buy legacy technology. Like, I don't. I think it is a much better idea to buy, like, really, really good teams, regardless of if they worked in your space. Doesn't matter, right? Um, and so that's, like, if you look at the aqua hires that we have done, they actually haven't been in the legal AI space or, like, legal tech. They've been outside of it, but they're really, really good teams that could work on a problem that we have, right? Um, and that doesn't mean that I won't do legal tech Acquisitions in the future. But I do think that right now, if you are making an acquisition, like, the number one thing you should be looking at is just talent. Because you can build things so much faster now, right? That it should literally just be talent. Like, are you buying a team that is really, really good, and are they going to align with your cultures? Because the other problem is, like, we're not even four years old. If we go and absorb a bunch of teams, like, our culture is still being built.

AI assessment note: “I think it is a much better idea to buy, like, really, really good teams”

Q any country in minutes, hire anyone anywhere, get visas handled fast, and get back to building. Visit deal.com slash sorcery. That's deel.com slash sorcery. So you started this company post COVID. It's nearly four years old, and it was AI native from the start. So I just want to know, like, what, what is like fundamentally different about building a company today in the AI era versus the SaaS cycle?

A Yeah. Um, I mean, I think pace is really important. And one thing with pace that I think people struggle with is like, You have to just assume that most things are two-way doors, and people really hate this. Like, they really like to kind of, like, think about a decision, get to, like, 90% certainty, and then make a decision. The reality is, like, you're probably gonna have to make a bunch of decisions at, like, 51% certainty, um, which means you're being wrong, right, right? And can you deal with being wrong and then quickly pivoting, right? And I think you have to do that way more than you used to in the past, A. B, I think there's a huge difference on enterprise, like, massive difference on enterprise, which is you cannot get away with, like, going in a room, building some product, and then selling it, and then never improving the product over the next, like, XYZ years. The users matter so much. Like, the alternative is Quad, ChatGPT, these other things, like, those are great products too. And so I think that the product bar for enterprise is astronomically higher than it used to be, where like you a hundred percent have to be constantly innovating and creating the best product for the end user. And I think in the past you could kind of get away with doing some things like really long sales cycles and things like that.

AI assessment note: “I think pace is really important... there's a huge difference on enterprise”

Q We're going to be interviewing Gabe after this. What question do you think I should ask him?

A Um, I think that the best question actually is why are benchmark, like, why are, why are the current benchmarks for most verticals bad, right? Um, and if you, if you look at, like, the benchmarks that have existed for legal for a long time, I mean, half of them are, like, can it pass the bar? Multiple choice questions on, like, community property law and things like that, right? And I think that we haven't actually until now had, here is a very, like, Very good set of data that doesn't a legal task from end to end, and we're missing this in most articles other than coding. Basically, coding is the only one that has like a good saturated benchmark.

AI assessment note: “I think that the best question actually is why are the current benchmarks for most verticals bad”

Q How do you compete with those large models that are offering competitive services? Like, what is the thing about Harvey that you cannot copy?

A Yeah. I mean, the, the best way to think about this is like, there's the product side, and then there's the intelligence or model side. And, and you could say on the product side, it's like the harness side. Um, I think that's like conflating things a little bit too much on the product side. You just go very vertical, right? So you build solutions that are really good at like diligence in a particular space, like all of those things. And I think it's gonna be hard for the labs to get to that level of specificity on the product side. On the model side, you actually do a similar thing, which is you basically build a bunch of models that are really good at specific legal tasks, and then you optimize them for cost, right? Like, there's a world in which GPT-X is more expensive than a lawyer. There's a, that's like a very possible world, right? And so where we have is basically you can think of like the frontier models here, commoditized models here. I think a lot of the economy is actually in between these things, right? Because the frontier model might be like too expensive to basically put in terms of every single piece of work. And so you have this massive space that is like all these vertical companies.

AI assessment note: “You just go very vertical, right? So you build solutions that are really good”