E50 - Inside Harvey with Winston Weinberg, Founder & CEO of Harvey

We have made it to episode 50, and we've got a very special guest. Harvey Founder and CEO Winston Weinberg. Hey there, I'm Zach Abramowitz and I am legally disrupted. And here is my hot disruptive legal take on everyone's most favorite discussion and legal. Harvey. Now, there has never really been a company quite like Harvey. Attention is all you need and Harvey has been the king of getting attention. In fact, not only has Harvey got an attention for itself as a brand, it has put a spotlight on the entire legal software vertical. The Harvey tailwinds have directly and indirectly led to other legal startups launching and raising capital, very often from VCs who have FOMO over missing out on Harvey. Now, I realized this early on because anytime I wrote a newsletter or published a podcast that had Harvey in the title, it would get like double the eyeballs. Silicon Valley was never interested in legal. But Harvey has broken so much of the VC dogma about legal tech that funding has exploded, not just for it, but for competitors like Legora, a catalytic, Solve, Centauri, Spellbook, Sandstone, a whole slew of others. At Legal Innovators London in 2023, I told the crowd of entrepreneurs they should get down on their hands and knees and thank Harvey for creating this kind of momentum. I've been in legal tech for a while, this didn't always exist. Now, to be clear, I am not a Harvey come lately. I have been predicting that LLMs were going to change everything in legal since December of 2022. And for as long as I've been writing and podcasting about LLMs, I have also been doing the same about Harvey. Back in February 2023, when A&O Sherman announced it was rolling Harvey out across the firm, I wrote at the time that this did not feel like the typical law firm innovation press release. The scale was different, the conviction was different, and more importantly, the implications were different. This wasn't just another legal point solution. It felt like the beginning of a shift in how the legal work itself was going to be performed, and this was going to be across verticals. This wasn't like a tool for contracts or a tool for discovery. It was sort of one ring to rule them all. Now, a few days later, I wrote a second newsletter doubling down, arguing that legal was going to be one of the first industries that was truly transformed by generative AI. My thinking was simple. If AI can read, understand, analyze, issue spot, and draft responsive documents, then it starts to look a lot less like software and more like a kind of scalable digital lawyer. And once that happened, the implications were going to be enormous. Now, why am I bringing up my previous predictions? Well, partly because I enjoy flexing on my correct predictions. But mostly because Harvey has become the company that everyone in legal tech wants to debate. Is it the future of legal work? Is it overhyped? Is the vertical AI application layer real? Is Harvey going to be worth a trillion dollars? Are going to be crushed and go bust because of insert narrative here? Now, nearly four years later, let's be fair. The only evidence we have is that Harvey keeps growing. And yes, I have serious FOMO because despite seeing this trend early, I didn't invest. I didn't even try. Now, to be fair, I still think there are real questions facing vertical AI companies like Harvey. But the question I always ask the skeptics is, did you predict that Harvey would get this far, this fast? So I sat down with Harvey Founder and CEO Winston Weinberg to go deep on all of it. We went for over an hour. So buckle up in the conversation, absolutely delivered. This is episode 50. And we are diving headfirst into everyone's favorite legal AI debate. Is Harvey building the future of legal services? Or is there some existential threat waiting around the corner? So without further ado, let's go deep with Winston. Let's get disrupted. Before I ask you any questions, I feel like I need to say something at the beginning of this podcast. I'm not an investor in Harvey and/or LaGoura. And the reason I say this is because I feel like no matter what I write about either of you guys or either of the companies out there, the first question I always get is, are you quietly an investor in Harvey? And I'm like, you guys don't know me. If I were an investor in Harvey, I would be flexing on it a lot. And I look at the fact that I didn't meet you guys or hear about you guys early enough to invest as something like a little of a hole in my process. So I'm not an investor in either company. I obviously admire the execution of both. But I kind of wanted to start with that just in the background. You guys have had an extraordinary run here. Like three and a half years later. And I guess you guys are turning four now? It's like August 15th, I think, or mid-August, yeah. I saw some cookie with your face on it last year. That was great. And I said to someone, I was like, this is swag that I need. I need the Winston Weinberg face on the cookie. We give bobble heads to folks that have been here for basically a certain amount of time. And so that was our first three-year bobble heads and then now every employee gets bobble heads once they've been here for three years. And then we put it on like cookie, like a face on cookies and bobble heads and all that stuff. So it's amazing. So I've listened to you in a bunch of different interviews. And you're 30. You're 30, 31. - There you go. - Yeah, 31. And this is your first startup. And it doesn't feel based on the interviews I've heard like you ever necessarily intended to start with startup. And you feel to me like very, very naturally good at this. But I'm wondering have there been specific people, VCs or other who have tutored you? Because it feels like you have gained like a knowledge that would have taken others, you know, maybe a lifetime of running startups. And it feels like either you're just naturally good at this or someone's been working with you. - Yeah, thank you. I think like, to be honest, we have a lot of like amazing investors and they've been super helpful like across the board. And I think like I've looked up to a lot of them. I've also looked up to a lot of people I've never met. So like I know you had like Keith on your call recently. And I've like looked up to him a bunch and I've actually never met him in person at all. And so I do think like there are a lot of people like that I've looked up to like Brian Haligan is someone I've looked up to a bunch. And he is awesome and I have met him in person. There's also a decent amount of people. Michael Dell is incredible and I have met him in person. So there are like some fantastic people that I've learned from. But I think to be honest, like you can learn a lot from just watching other folks that you massively respect who have built like massive companies or done other things that are super impressive. I also think to be honest that the way, the main way you learn is you just fail like a lot. And I think from the outside, it looks like we've been had like an crazy, crazy run. But like trust me, there have been, I mean, so many periods of the company is like oh, like oh my God, it seems like it's gonna be over. Like that happens all the time. And I'm glad you mentioned that because you know, I've written about this and I think this has been you know on social that there's been like a lot of weird hate directed at Harvey. And I think that one of the reasons that that's the case, and you can tell me if you agree, is I think there's this impression that like they never really suffered. It's like just always gone like hockey stick growth. It's only gone one way. You know, it's where the growth that took other companies, you know, it took like net documents 20 years to hit a billion dollars in valuation. And you know, you guys are already at 11 and I hear you guys are working on them around at 16 billion. So, you know, it does feel I think probably to the outside world like they've only had it good. Can you tell me about like some of the worst days at Harvey because I know he was starting to start up. It's like it tends to be a roller coaster. And yeah, and things can like tend to just look horrible, you know, even for you know weeks at a time. - Yeah, the main problem that we run into just again and again and again. And luckily, so it's the same problem, but the fix is different each time, which is good because if it's this, if it's like basically the same problem and then the fix is the same every time that means you have not improved or learned in any way shape or form. The problem every single time is in scaling, like literally every single time and it's something to do with scaling. So I'll give you an example of this. Last year around this time maybe a month earlier in like April, we had like a sub zero on our platform. Like the platform went down for like an hour every other day probably, like something like that. Like that's how bad it was. And the main reason was we had this massive explosion and growth, like A, for people using the platform, B, we had released Vault, which was basically like, you can ask, you know, over 100,000 documents, 20 questions or 50 questions, things like that. And people were uploading like a terabyte of data. Literally. And our system like absolutely could not handle that and we didn't put caps on it, right? And so there are, you know, there's been so many periods when that happens. You probably were like more concerned with people like not adopting and probably could forecast. I listen, we can release these products because we know like it's already hard enough to get people to use the product in the first place in a major law firm. We're never gonna run into scaling, bro. - Yeah, oh, 100%. And I think like, but I'll give you an example of how this has helped, where last year because of that, like we're hired to CTO and he has a good insane infrastructure background and a lot of the principal engineers that he's hired are all infra right so like helping with scale. Now that we've switched to cloud agents are usage is basically doubling month over month like literally month over month and in terms of like documents uploaded time spent in platform everything like that and we're not completely crashing and we've definitely like hit some problems earlier this year. But my God is probably in the next like three weeks or something like that we'll be able to support about a hundred times the usage that we have right now right and so I think that this stuff you know it takes maybe six months or a year for this to actually pay dividends but you really the way that you improve so quickly and you learn is you just you go through the process that a company normally would go through it in 10 years like every six months like that amount of switching. And then you just mess up absolutely everything and it just all collapses and then you learn from that so that in the future next time you grow you know double or whatever you're set up for that and I'm using the infrastructure as an example because that's like top of mind right now. But it's happened in everything it's happening like how we design our product how we do our roadmap it's happened in how we do sales how we do go to market how we do post sales all of these things and I think the the thing that people don't see is just how many times the scale breaks like just completely breaks and then you have to fix it and I think part of that reason people don't see it is because. Like when you're running the company you see scale breaking six months before it starts breaking right and so you are having like a panic attack like I'm having like a panic attack about like six months from now and nobody else sees that because it's six months from now right and so I think that's why from the outside it looks like great but from the inside you're always. You're like you're like you're like the you're like the parent who is like seen the child injure themselves on the playground right and so like when you're there you're like oh gosh I gotta like I gotta like be like a helicopter parent you know around around this kid and that's what you're looking at you know six months down the line. And some of it honestly like the narrative stuff is also weirdly helpful for a company so I would like the reliability problems that we infrastructure problems that we faced early last year are going to pay mass amount of dividends this year I'll give you another example of I think what's going to pay dividends there has been a tweet about Harvey going to zero every six months that has like a million impressions for the past three years I think. Like literally I would say if six weeks go by and there isn't one of them I'm like nervous like I'm like what is going on and I actually can't think of a period of time when it hasn't been about a month and a half maybe a month every month or something like that right. So I like to say to people when they ask because I get this question a lot like you know is Harvey going to go bust it's like for the weird binary but thing I always say is like I can look at this question like on paper. In terms of like the shrinking cost of software development but I don't know like I don't want to bet against you guys even you know even in bet even in terms of like giving a public opinion not like an actual like you know I don't know how to short Harvey I'm not sure there's a mechanism for that but you know I I do think that there's that that is something that that people think is like a real possibility but I always ask people like listen if you have this prediction that Harvey is going to go bust just just you did you predict three and a half years ago that they would get here right if you did like if you have like a live prediction that says I think that Harvey is going to get you know hundreds of million in AR are get 50% of the top am law firms get to us you know whatever the valuation you guys are currently raising I'll say 11 billion valuation where you are right now if you predicted all that and you also have predicted that Harvey is going to go to zero then I'm like more interested but I find it like to hire that person. Right but I've found typically that that's not the case is that usually when when people have this like bearish attitude either towards AI in general or towards like Harvey in particular it's not a person who also predicted the rise of Harvey and I think that's like where predictions are important you know you talk about learning I the reason I feel like predictions are so critical is look back and figure out what you got right but also it's a great mechanism to figure out like what you got wrong. So totally agree but I agree with all of that and the point is actually making was if you go through that a lot your team gets really strong like your this doesn't bother your team as much right and so like the first time that happens your team freaks out and then the company does well and they're fine and then the second time same thing right and it just keeps happening and at this point it's happened so many times that I'd actually think it's a massive advantage that we have where if the more the outside world says oh this company is in trouble the more our team just goes let's prove them wrong like we've done every six weeks for X amount of years and I think that that becomes more important because a lot of the the Harvey is going to zero has been like really silly like it's like all over the place with with levels of silliness I put it like as great like very very when I saw someone X the other day who said that there are Harvey's top hater but they're about to get a product demo finally and I was like what do you mean you're the top hater and you've never even seen a demo like that is like the super yeah that's really common especially there's a lot of folks that are like plot does this better than Harvey or stuff like that and they're not a customer they've never had a demo or or tried the product but I think the more important piece I'm trying to say is there will be times in the company history like there will be other really hard times for the company it will happen like there's no universe in which like we're past the hard times and my point is I would prefer that we have spent four years or whatever amount of time five years six years seven years eight years nine years ten years with people saying that things are going to go wrong like and that happening and then the company powering through and it then and that's proving them wrong I would much prefer I've had built a team of four years of people that have been doing that then a team that has never experienced that before and then things do go wrong and there's no confidence right like that I think is really really important and I think it's underrated in terms of you you have a team like has the learned experience of like having seen the Harvey going to zero narrative and send like yeah but but it didn't and that's really powerful and I like I think that's something you know as we compete more with with the model providers I think that's our actually our biggest advantage is a company competing against the model providers or compared to maybe other companies as well is we have been battle tested in a lot of instances and I think that the company feels very confident internally when we get to the next battle right and I wouldn't trade that for the world like if anything I wish that it happened more times like I wish there were more competitors earlier on I wish the labs had come out there us earlier like you got to build this up earlier because every stage it's harder right like every time your company gets bigger the threats are way stronger way more powerful more targets on your back yeah 100% and if you face none of those you're going to have a team that doesn't have a backbone and I think that the backbone is really really important for the next couple of years how is your like day at Harvey change like since you started the company you seem like you've been very involved in both like the brand side and the sales although you know some of the things you sent me recently it feels like you know maybe getting a little bit more into the weeds with the tech these days but has your sort of routine at the company changed over time is it fundamentally different yeah I think it's changed a lot I'd say in the beginning like the way that Gabe and I ran the company is Gabe basically ran engineering and then I ran it like basically everything else and that was kind of how he did things and so I spent a lot of time on GTM and things like that I'd say in 2025 I switched to about I'd say it was like 60% product and 40% GTM something like that and I'd say starting in November of last year it's been 95% product like basically the only thing I've been doing is engineering product and design there's a couple reasons for that one is I think that we're getting to like a much different scale as a company and I think that there are a lot of product decisions that really really matter because the underlying tech is changing just so much more the second reason is I think that we are in a the past four months have been a transition where if your company does not go through this transition as a startup like I think you're just done and I think most software companies are done if they do not do this and that transition is basically moving everything to cloud agents like literally your entire company to cloud agents infrastructure and the way the main way to think about the Delta here is the old way to do this was basically just API calls right and what you do is the you know the user would type something in and then you'd either call the model directly and you know create an output or with retrieval you basically like say actually use all this data first and then pull call the API right and then you could do a tool use to but the tool use wasn't super sophisticated and things like that but those are basically like the three options The difference that we're moving towards is you create the environment with all of those things, with memory, context, basically everything, all of your data. And you basically throw a coding model into that, and you pull all of the pieces that are necessary for getting the task done. And the delta in this is like astronomical. To the point where, I mean, the best way to say this is like, we started working on this about December, like mid-December, 'cause the coding company started doing it, and we started realizing how important this was. And the labs at first, and then the coding companies, like a certain cognition in those folks at it for after them. And it's getting to the point where again, like our usage now is starting to literally double month of a month, like the amount of documents uploaded, created all of these things, time spent in the platform. And there are other companies that have done this, adoption is the same thing. And so I think right now, if you're a founder and you are not spending almost your entire time with the company making this transition, I think you're in for a problem. - And what you're describing sounds to me a little bit like what they're talking about with harness engineering. - Yeah. - And the harness itself, can you explain that for, you know, you sort of how do you see that? - Yeah. - And what the harness means for Harvey, because the way I sort of explain it to people is, you're not talking about model improvement itself, the model could sort of stay as it is. But everything around it, what it connects to the tools that it uses, you know, that's, and again, that's going to be something where a company whose, you know, understands the legal workflow can really make a big impact and do something. - The way that you're describing it is like perfect. Like the best way to think about this is just, you can get, like here's the IQ of the model, right? And then here, like the steps that it goes through to reach a decision, right? Or like all the different tools, contacts, everything that it uses, we're getting to the point where optimizing this can make it so that a model that is less good than the frontier model actually performs at the same quality, but it's way less expensive, right? So that's number one. The second thing we're getting to is open source, it is getting very good. And there's a different, a couple different ways that people are using open source and post-training to make things really good. One simple way is that they're actually making their classifier based off of an open source model that has been post-trained. The classifier is just a routing system. It's like the orchestration layer. So it's basically based off of this user, what are they trying to do? And then route to the different harness or to the different model and things like that. That part of the product is actually really expensive. And so you can reduce a lot of the latency and things like that by just actually having like a classifier that's been post-trained. And then people are actually starting to just post-training on this, right? So like cursor has done a bunch of this with regular like composer two and things like that. And my strong sense of what's going to happen is I think that every industry is going to have a bunch of post-trained models. And I know like we got tons of flak for doing this in the beginning. - Oh no, I'm so glad you brought this up because a lot of these things are cyclical. And one of the questions that we have been discussing internally on our company threads is when will fine tuning ultimately make a comeback, right? 'Cause I, the fact that you guys, and I know this is almost like a LaGora talking point at this point about like the mass amount of time you guys spent with like fine tuning. But there were a lot of, you know, there was the Bloomberg GPT that fine tuned and then 4.0, you know, it was just better at all financial tasks and fine tuning kind of went out of style. Do you see it coming back in? - Yeah, for a couple of reasons. One of the biggest problems with legal is there's no e-vails. Like the e-vails are just atrocious. Like, and I don't know if you saw that we basically produced something called lab, legal agent bench. And the difference is-- - Which could be like one of the killer feature coming up, meaning like that. The e-vails are not just sort of like the side part of the product, but actually like-- - No, core product. - Or massive product. - I think they're gonna be core to the product. And we open source these so everyone can look at them and we're gonna continue to open source all of these. But basically the problem was you never had e-vails that were something like automated fund formation. Like there's just no way to create an e-vail for that. The main reason is you can't train on customer data, right? And so one of the things that we started doing is you can get some customers that are interested in doing this and you can train on their data in some instances, but then you can't give it to anyone else, right? But the second way you can do this is you can create synthetic data sets using the coding models. It's really expensive and it's hard to do, but you can basically create like a fake data room, right? And that is like the way that we're starting to go. And we're going to every single use case across long. Like we're just trying to do all of them. And we're benchmarking all of them right now with like Anthropic, OpenAI, NVIDIA, Google, like everyone's being involved in this. And our sense is what's gonna happen is these benchmarks are not saturated at all. Like they're gonna be at like some of these use cases are gonna be like 2%, like it is like not gonna work at all. Which is great for the legal industry, right? I think the second thing that's happening is like let's ignore performance. Let's just talk about cost, right? These models are getting to like frontier models are increasing at like a massive, massive rate. Like it's about like 20 to 30% increase in price in each launch. And we're getting to the point where the launches are gonna be every six weeks from Anthropic and OpenAI. That's like about the launch cadence. So you can do the math on how it's done. So that's gonna get, right? And then these agents, especially if you're using cloud agents, they're costing like an crazy amount of compute. And so what's gonna end up happening is like you do not want mythos or GPT-6 or whatever to do every single part of your company or law firm. Like you don't want that because that will cost more than the human labor. Like that will literally cost more than the human's doing it. Like I think there is an article recently that like Uber ran out of their entire token budget entire thing in a month. And it was like a year or 60 months or something like that budget and they just ran out. And so I don't know how long if you worked long enough at a law at, you were no melvety, right? I think if you worked there long enough to have this experience, but I know that when I was at Shulti, you know, we were very often trained like how to use Lexus and West law in a way where it wouldn't be too expensive. 'Cause if you would do these searches at the time and that wasn't a compute issue, that was just like their business model. But it feels like we've come full circle to that right now. - Yeah, it's like you had to do Boolean search so you just got outputs because if you put like a natural language search in, it would cost more or something. It was something like that I feel like I remember. - It was just that every single search you did cost money. So you had to like learn how to search in a way that was most likely to get you to the answer that you needed without doing too many searches, right? - Yeah, you do like filtering first and then you did a search. - Yeah, it was a deeply, deeply painful process. But yeah, but like I was just talking to a client about this the other day and they were telling me about a product they were thinking about demoing and this is a client who's been very good about like, hey, like we want to know everything. We want to know every product that's out there and I said to them, I'm like, you know, try it out by all means, but I said, I have to tell you something. Like you're gonna have to start cutting back because these bills are going to start like piling up and I think that this is like the, you know, you talk about like the six, the thing that's gonna be, you know, six months ahead. I have to think about this also because very often I get like I got asked to speak this year at Iltecon, Europe and they asked me for the title of my talk in six months from now and like, like how am I supposed to know? So I always joke that the best thing to do is just to say like, is there an AI bubble? That'll be your talk. - Yeah. - You know that like, no, how amazing. - No matter what someone will be discussing it. - Yeah, no matter how amazing AI is, someone's gonna be out there saying, oh, but are we in a bubble, right? So it's like a, it's one of those evergreen topics, but yeah, it's very, but I do think that one of the things that we're gonna be talking about a lot more six months from now is like, how do we like mitigate the cost? And I think you could start to see at that point firms potentially jettison some of the pilots they've been doing with certain companies and like holding on to others. And I do think that there's this, you know, for you to be able to figure out how people can use Harvey in a way that's like also not cost prohibitive is gonna be critical. I think there's nothing interesting about, you know, the fine tuning and the harness and everything goes with it. I heard you, I wanted to say this was already in like in 2024. Daniel Gross did an interview with Ben Thompson from Stratekery. - Yeah. - And he made the point that even right now, if the models stopped improving, then we would like, it would still be massively disruptive and be like massive progress. And I've thought this for a while that actually like the continuous model improvement probably hinders adoption because people feel like, hey, I don't know what I'm gonna be split, what I'm supposed to do six months from now because the things could change so much. Have you seen that as well? - Yeah, I think that's one of the limiters and then I think the other massive limiter that we're gonna have this year is everything is measured on individual productivity and that's gonna be a huge problem because like you need institutional productivity. - Right. - So like the problem with this. - Stuff that shows up on a P and L. - Yeah, 'cause the problem is basically, like everyone is using these things for different things and they're not coordinating with each other and then it creates actually more coordination costs, right? And then it's like anything you're actually getting from individual productivity could be worse. 'Cause you're basically like, you know, like you need everyone to row really fast and in the same direction. making people row 10 times faster, but now they garden more separate directions that doesn't necessarily increase the velocity. And so I think that's going to be important. But I think like even step out of legal, like something that Gabe and I talk a lot about is the problem that the billable ourselves is weirdly the problem that the entire world economy I think is going to end up in, which is, and yeah, that sounds a little grandiose, but I think it's true in a way. Like, okay, say you have a token budget, right? And you're a large enterprise and you spend a hundred million dollars on tokens a year, something like that, right? That was your plan. And then all of the sudden, it like you spend a hundred million in a month on tokens. What becomes the number one thing you need to figure out? It's ROI from basically token allocation. Like this becomes really, really hard. It's like, oh, I'm spending this amount of money on intelligence. What is the ROI of that intelligence? Which in a way is actually what the billable hour does. Like the billable hour basically says, I'm going to pay you three thousand an hour, but I want to know what the ROI is on your intelligence for that hour, right? That's like basically what you're doing. And I think that this is going to be a massive problem across the entire economy where people are going to have to figure out, okay, this team is spending this amount of money on tokens. That's like more than actually we're paying the team, right? What is the ROI of that? And I think this is where vertical companies are going to have a massive advantage because with all these evils and everything else, you can get to the point where if you create these products that help a whole team work, right? You can start figuring out, okay, this is where what is the ROI of that intelligence? Yeah, exactly. And so it's interesting where it's like we've been talking so much about the death of the billable hour. And it's like, hey, I don't think that's going to happen in legal anytime soon. I agree with you on this and be in a way of the problem that the billable hour is trying to solve is actually the problem that I think is going to be across the entire every single industry when tokens are being used and consumed at this massive scale. So this is super interesting for a bunch of reasons. First of all, there's this amazing study from Wharton at the end of 2025 that showed that legal, legal increase, it was the biggest increase in usage of AI from all the verticals, but they are the worst at measuring ROI. Now, I think this is like people might make a lawyer joke here, but actually it's kind of hard to measure the ROI with legal teams because the legal team has never been measured in efficiency. The whole point of hiring a legal team was like every nook, every cranny, make sure to take your time. This is where we can't afford screw ups. So it's not, it doesn't lend itself very nicely to ROI. And then the only kind of like, I think that lawyers don't necessarily have the greatest imagination when it comes to thinking about ROI. They tend to like kind of quickly go to like, this is mean I'm going to have to like lower my race. This is mean I can do a fixed fee. And if like those two things aren't satisfied, it's like, well, they kind of give up then. And I think that what has happened over the course of the last year, and I've been talking about this publicly. So I think, I probably influenced it somewhat, but I think just in general, people are getting smarter about this is, first of all, realizing what you said is that like the the easiest ROI to measure is the individual productivity. And if someone, and like the way I do hear this from Harvey users, which is like, I couldn't possibly go back to doing it the old way, right? I'm not going to stop driving with ways, right? That's, you know, so, you know, or I've given the example like, what's the what's the ROI of two screens, right? Yeah. Now I should say, I wasn't one conference where I mentioned this and someone like came up to me afterwards, like nerd CIO is like, well, I want you to know we actually measured the ROI of two screens. I was like, all right, okay, but like, come on, you get like the overall, you get the overall point here. So I think you can have more Harvey tabs open. So infinite ROI. Yeah, absolutely, but I listen, and I think there's also it's tricky because you talking about usage and I want to get to this, this, this, the sass, the saster article about like Harvey usage, but I also sometimes think that like the, the, the value of the AI is sometimes just also even when I don't use it, but like knowing it's there, if I do need it, right, is, is to me like critical. So I think you have that problem generally speaking, trying to measure ROI in this field. And I think that one of the things, and I'm sure you guys are thinking about this a lot in terms of how you're advising firms. I'm certainly thinking about it myself is like, what other ROI is meaningful? And like, how else will this like continue to show up? But the other part of it is I have actually been structuring more of my engagements recently, not on a billable hour, but on, on hour or like 90 minute sessions where I tell clients, I will, I'm going to spend this amount of time reviewing whatever it is you want me to review. And I'm also going to spend this specific amount of time with you kind of going over your key problems. And what I found is in a, in a world in which AI can like put the deliverables together like pretty quickly. Actually, my human judgment, my like, yeah, hour just focusing on you using whatever AI I'm going to use to review, to put, to synthesize, that that's exceptionally valuable. And I think that's what a lot of people are going to want from their lawyers moving forward is saying like, okay, no, I know you can like put together like the the motion, the brief, whatever. But like, I need then you super talented lawyer who's like not gotten less than an A since the eighth grade to just sit and focus on this for certain amount of time. How much time do you think you're going to dispense? I think I don't think that we're going to kill the billable hour. I think we might revive it in many ways. I think it'll be, I think that we will see higher billable hour rates. And then there'll be like per project, you'll see fixed fees for a bunch of parts of a project like a diligence or something like that. But my guess is the advice, like the advisory part, you can actually just charge more for than my guess. Like it's a repackaging, right? Like when you, when I, like we use a bunch of law firms and like we don't do everything internally with Harvey, we still use law firms. And if you think about it, it's kind of an interesting data point, isn't it? Yeah. Oh, 100%. And but we go to them for advice. Like there's basically like two types, right? Like there's like do the work for like professional services in law and legal, but I think broadly professional services, there's do the work you pay for like work to be done or you pay for advice. And normally actually when you're writing a check to a law firm, in your brain, 99% of it was for advice. Like as the consumer of legal services, my brain is actually 99% of it was advice. My, the bill is 99% work. Like that's the, that's like the case. You say the similar thing like in, you know, with consulting firms like McKinsey, right? Like you, you pay for their advice. It's just that like part of what you expect is for them to produce a bunch of fancy slides. But really if they would just give you, you know, the, the sort of the bottom line do this, do this, do this, that would also be like extremely valuable, probably more valuable than the slides. Exactly. Yeah, exactly. But I think going back to your ROI part, like the first thing that we need is just incredible evals and benchmarks across the industry was just having existed, right? And I think everything on the ROI side will like flow downwards from that. You, you talked before about like the model companies and thinking about the model companies as, as your primary competitor. I heard this very interesting analysis from Moore Shlomo as the founder of base 44. And he was getting the same kind of question and coding that you I'm sure you get a lot, you know, about legal work. And he said, from my perspective, as long as the model providers have to compete with one another, there will always be room for companies who are building on top. If one company pulls away, and actually this is a, nearly from earlier this year, and he said the company he was most concerned with was Google, like you see how much changes, you know, in just a matter of months, he said, if one company pulls away, then I think we are really in trouble. But I don't see that one company is going to pull away. Do you see things the same way? Yes. I also think that in these verticals because the models are getting better, you can scratch, you can actually like create a lot better vertical solutions than you could in the past. And so what I mean by that is like going back to these evals and post training models and things like that, we still haven't gotten to the point where any of these tools help a team do like an entire matter from start to finish. That doesn't exist. Like we're not there yet, right? And we have a bunch of the pieces for it and things like that, but it's not quite there. And that is the amazing piece that I think you can build and vehicle is you can get to the point where like I think the frontier system, call it a harness, call it models. I think it will be a combination of both, by the way. I don't think it'll just be harness. And then you're only using anthropic or open air Google models. I think it will actually end up being post trained open source models as well. The frontier system I think will be different per vertical. That's what I think will end up happening. And I think that that'll be because of performance, but I also think it'll be because of performance relative to cost. And you can think of this as like what the labs are going to, there will always be I think a need for frontier intelligence, right? But you do not want to use frontier intelligence on every single problem. you will bankrupt yourself. I mean, the best example of this is-- - Or have some without guard reels. - Yeah, yeah, exactly, exactly. But think of just even the cost per part where think of you doing, you're doing an M&A and you use a partner for change of control review. It's like 3,000 an hour. That's what you're doing when you're using, that's what these systems are doing, right? And so I think that that is a massive advantage for vertical companies if they can move on that very quickly. And then I also think that what's gonna end up happening in these super regulated verticals is the way that you build these systems is just completely different from like a security standpoint, infrastructure standpoint. And I think that's going to get really tough. If the labs put all of their resources on a vertical, yeah, I think they'd probably beat you. Like that's probably right. If they put a tent of their resources on it, a fifth of their resources on it, I don't know, right? And I think the domain expertise is so important here where a lot of what we're ending up doing as we move to this system that is for individual productivity to like how do we change how your firm operates? Is like we go in there with legal engineers in the different practice areas and we connect it to all of your data sets and all of these different things, right? That's gonna be hard for the labs to do. And again, I think they'll start doing that to some degree, but I would bet that the vertical companies can do this at a much, much larger scale and create a machine for it way faster than the labs can. - I do get that. The question very often should we go with Harvey? Should we? And I've dug in over the last few months to what are the killer features. Am I wrong to think that your killer features are your security data retention policies and SLAs? - I think it's important. I think the reality is like we're getting to a point where there's a bunch of features on our platform that would be very hard for the labs to do its scale. So I'll give you an example. Our vault product is getting to the point, and I talked about this earlier, where we're getting to a point where it's like a single customer will upload like a terabyte of data, right? And the amount of like processing power that you need from your infrastructure to just go through all of these documents and basically do legal queries over them is gonna get really, really high. And I think this is very different than like us using like a review table, right? On for like a thousand documents and you do something easy. Like I'm talking about I have a diligence for an entire discovery and I need to process all of the documents in that. That is going to be very, very, very hard for the labs to do at a massive scale. And I think that that's a pretty large advantage. I think the other thing is all of these platforms are turning into matter first, right? And what I mean by that is basically how do you take all of the context out that's relevant for your matter and operate over all of it simultaneously, right? And I think that's what shared spaces is gonna turn into, right? And in that as well, that's how you collaborate with your basically in house client, right? And you're doing both simultaneously. And so I do think that you could say that the killer features on the productivity layer, like those are all gonna get commoditized. And if I think if you're running a legal layer company and you think that the productivity layer is not going to get commoditized, I like don't know what you're building for. Like you're building for winning pilots and demos and you're gonna have a problem later on this year. The thing that gets really interesting is how do you actually operate matters internally at a firm and process all the documents with super high accuracy and not spending more money than a human on doing this. And I think we're gonna see these compute costs just go to the moon for this frontier intelligence. So basically like in one sentence, I think that in every single vertical, you can create a company or companies that create legal intelligence or frontier intelligence for that vertical. And I think you can help the law firms in our vertical create their own frontier intelligence. And I think that's gonna differentiate them, so them. And I think this is something that we tried to do in like 2024 and the systems weren't good enough, open source wasn't good enough. And this is something that I think is going to start to explain. So let me play like either law firm managing partner or CIO, whoever's gonna be evaluating the decision. So I understand, I think it was very, very smart for A&O Sherman to partner with you guys very early. But here I am looking at tools right now and how valuable is it for me to be like firm number 200 and whatever that subscribes to Harvey? I'm not getting in with you guys super early, I'm not gonna have that special relationship. How do you continue to justify for a firm that it's just as smart for you guys where it's the right move for me today, even though I'm not one of the sort of early adopters. - I think the question you're asking here is less about like how is it good for the users or the actual lawyers? You're asking more, how is it a differentiating factor? - Right, and I think that's sort of connected with the Claude question is like a lot of law, there are firms who just want to do what everyone else is doing but there are definitely law firms who want to be bespoke and have their own proprietary solutions. How can they get that with Harvey or does that mean they have to sort of like no, they're gonna have to build with Claude from ground up? - So I'll talk about how this works now and I'll talk about how I think this is gonna work in the future and like relatively soon. So right now we have, I think it's 170 legal engineers on our teams, it's a lot and we're hiring really fast and we have a transformation office as well. And so what we're doing at scale is kind of like Palantir-esque, right, of going in and helping you basically like set up how, what is an AI native M&A practice? What is an AI native X, Y, Z practice? And we're doing that also in different verticals. So we'll do it with banks, we're doing private equity, we'll do it with pharma, we're doing these other areas, right? That's super helpful and that's gonna continue and I think it'll get better and better over time. The thing I think this is going to transform into is earlier we were talking about how the harness really matters, right, and how there is basically going to be like each company is gonna have their own harness and things like that. I think each law firm is gonna have their own harness. And in a way, they have one today, like they do. They have like, this is how we do a transaction. This is how that partner does a transaction. This is how we train our associates. This is the type of data we use. This is how we format all our documents. This is how we do this step before this step. This is how we do market research, right? I think you are going to be able to scale these systems in the same way that we are like building our own harnesses, building our own cloud agents. You will have a way to do this, per firm. Like I actually think like the frontier law firms of the future are going to be honestly building like harnesses in conjunction with a Harvey or somebody else to help their firm. Like it becomes a system throughout the entire firm. It's like their infrastructure as well. And I think we will start partnering with a lot of firms on this and large corporates too for those groups. And you know, that looks like a combination of traditional like B2B software, right, like where's your system of record, like pulling all your data and a press search, all of those things. But combined with honestly these environments and open source and eVal systems that are thinking to be really, really interesting. And I think eVal is the, eVal to me is like, if I could say one thing that is like the theme of this year that's changed is you can start eVal-ing like in M&A, right? And you were never able to do that before. And that changes, I think, everything. That's a big change from your perspective. Massive, like it's the biggest difference between coding and legal. Like there's just no way to eVal whether this stuff is working and what the ROI is. And so it's vibes. It's usage stats. It's like, if I'm saying-- Can you turn like an M&A workflow into code work and you can figure out like, it doesn't work. Something like broke there and give it kind of like ID experience. Yeah. You can turn the entire matter space into coding. Like everything that is in a matter, you can turn into code. And you just basically, like the data room becomes an environment. That's like basically the best way to think about it, right? Or the discovery corpus, like those things would come in the environment. Yeah. A lot of the use cases, I think in the first few years of legal AI, have been like very focused on transactional. But yeah, we sort of saw this at the beginning of the year. There's going to be a move towards many more litigation use cases. And I've been excited about that because I've always thought litigation is actually a much better use case for the AI because it tends to be much more like strategic thinking, which creativity, which the LLMs are very good at. And sometimes you knew Asuka, an LLM to draft an NDA. And it's like, yeah, it did it. But I had a pretty good template to do an NDA beforehand. It's kind of like a more of a boring mundane use case. Can you talk about that a little bit in terms of how Harvey's being used in disputes right now and that entire transition? I can talk a little bit about what the problems were and why this took so long. So I think one of the pieces is most litigators think of litigation tech as case law research, and that's it. And so in the beginning, they don't think about that now. But I was a litigator. And when I pitched to all of my friends, and I showed Harvey for the first time to folks, the first thing every single litigator did was ask the case law research question. That's it. They didn't use it for drafting or anything like that. It was that. And so I think that a lot of litigators have been turned off from this technology from 2023. because they initially tried everything as case law research. And then they were like, this is obviously not good. Like, this is wrong, this is bad. And so I think like now that people are part of this is just a mental switch of like, oh, you can use this for discovery, you can use this for drafting, you can use this for, you know, high level research, even if it doesn't give you every single needle in the haystack case, things like that, right? So I think that's part of it. The other part is there's actually just like a bunch of technical blockers. So like, the best way to do to have done like discovery in the past, like a massive, massive corpus would probably be to like pre-index the corpus, which would take like a two weeks, basically. And then you could do search on top of it and then apply the models on top of that, which is like a terrible process for discovery. You don't have to like do that through each one. You basically like have to do what you do for enterprise search for like each discovery. Because the entire sort of EDR, you know, I don't know if folks are familiar with this process, seems, and I've written this before, seems entirely irrelevant in a world where AI exists. It's basically, it is created in a world in which the only people that would be able, the only way you'd be able to find certain files and documents is people slash, you know, Boolean search. And as a result, there's like an entire process that goes into that theoretically in a world of AI, you might do it entirely differently, but the problem, I think you're working up against there is that there's just this entire entrenched culture of the discovery teams who don't really have like an imagination for what things might look like in a future world, and I think you've seen, you saw a couple of discovery companies kind of start up at fail to launch and kind of like fail to really get off the ground. Yeah, like really serious traction. Yeah, I think the thing that's going to happen is people are going to use some of the old school tools plus Genai on top, which I think will get good and I think the biggest thing is just rag going away, like rag just is not going to solve discovery. And I think like a lot of people tried to use rag to solve discovery and it just it didn't work and I don't think it up a will. Going to cloud agents instead to do it this way will work quite well and but you need a combination of basically like filtering and then basically having the cloud agent run over top of a corpus and that I think will work better on just like massive scale of documents. And then I think the other problem that you need is you need a completely different UI for discovery, like I think this is really important and I think what you're actually going to start seeing is by matter types, you're going to have different UIs. The reason why is the review process is going to look different for each matter type right and so you want the user to actually have a review process that makes sense for discovery and then for makes sense for disclosure schedules or for makes sense for something else. And so I and that hasn't really mean like every like every discovery is a little bit different. So like whatever you spin up for that specific matter might be different. In other words, if you're reviewing one, you know, a case that might be a former case versus a case that's a, you know, antitrust or is that is that what you mean that each, each matter is different. Yeah, that'll, that is a step past this. I'm talking about just like like vault right now our current version of all you can a lot, it's being used tons in litigation and discovery, like a lot, right. But it actually the UI has not been built specifically for that, right. And we kind of have this UI that like works for a data room, it works for discovery, but it's not like perfect for all of them. And my point is you're going to want a different UI for discovery than you're going to want for every other type of part of a matter, right. And then yeah, you could potentially even transform it more for like types of discovery and things like that. But I'm just saying that first layer, right. And the main reason I'm saying this is imagine your discovery has like a million documents in it. The purpose of Harvey, if you're reviewing a million documents is how do you check that it's work is accurate. That's actually the purpose of the product at that point, right. Like that's the purpose. Like sure you can interrogate that set of documents and things like that. That's going to become important to, but even in that interrogation, really what is the point. How do you make sure this is right. Like that is what is going to be the most important like UX UI. And that is going to feel different for a discovery than it will for a data room and for different parts of M and A or things like that, right. And I think that is going to be a huge difference. And when we move towards this like matter first environment where the UI for different matter spaces for working in a different matter space is going to look different because you have to make sure it's easy for the human lawyer to review all of these pieces. Right. But both those things you discussed, you know, case law research and discovery, these are sort of old older processes that you would have done with some amount of tech and theoretically you can do them better with an LM. But it feels like there's now use cases that go beyond those two within litigation and things that people might have never used tech before that they are beginning to think about now. Yeah, I mean, one of the main ones I'm seeing some folks look at is like proactive internal investigations. So like basically, how do you run more internal investigations before you have an internal investigation, like starting to see that, which is like a huge use case for a lot of large corporates. And so, and it kind of like great Geven's paradox like, well, what if it wasn't a big deal to do like an internal investigation. Same with another one that has this is like, like we've been working on a horizon scanning product for a while and it's like that is another area where people are thinking like a hedge fund or a bank or something like that will think about doing different like higher level horizon scanning or in different countries that they normally wouldn't have because of like cost, right. So I think that there's a lot of stuff that is more like look around corners that I think is going to become super important. But if you talk to like chief legal officers, a lot of why that role exists is they become like look around corner like it's not as much like, oh, I'm practicing law or at least like when you think about the practice of law, it's like I'm ready all these documents and doing these things. They're a business partner in that board meeting being like, hey, we should consider this because in six months, this is going to be a problem, right. And I think the technology is moving more towards that. And the reason I'm using in house is the example is because whatever in house is doing means law firms are going to support them at doing it. And that's like the thing that I start seeing another example, and this is outside of discovery or litigation, obviously is private equity is looking at more companies like they'll look at like normally say they look at 10 a quarter. It's like, oh, wait, with AI, I can look at 20 now, right. But because it's expensive to look at these companies, right. And so I do think that there's going to be an incredible amount of explosion of just, oh, these are all these things that we couldn't do in the past. And now we're going to take them on. Yeah, and I think it's a great, you know, wait, when I tell lawyers to think about AI, I try to tell them, don't think as much about can AI do the thing that you've been doing up until now as well as you. It probably won't be able to like it, it'll look, it'll look cool, but ultimately you'll find some like net with it. Instead, try to focus more on things that you literally never even thought possible or reasonable. You know, the example I give is I've got law firms who, you know, care about what some of their peer firms are doing. And I've created searches that go every single day and read the news releases at your site, and it look go on several others, but they also go to like those specific firms and read like their press releases so that I can know, you know, if one of, if one of my clients, peer firms is doing something that knows going to be interesting. I'll flag it and say, hey, listen, you're probably going to get a question about X Y and Z tool, because I just saw like one of your, you know, top, you know, top competitors announced a pilot with them. Here's what you need to know. And again, those are things I never would have done because it just wasn't, it wasn't like a reasonable thing to go and read the news, you know, the press release of every single law firm and read every single legal news website. I don't like a Google search, but we've been limited. So I think it's just like thinking beyond what you can do, because then you also have like once you start doing work that is totally different than what you've might have been doing beforehand. You have a better incentive to try to automate the stuff you've been doing up until that point. Right. So I think that tends to, you know, to be a better way to think about it. When you were, you were, you were talking about the billable hour before, but another thing I think we tend to agree on is the value of junior associates. So I know that like again, whenever one of the main questions I get from every firm is should we stop hiring juniors. And I've heard firms kind of going in this direction. And I feel like, man, there's no way to like become irrelevant quicker than stop hiring youth. Right. Like I would not go in that direction. And I feel the opposite. I feel like had I had AI as a junior associate, I would have been able to get trained up really quickly. I would have I would have been like a senior attorney level six months in, but the training resources at the firm. And in most firms just weren't great. I mean, I joke all the time, but I was stuck on this deal, you know, the first deals I worked on. And I had not summer to the deals group. I had worked in litig and disputes. So I never thought I was going to be a deal lawyer. I had no idea what it was they did. And I got told, hey, you need to get the disclosure schedules, you know, all prepared for the signing tomorrow. And here I am at 3 a.m. in the office calling doc services. I mean, do you guys know what disclosure schedules are? I have no idea what they are. Did anyone, does anyone know what this is? And like, you know, with AI today, if I had Harvey, I'd be like, hey, what are disclosure schedules? What am I supposed to do? What does the partner want me to do? What are things I should be looking for? And I think there's this weird thing where we're firm. So like, how will we possibly train lawyers without AI? I'm like, no, the question is like, how did you ever train lawyers in the past without AI? So I agree with this so strongly. I also, you can just use non legal examples of this. Like, I know a bunch of junior associates that have been on client calls and, you know, their screens blacked out, so I don't have to say anything. But they've had like a pharmaceutical company. And they're talking, the pharmaceutical company is like talking about a drug. And they go into chat, GPT, and they're like going to like chat GPT pro thinking mode. And they go into like this super long explanation of this drug with all of like the press releases on and things like that. And then they ask questions in the chat that are related to that drug. And I've had multiple instances of like, the client ends up being like, wait, that junior associate, like knows more about my company than like some of the other people on this call, right? And it's, so I believe this, as you can tell, I'm very passionate about this. And a lot of it is because I was a junior associate like three years ago. And I think the, the way that I talked about this to firms is you should be more focused on time to partner. And what I mean by that is I think like from everything I've heard, like especially lawyers have been lawyers for a long time, like 40 years, 50 years, right? They told me that it used to be more of an apprenticeship model than this today. Like way more like it was, exactly, like way closer to that. And you know, like junior associates literally sit at like a partner's desk and they just sit there and absorb everything, right? And I think that firms, because there've been so much work, like what, what happened is the complexity at legal work got so high that they were just like, ah, we got to have higher armies of associates to do these things, right? And so I think they had to do this, but what happened with that is now you have this crazy leverage model, right? That's like one to 20 or something and saying like that. Yeah. And there's no time for any sort of like teaching or apprenticeship or there's some, but it's, it's hard, right? And I think that firms with AI are now going to have to be like, wait a second. We have partners that build like a crazy amount of money, right? But the junior associate works, some of it's getting commoditized. Well, then what is the goal of our firm? The goal of our firm is to make it so that we have as many and as best partners as possible. That's the goal. Like if I am running a firm and it is a business, what is the unit? Like what is the thing that we are providing? It's actually time to partner. It's like these, like if you have worked at our firm and you've grown up in our firm, you are this type of partner that represents our culture, that represents like our skill, that represents everything, that is my product. Like that's the product of the firm, right? And I think that if firms go back to this, we will also solve a lot of the problem that is happening in the legal industry for the past five years, which is just no loyalty and no allegiance to firms, right? Like you see this all the time. It's like a lateral partner leaves and every single one of the clientes with them. And it happens 24/7. Everyone's just like ping ponging back and forth through firms. And I think that if you go back to this apprenticeship model, it's like lawyers are human. They will remember like I have loyalty to this firm. I learned so much from this firm. They put so much effort into my professional development. And that will pay massive amount of dividends down the road. And I, but I think firms need to think about that. They need to stop doing like everyone has to spend 10 years doing this. And then they will maybe be partner. It's like, no, there's some people who deserve to be partner sooner. And you need to let them on that path sooner. And you put more effort into your training programs and you need to make sure that they're not blocked out on calls and all these things. So first of all, our mutual friend, David Fox, it told me this was a huge part of the, the secret sauce of Kirkland was, was building this kind of like loyalty and love with their attorneys. And he said, you know, people like don't want to believe that it's like that sort of simple, but it, but it really was. I think that, you know, when you, when you talk about this, you know, that this, um, this apprenticeship model and, and building, and building that kind of loyalty, there's also the, you know, people think about like always like using AI as like the measure of like, are we staying ahead? And I try to tell firms like, try to think also about like because AI, like what do you need to start doing strategically because AI now exists? And I think that one of the things that firms need to think about a lot, and I've said this out loud and shout out from the rooftops is, how are you going to keep your best talent? Because your best talent now is like, they don't necessarily need the offices that they don't like commuting to, right? And remote work has, has been like a kind of a mixed bag for the firms. Um, they don't necessarily need the army of associates to lever up a matter. Um, with AI, they can probably, you know, get a lot of the technology tools that they would need without a lot of the other, you know, sort of big firm infrastructure. And I think that you're, what you're talking about, and creating the ability to make partner at an earlier, um, at an earlier age or after a few or years is a, could, could be a massive, massive win for firms in a very smart way to like think about like, Hey, you know, once upon a time, this would have taken 10 years, but now with AI, we might start making people partners after three and four, you know, like why, why not? And I think that's like, we talk about where people will then say, okay, now you've given me a reason to stay at a big firm that I maybe didn't have before I definitely, I definitely think that's one of them. I 100% agree. And the reality is that's all that matters for software companies or AI companies as well. That's it. Like at the end of the day, it's like the main thing that matters is how good is your team and what is the rate of improvement of your team? Like how good is your team rate of improvement of your team? And are the quality of people entering your company better, the same or worse. And if it's the same as in like you, you know, you're like thinking that people aren't improving. So you have people internally that are improving at your company like this, but you aren't hiring better people to match the improvement of the people like your company, your screw. If it's decreasing, you're very much screwed. And so it should just be, are the people at your company internally improving as a team and as individuals like really, really fast? Is that quality going up? And are you matching that by hiring people that match that quality externally? And that's it. Nothing else really matters. Like I genuinely do not think that any of the most stuff we were talking about or anything like that actually really matters that much compared to that. You were, we were, I was talking about the very beginning of our conversation. Some of the like the hate that's been directed at Harvey and you made a good point to me recently, said like for the legal tech crowd in particular, like shouldn't they want to root for a kind of a legal specific company, as opposed to like the base models? Because ultimately, right, this is going to be us against the base models. What would you say to folks, you know, to like, to sort of combat some of this, like weird, shouting Freud or, or, or sometimes like Harvey's going to zero? Like, what would you like to just tell them? There's a lot of them listen to this podcast, you know, to sort of get them to like, think about Harvey, maybe a little bit differently. Yeah, I think like the, the main thing that I want to say is, I think like get to know the company first, like that would be the main thing. And, and by the way, like, if you've gotten to know the company and you've used the product, and you also have like a lot of hate towards us, totally fair. And I'd love to hear that feedback, like 100% would love to hear that feedback. I think in most instances, these are folks that have never interacted with the company or use the product. And in those, I think like the, the part of frustration that I think is actually super justified is we kind of, we got actually way ahead of our skis when we launched. Like we onboarded in a Sherman, when we had three people at the company. That was 4,000, 4,000 attorneys that were being onboarded. And we have literally three people. And by the way, the third person, including my co-founders, I just want a person. And he had joined, he joined three weeks earlier, right? So like, I think that one thing we did do incorrectly in the beginning. And we're trying to hopefully resolve it or like helped resolve it in the past like year or so. And we're trying to do more is like build with the community and build in the open. And we could, we should have done more of that. And I'm not using any Sherman as like an excuse, but I do think people don't understand the perception. Oh, they don't understand that like part of this was we've supported one firm and we had no people. And then everyone, all the firms reached out to us and like we couldn't hire fast enough. Like it was impossible to actually support and do all the podcasts, do all the legal tech events and do all those things. So I don't regret it those first like couple of months. But I think like six months after once we had like built up more, I think we, I would have should have done more. If you've never worked at a startup, you would have looked at this and been like, oh, they're being super stealthy and they don't, you know, they're, they're like two. good for us. And I'm like, I looked at this and I'm like, they probably have been off like way more than they can do. They're like heads down, just trying to like put out fires everywhere. And I think that that's that's off that was that was very clear. The case I am putting up this this little bit of a flex because I did say it was a really big deal at the time. This firm doing a full rollout. I mean, it's it's wild to me in retrospect. You know, talk about legal being this like slow adopter right before like large language models launched. I was giving a talk at conferences, basically saying stop blaming lawyers for the non adoption of like tech products. It's it's the products. It's not the lawyers. And so like when I saw this come out, I'm like, wow, this is like a this is like a really like, you know, serious rollout and needs to be taken seriously. And certainly it was. I want to end with with a with a question that I think is like, you know, maybe a little bit of a of a tough question. Um, but if I was a law firm, one of the things I would be concerned about is how invested Silicon Valley is in in Harvey and how much of the company they own. And what promises have been made to Silicon Valley that this company is going to be worth, you know, not you're not promising, you know, even at an $11 billion valuation, which I think you're going to be passed quite soon. But like, let's say that's the case, right? You're you're effectively promising these firms that the company is going to be worth 50 billion, 100 billion, maybe more. What, you know, should should I as a firm be concerned, you know, at Silicon Valley's involvement? And how do you, how do you get someone like me comfortable in that, in that situation? Yeah, that's a good question. I'll start by saying something maybe my, you see, I think my BCB finally saying this. So I have full control over the company. Um, and so if there is any fear or like about that, it should be directed towards me, like 100%, like I take ownership for every single decision, I technically have in full control of the company. Like it is, it is all on, on me, if I am doing something that is not good for the legal industry, like it's, it should be directed towards me. Um, so I'll say that first, because I think people, there's a lot of rumors in the market that because we've raised a bunch of money and things like that. Um, but I have enough voting rights where I have full control of the company. Um, I imagine, I imagine some of you guys have, have taken some ships off the table. That would be, I would, if that's not the case, I would love to know that not really. Um, but, but I think the more important, the more important piece is, the, the more important piece, I think is like, we are not rampied of like VCs, like it is rampied by us, like we have full control over the company. It all do the decisions that we make or ours. Um, that doesn't mean I don't respect our investors and I massively do. But at the end of the day, like the burden of making a decision is entirely on us. Um, so I'd say that first, the second thing I would say is I think like a lot of folks are worried about like, oh, it's all labor displacement and massive VCs want and all these things. One thing I think we all have to remember is there is zero tech penetration and illegal. It's like, it's the lower than any industry I can ever find. Like I can't find, I've tried, I've tried to find lower industries, right? But it's like, the legal tech market is around 25 to 27 billion. Sometimes people say it's 30, right? Something around that range. The legal market's trillion. Yeah. There's no, like, I think the most recent report and like, and likely to get bigger. Yeah. And it was like, and likely to get played bigger, right? As the economy grows. And so the thing that, and I read a report recently, I think the most recent one, like most expensive one was from 2024. And it was like for law firms, spend 2.4% of their revenue on technology. It's just like nothing that's like completely irrelevant. And that includes like, Wi-Fi and stuff. Like this is like, it's insane. And so my point is, like, you don't need to pitch, like, labor displacement. All you need to pitch is say, Hey, look, tech tech is going to be a big, bigger part of this. Yeah, it's going to, or go from 2.5% to 5, like just to 5. That is a crazy amount, right? Of growth, insane amount of growth. And I think people don't think about it that way. Because there's so many articles that are like, ah, lawyer displacement, like all these things. But that's all you'd have to look at is just like, all that has to happen is quite literally the tech budget of the legal industry, both in house and law firms, just doubles. And it's actually not even that high of a percentage. It's like it gets to 5% and the legal industry also grows. Like, that's pretty much it, right? And so I think that there's just, there's a lot of fear. And I think there's a lot of fear in the whole world right now with how good this technology is getting. But that's how I would answer that question. Yeah, Winston, I could geek out with you for hours more. But I appreciate it. And look forward. By the way, folks should also follow you on X now because there's there's much more of a conversation these days about AI on X. I've noticed this is way more. I think I think that's sort of an interesting data point in its own right that sort of like the lawyers who may not have cared about this in the past are deeply starting to care. Obviously, it will be we're following following the progress of the company very closely. And thanks for joining. Yeah, thank you, man. Thank you for tuning in to Zach Abramowitz is legally disrupted. If you enjoyed today's episode, please consider leaving us a rating and a review on your favorite podcast platform. Don't forget to subscribe. See you never miss an episode.