1 00:00:00,150 --> 00:00:02,880 Jordan, thanks for joining us this afternoon. 2 00:00:03,090 --> 00:00:04,019 It's good to have you on. 3 00:00:04,350 --> 00:00:05,460 It's great to be here, Ted. 4 00:00:05,670 --> 00:00:09,420 Thank you for inviting me to the preeminent legal Tech podcast. 5 00:00:09,930 --> 00:00:11,550 Yes, thank you very much. 6 00:00:11,640 --> 00:00:16,590 Uh, I think we're one of the few, if not maybe even the only, that is 7 00:00:16,590 --> 00:00:18,450 solely focused on legal innovation. 8 00:00:18,930 --> 00:00:23,820 So, um, which is a hot topic these days with the way AI 9 00:00:23,820 --> 00:00:25,020 is kind of shaking things up. 10 00:00:25,620 --> 00:00:25,950 Yeah. 11 00:00:28,275 --> 00:00:34,575 Um, yeah, I, I've been in the legal tech space for five years now, and I think 12 00:00:34,725 --> 00:00:38,985 it's really exciting to see that there's actually appetite for the industry now, 13 00:00:39,315 --> 00:00:43,935 because when we started in 2021, there, there weren't like legal tech podcasts. 14 00:00:43,935 --> 00:00:48,375 There wasn't, there wasn't an audience for, for that type of thing. 15 00:00:48,585 --> 00:00:52,695 So, yeah, it, you know, it's, it's funny, legal innovation. 16 00:00:53,385 --> 00:00:57,915 I joke and say, really used to be an oxymoron, like jumbo shrimp 17 00:00:57,915 --> 00:01:03,074 or military intelligence, but it, the market's coming around it. 18 00:01:03,074 --> 00:01:03,135 Um. 19 00:01:04,290 --> 00:01:09,300 You know, we, we jokingly would call the CINO role, which is Chief 20 00:01:09,300 --> 00:01:14,520 Innovation Officers Chief in name only, because so many in the early 21 00:01:14,520 --> 00:01:16,650 days were just not properly empowered. 22 00:01:16,710 --> 00:01:20,100 They had good people they would put in those roles, but they just 23 00:01:20,100 --> 00:01:22,470 wouldn't properly empower 'em. 24 00:01:22,710 --> 00:01:23,280 Um. 25 00:01:24,045 --> 00:01:26,835 You and I got connected because you, you wrote an article that I thought 26 00:01:26,835 --> 00:01:29,775 was, was really good and provocative. 27 00:01:30,345 --> 00:01:34,185 Uh, and I like when people push the boundaries a little bit 28 00:01:34,215 --> 00:01:35,865 and challenge the status quo. 29 00:01:36,405 --> 00:01:40,425 And your article was called Why Can't, 4.3 billion. 30 00:01:40,899 --> 00:01:45,640 In legal AI investment, outcompete $20 a month for chat CPT. 31 00:01:46,119 --> 00:01:49,780 So, um, we're gonna get into that, but before we do, why don't 32 00:01:49,780 --> 00:01:53,440 you tell us kind of who you are, what you do, and where you do it? 33 00:01:54,190 --> 00:01:54,670 Sure thing. 34 00:01:55,360 --> 00:01:56,289 So my name is Jordan. 35 00:01:56,350 --> 00:02:01,630 I'm a co-founder and CTO of version story and we are a startup 36 00:02:01,720 --> 00:02:03,610 building get for documents. 37 00:02:04,060 --> 00:02:06,729 So we're trying to solve legal version control. 38 00:02:07,125 --> 00:02:11,055 We're trying to make it so that you as a lawyer, always have confidence about 39 00:02:11,055 --> 00:02:12,405 which version you're working off of. 40 00:02:12,825 --> 00:02:16,515 You are never fearful that you're missing changes when you send a draft 41 00:02:16,515 --> 00:02:18,105 to a counterparty or to your client. 42 00:02:18,885 --> 00:02:25,755 And, uh, the inspiration for our product is Git, which is the defacto 43 00:02:25,755 --> 00:02:28,935 version control for how any software engineer in the world works. 44 00:02:29,715 --> 00:02:33,045 Uh, it's kind of the backbone upon which. 45 00:02:34,049 --> 00:02:38,160 Vibe coding, for example, can be built, is having reliable version control. 46 00:02:38,850 --> 00:02:41,100 Anyway, so that's what we work on. 47 00:02:41,190 --> 00:02:43,799 Uh, I am, like I said, I am the CTO. 48 00:02:44,310 --> 00:02:48,359 We've been working on version story for, uh, a little over five years now. 49 00:02:49,560 --> 00:02:54,329 And, um, we're based out of Brooklyn, New York, and yeah. 50 00:02:56,489 --> 00:02:57,059 Good stuff. 51 00:02:57,059 --> 00:02:57,299 Yeah. 52 00:02:57,299 --> 00:03:01,920 You know, it's interesting, GitHub, I, you know, I'm, I own a software 53 00:03:01,920 --> 00:03:03,690 business, so I've known about gi. 54 00:03:03,690 --> 00:03:08,790 I used to work for Microsoft 26 years ago, so I'm very familiar. 55 00:03:09,930 --> 00:03:14,250 With GitHub, but I feel like GitHub has really gone mainstream. 56 00:03:14,520 --> 00:03:18,269 You know, to use Claude Code, you need GitHub. 57 00:03:18,660 --> 00:03:23,430 Um, right, they, they've now released Cowork, which is 58 00:03:23,430 --> 00:03:25,290 literally like two days old. 59 00:03:25,290 --> 00:03:27,870 By the time this airs, it will have been out for a little while. 60 00:03:27,870 --> 00:03:29,490 But, yeah, it's funny, man. 61 00:03:29,490 --> 00:03:33,450 I see non-technical people now with repos on GitHub because 62 00:03:33,720 --> 00:03:35,070 they want to use cloud code. 63 00:03:35,070 --> 00:03:37,050 So GitHub is now. 64 00:03:37,350 --> 00:03:39,990 Kind of a household name, which is pretty cool. 65 00:03:40,710 --> 00:03:41,460 It is, yeah. 66 00:03:41,550 --> 00:03:48,240 Uh, and it's really powerful and it is ki it is essential to, to vibe coding. 67 00:03:48,330 --> 00:03:53,190 I you one kind of beginner mistake that I've seen from a few times 68 00:03:53,190 --> 00:03:57,240 from non-professional software engineers who dip their toes in. 69 00:03:57,885 --> 00:04:01,785 They don't know that they need to set up Git first and they end up 70 00:04:02,115 --> 00:04:06,225 having cloud code go in there and rewrite their whole code base, delete 71 00:04:06,225 --> 00:04:07,605 files, and now they can't go back. 72 00:04:08,115 --> 00:04:13,725 Um, and so it really is imperative to the vibe coding workflow. 73 00:04:14,445 --> 00:04:21,615 And so I think what we're building, uh, is going to similarly be imperative for truly 74 00:04:21,615 --> 00:04:24,885 unlocking AI empowered legal drafting. 75 00:04:25,680 --> 00:04:26,580 Very cool man. 76 00:04:27,270 --> 00:04:29,909 Well, let's talk a little bit about, uh, about the article. 77 00:04:29,909 --> 00:04:34,229 So again, the title was WA Can't, 4.3 billion in Legal AI Investment, 78 00:04:34,229 --> 00:04:37,320 outcompete Chachi PT for 20 bucks a month. 79 00:04:37,919 --> 00:04:42,240 And, um, you know, that's a very attention grabbing headline and I think 80 00:04:42,240 --> 00:04:50,159 it's one, it touches on a nerve that a lot of people have been, you know, uh, 81 00:04:50,190 --> 00:04:54,000 pontificating on, which is, you know, how. 82 00:04:54,405 --> 00:04:55,455 Where's the moat? 83 00:04:56,474 --> 00:04:59,805 Um, you're too young to remember, but in the eighties, may maybe 84 00:04:59,805 --> 00:05:03,735 you've, you've seen reruns of the commercial, uh, for Wendy's. 85 00:05:03,735 --> 00:05:06,224 There used to be an old lady and she'd say, where's the beef? 86 00:05:06,974 --> 00:05:10,995 Um, well, investors are going, where's the moat? 87 00:05:11,505 --> 00:05:14,835 Um, but they still keep, they keep writing checks at, right. 88 00:05:14,835 --> 00:05:18,585 You know, a DX valuations so clearly. 89 00:05:18,990 --> 00:05:22,620 They're willing to look past the absence of a moat, which I 90 00:05:22,830 --> 00:05:24,750 don't think really exists today. 91 00:05:25,110 --> 00:05:30,630 I think the moat is really distribution and install base, right? 92 00:05:31,170 --> 00:05:38,005 And, um, that's, and it's not a bad strategy because I think when you're 93 00:05:38,005 --> 00:05:41,100 first to market and market, if you can, if you can generate enough 94 00:05:41,100 --> 00:05:45,480 momentum and create a large enough install base, it gives you time. 95 00:05:46,140 --> 00:05:47,310 To figure out your mode. 96 00:05:47,460 --> 00:05:51,150 And a lot of businesses, some businesses do that before they 97 00:05:51,150 --> 00:05:52,650 even figure out how to make money. 98 00:05:53,040 --> 00:05:54,150 I mean, look at Facebook. 99 00:05:54,390 --> 00:05:57,150 Facebook didn't have a revenue model for years. 100 00:05:57,599 --> 00:06:01,409 Git like GitHub didn't have a revenue model. 101 00:06:01,890 --> 00:06:07,050 Um, it was really catering to a lot of open source, um, development projects. 102 00:06:07,080 --> 00:06:09,030 And, you know, over time. 103 00:06:09,570 --> 00:06:11,250 You, you can kind of figure that out. 104 00:06:11,250 --> 00:06:16,920 But tell us just a high level what the, what the premise and the 105 00:06:16,920 --> 00:06:19,350 motivation for, for the article was. 106 00:06:20,250 --> 00:06:20,790 Sure. 107 00:06:20,880 --> 00:06:27,630 So I'll start with the motivation and kind of what led me to write it. 108 00:06:28,830 --> 00:06:34,920 Uh, I think a lot of the ideas in the piece were stewing for quite a while, 109 00:06:35,100 --> 00:06:36,780 uh, but it kind of clicked for me. 110 00:06:37,575 --> 00:06:42,165 After having a few conversations with some friends of mine, these friends of 111 00:06:42,165 --> 00:06:45,075 mine are software engineers and founders. 112 00:06:45,344 --> 00:06:49,995 I would characterize the way they think as being fairly reflective of like the 113 00:06:49,995 --> 00:06:52,215 Silicon Valley mindset, if you will. 114 00:06:52,724 --> 00:06:54,344 I mean, they're, they're on tech Twitter. 115 00:06:54,795 --> 00:06:59,230 Um, so like, I kind of think of them as, um. 116 00:07:01,484 --> 00:07:05,354 Yeah, kind of a reflection of what, what Silicon Valley is, 117 00:07:05,414 --> 00:07:06,734 is thinking at any given time. 118 00:07:07,455 --> 00:07:13,094 And I noticed that over the course of like a couple months this past summer, 119 00:07:13,484 --> 00:07:16,664 I, I felt like I was having the same conversation over with a few of them 120 00:07:17,234 --> 00:07:24,075 where they were asking me like, how worried are you about like these, 121 00:07:24,554 --> 00:07:26,895 these big name legal AI companies are. 122 00:07:30,315 --> 00:07:32,085 Don't you need to be doing the same thing? 123 00:07:32,955 --> 00:07:41,085 Um, and I, I kind of felt like, well, I don't exactly think we 124 00:07:41,085 --> 00:07:43,515 are approaching the same problems. 125 00:07:43,664 --> 00:07:46,455 We're, we're not really trying to do the same thing. 126 00:07:47,145 --> 00:07:52,424 Uh, but the, the point that these friends kept reemphasizing is 127 00:07:52,424 --> 00:07:57,554 that in, in the AI world, all that matters is distribution and that. 128 00:07:58,350 --> 00:08:05,640 It's really a winner take all market where you need to have maximum 129 00:08:05,640 --> 00:08:08,430 distribution and product is secondary. 130 00:08:08,730 --> 00:08:11,880 That was like kind of the, the repeated idea. 131 00:08:11,880 --> 00:08:17,250 And so I, I kind of, um, was curious about this and I, and I was pushing 132 00:08:17,250 --> 00:08:22,410 back and trying to understand where they were coming from because disagreed. 133 00:08:24,825 --> 00:08:27,794 The way we built our company is starting with building product first, 134 00:08:27,794 --> 00:08:31,245 making sure it's really solid, and then figuring out distribution later. 135 00:08:32,595 --> 00:08:40,485 And the idea that kept resurfacing is that, um, it is a land grab 136 00:08:40,485 --> 00:08:46,694 opportunity because all AI companies are betting on the future 137 00:08:46,694 --> 00:08:48,525 improvements of foundation models. 138 00:08:49,215 --> 00:08:54,915 So the move is you capture distribution as quickly as possible so that once 139 00:08:55,185 --> 00:09:00,675 OpenAI releases whatever, you know, the model is that unlocks a GI or 140 00:09:00,675 --> 00:09:02,415 something similar to a TPT seven. 141 00:09:02,805 --> 00:09:07,965 You have, you are in the, the market position to be ushering in like the 142 00:09:07,965 --> 00:09:10,275 total revolution of the industry. 143 00:09:10,965 --> 00:09:13,455 Um, and yeah, once. 144 00:09:14,760 --> 00:09:15,810 That clicked with me. 145 00:09:15,810 --> 00:09:24,510 I think a lot of other dimensions of the big player strategies made sense and, 146 00:09:24,720 --> 00:09:29,790 um, aligned with my understanding of, of how the venture capital world worked. 147 00:09:30,270 --> 00:09:33,325 Uh, my experiences with venture capitalists over the years, et cetera. 148 00:09:35,055 --> 00:09:37,334 Yeah, so well, let's talk a little bit about that. 149 00:09:37,334 --> 00:09:40,665 So we are largely bootstrapped here at info dash. 150 00:09:40,665 --> 00:09:45,375 We've taken a little bit of, I'll call it, uh, we've done a couple of small seed 151 00:09:45,375 --> 00:09:53,805 rounds with the legal tech fund, and we're mostly, um, founder, founder backed, and. 152 00:09:55,110 --> 00:10:00,690 I have a natural aversion to especially the West coast kind 153 00:10:00,690 --> 00:10:07,230 of VC mindset, which is tends to lean towards growth at all costs. 154 00:10:07,740 --> 00:10:12,420 And there's a very limited focus on capital efficiency. 155 00:10:12,930 --> 00:10:18,150 And as an older founder that you know somebody who's lived through, um. 156 00:10:19,335 --> 00:10:20,880 Thin times, we'll call 'em. 157 00:10:21,105 --> 00:10:27,735 Um, I, I know how valuable resources are and I I'm not willing to trade, 158 00:10:28,905 --> 00:10:35,205 you know, my cap table for a hope and a prayer that I can achieve the growth. 159 00:10:35,385 --> 00:10:40,785 I wanna be able to demonstrate and systematize my mechanisms for growth. 160 00:10:40,785 --> 00:10:44,055 And then if I can raise money and I've got a predictable and 161 00:10:44,055 --> 00:10:46,965 demonstratable go to market machine. 162 00:10:47,625 --> 00:10:55,545 Then I feel good about A, taking the investment and, and b, my willingness to, 163 00:10:55,635 --> 00:10:59,235 or my ability to deliver to expectations. 164 00:10:59,895 --> 00:11:06,944 And, um, but the VC mindset oftentimes is like, you know, we. 165 00:11:07,335 --> 00:11:10,155 We get routinely solicited. 166 00:11:10,185 --> 00:11:12,314 When I say routinely, I've got a spreadsheet. 167 00:11:12,705 --> 00:11:14,445 My listeners have heard me talk about this. 168 00:11:14,445 --> 00:11:16,605 I think there are, there are over 40 bcs. 169 00:11:16,665 --> 00:11:19,905 I started not, not like touches by bcs. 170 00:11:20,025 --> 00:11:22,845 40 different funds have reached out to us. 171 00:11:22,875 --> 00:11:26,415 I started tracking it right before ilta, so that would've been like 172 00:11:26,415 --> 00:11:28,485 late summer, like so mid August. 173 00:11:28,995 --> 00:11:30,675 So in four months. 174 00:11:30,765 --> 00:11:31,365 Um. 175 00:11:33,180 --> 00:11:40,650 Roughly 10 different funds a month and they keep, and it's, uh, it's, it's 176 00:11:40,650 --> 00:11:44,730 really indicative of how aggressive. 177 00:11:45,330 --> 00:11:45,570 Yeah. 178 00:11:45,810 --> 00:11:50,070 You know, and it's just like, but I think you had a good point, which is like the 179 00:11:50,070 --> 00:11:54,840 VCs really favor, like moonshots over building a viable, sustainable business. 180 00:11:54,840 --> 00:11:56,370 Do you, is that your take as well? 181 00:11:56,775 --> 00:11:59,895 It is, and I think it's worth noting that it didn't use to be that way. 182 00:12:00,225 --> 00:12:04,785 Uh, VCs reaching out to legal tech startups wanting to invest Oh, yeah. 183 00:12:05,925 --> 00:12:06,585 In particular. 184 00:12:07,605 --> 00:12:13,965 And so, um, the reason for that is I think downstream of the incentives 185 00:12:13,995 --> 00:12:16,065 of the two industries respectively. 186 00:12:17,790 --> 00:12:22,155 So VCs are incentivized to make moonshots. 187 00:12:23,235 --> 00:12:28,694 Uh, the way that their portfolio map works for those who are listening who perhaps 188 00:12:28,694 --> 00:12:33,585 aren't as familiar with venture capital, is that a venture capitalist, a venture 189 00:12:33,585 --> 00:12:39,495 capital firm, will raise a fund and they will make a limited number of bets on, 190 00:12:39,495 --> 00:12:47,655 say, 20 startups with the expectation that 18 of them will go to zero, and 191 00:12:47,655 --> 00:12:52,545 one or two of them will be worth tens or hundreds of billions of dollars, and. 192 00:12:53,175 --> 00:12:57,315 They're carry that the percentage of the profits that the firm, 193 00:12:57,585 --> 00:13:00,735 uh, earns will return the fund. 194 00:13:02,175 --> 00:13:06,555 And so because of that, VCs structurally aren't interested 195 00:13:06,555 --> 00:13:16,695 in viable businesses that are sub billion dollar outcomes, and they are. 196 00:13:18,420 --> 00:13:23,100 Willing to make, to take a lot of risk if it has some percentage 197 00:13:23,100 --> 00:13:24,840 chance of being the big outcome. 198 00:13:26,220 --> 00:13:31,800 I'll, I'll, I'll note that another dynamic that's worth considering is that VCs 199 00:13:32,250 --> 00:13:39,120 receive a compensation of assets under management as well, and they also will 200 00:13:39,689 --> 00:13:46,199 usually raise their next fund while a fund is, their current fund is still active. 201 00:13:47,250 --> 00:13:52,319 The way that they will raise that fund is by reporting to their investors, the 202 00:13:52,319 --> 00:13:58,290 limited partners, what the returns of the fund are, and so they have a, a short term 203 00:13:58,290 --> 00:14:03,720 incentive for their portfolio companies to go on and raise a new round, 18 months 204 00:14:03,720 --> 00:14:06,209 after investment at a five x markup. 205 00:14:06,615 --> 00:14:08,840 That helps them raise their new fund fund. 206 00:14:08,845 --> 00:14:14,445 So VCs have a short term incentive as well as a big risk taking incentive, 207 00:14:14,715 --> 00:14:20,025 and that combination is diametrically opposed to the way lawyers work as 208 00:14:20,355 --> 00:14:22,485 sure most people listening understand. 209 00:14:22,995 --> 00:14:30,105 Uh, lawyers are naturally risk averse as a profession, which completely makes sense. 210 00:14:30,165 --> 00:14:32,200 Their upside is capped by. 211 00:14:32,895 --> 00:14:39,405 Whatever fees they've negotiated, uh, for a deal, um, by, you know, hourly 212 00:14:39,405 --> 00:14:42,915 part, hourly billing rates, et cetera. 213 00:14:43,155 --> 00:14:46,635 Um, they have uncapped downside. 214 00:14:46,665 --> 00:14:50,475 However, if they make a mistake that exposes their client to, you 215 00:14:50,475 --> 00:14:51,795 know, hundreds of millions dollars. 216 00:14:53,535 --> 00:14:57,975 Uh, so traditionally the. 217 00:14:58,410 --> 00:15:02,819 And then another component about the legal industry is that it, it takes a while 218 00:15:02,819 --> 00:15:04,949 to build a good product in, in legal. 219 00:15:05,520 --> 00:15:08,790 Uh, working with Doc XFiles is really technically challenging. 220 00:15:09,209 --> 00:15:12,599 Gaining the trust of lawyers, uh, takes a while. 221 00:15:12,930 --> 00:15:19,170 So traditionally, the legal industry has not been particularly 222 00:15:19,170 --> 00:15:21,089 favored by venture capital. 223 00:15:21,120 --> 00:15:23,615 When we went through Y Combinator, winner of 21. 224 00:15:25,755 --> 00:15:30,255 Pretty prominent venture capitalist was, was speaking with us and advised 225 00:15:30,255 --> 00:15:34,005 all of the legal tech companies in are bashed to pivot to a different industry. 226 00:15:35,145 --> 00:15:37,785 Um, so the times have changed. 227 00:15:38,475 --> 00:15:42,345 Yeah, it's, and you know what, it wasn't bad advice in 2021. 228 00:15:42,645 --> 00:15:47,355 I can tell you, as somebody who sold and sold technology into law firms 229 00:15:47,355 --> 00:15:51,195 for many years, it's a, it's a slog. 230 00:15:52,935 --> 00:15:56,324 Has been, yeah, it's difficult and you know, started. 231 00:15:59,715 --> 00:16:01,065 Had some difficult lessons. 232 00:16:01,065 --> 00:16:01,905 We had to learn. 233 00:16:02,145 --> 00:16:06,975 You have to eat table scraps for years, potentially to build the 234 00:16:06,975 --> 00:16:11,535 resume, to just even get the at bats at the top of the AM law. 235 00:16:11,925 --> 00:16:12,075 Yep. 236 00:16:12,105 --> 00:16:17,595 And you know, um, the Harveys and LARAs of the world and others who've, 237 00:16:17,835 --> 00:16:23,365 um, demonstrated incredible growth out of the gate have really kind of sh. 238 00:16:23,880 --> 00:16:24,750 Took a shortcut. 239 00:16:25,020 --> 00:16:28,980 They've shortcutted that that process, that historically has been in place. 240 00:16:29,400 --> 00:16:33,420 And my theory on how they've been able to do that is really twofold. 241 00:16:33,600 --> 00:16:34,980 Two main drivers. 242 00:16:35,490 --> 00:16:38,580 One is the VCs in their cap table. 243 00:16:39,360 --> 00:16:43,770 So if you look at Harvey, for example, I'm more familiar with their cap table. 244 00:16:43,860 --> 00:16:51,960 OpenAI, Sequoia, A 16 Z, I mean the who's who of Silicon Valley vc. 245 00:16:52,590 --> 00:16:53,160 And. 246 00:16:54,840 --> 00:17:00,330 A, those funds are big buyers of legal services, right? 247 00:17:00,360 --> 00:17:07,650 All their Port cos and, um, all their m and a work, and they have pull 248 00:17:07,950 --> 00:17:13,470 in the legal market that where they can facilitate, um, introductions. 249 00:17:13,890 --> 00:17:18,270 It's also a very strong signal of credibility when 250 00:17:18,270 --> 00:17:19,650 you come to the table with. 251 00:17:20,040 --> 00:17:24,930 Open AI and Google Ventures and Sequoia and a 16 Z in your cap table. 252 00:17:24,930 --> 00:17:25,020 Sure. 253 00:17:25,290 --> 00:17:30,360 It gives you in instant credibility and lawyers are very much focused 254 00:17:30,360 --> 00:17:37,770 on, um, precedent and brand and if you come and prestige. 255 00:17:37,950 --> 00:17:40,290 That's how the, that's how the law firm market works. 256 00:17:40,290 --> 00:17:43,110 Like the, the, the firms at the top of the AM law. 257 00:17:43,560 --> 00:17:47,430 How they, how they're able to get those rates are, it's, it's really 258 00:17:47,430 --> 00:17:49,050 their prestige and their brand. 259 00:17:50,264 --> 00:17:54,254 So that that translates really well into legal. 260 00:17:54,554 --> 00:17:57,615 And then the other dynamic that has allowed them to 261 00:17:57,615 --> 00:17:59,985 shortcut that process is fomo. 262 00:18:00,554 --> 00:18:02,865 There is a tremendous amount of fomo. 263 00:18:02,865 --> 00:18:06,705 There's this existential threat out there that law firm leadership 264 00:18:06,705 --> 00:18:09,705 looks at AI and is implement. 265 00:18:10,365 --> 00:18:14,865 A tremendous amount of change and transformation within the industry, and 266 00:18:14,865 --> 00:18:18,435 we don't really know how it's gonna shake out, but one thing we do know, as law firm 267 00:18:18,435 --> 00:18:20,235 leaders, we better be doing something. 268 00:18:20,595 --> 00:18:24,495 So let's just go buy something regardless if it's a fit. 269 00:18:24,500 --> 00:18:27,465 Do, do you agree with those dynamics or do you see others at play? 270 00:18:28,065 --> 00:18:29,295 Yeah, I completely agree. 271 00:18:29,295 --> 00:18:34,965 I think you, it's hard to underestimate the power of fear as a motivator, 272 00:18:35,535 --> 00:18:38,025 uh, in, in this conversation. 273 00:18:39,570 --> 00:18:46,410 Uh, the narrative that lawyers have been seeing really since the 274 00:18:46,410 --> 00:18:52,800 release of chat GPT, is that AI is marching on a trajectory to pose an 275 00:18:52,800 --> 00:18:54,570 existential threat to their livelihood. 276 00:18:55,590 --> 00:19:01,890 That there are claims that, uh, AI is going to automate legal reasoning, 277 00:19:02,400 --> 00:19:05,550 uh, in a number of years and. 278 00:19:08,490 --> 00:19:16,200 Personally don't agree with those claims, but I think that one can see how tempting 279 00:19:16,200 --> 00:19:23,010 they are for, say somebody like a venture capitalist who is regularly entering into 280 00:19:23,610 --> 00:19:29,100 legal agreements and they are reviewing term sheets on a regular basis, and 281 00:19:29,100 --> 00:19:34,440 then they drop a term sheet into JTPT and you know, it can do a, a pretty 282 00:19:34,440 --> 00:19:36,240 reasonable job of translating legalese. 283 00:19:39,675 --> 00:19:40,545 In a lot of cases. 284 00:19:41,175 --> 00:19:44,745 Now, obviously there are plenty of issues of reliability, of 285 00:19:44,835 --> 00:19:49,095 hallucinating of it, not having proper context to do an adequate job. 286 00:19:50,385 --> 00:19:59,145 But, um, I think it would, it, an idea arose amongst, uh, VCs amongst 287 00:19:59,145 --> 00:20:00,765 the tech community, which then. 288 00:20:02,670 --> 00:20:07,230 Uh, amplified out to the right, wider world and to lawyers in particular, 289 00:20:07,590 --> 00:20:12,780 that AI was, was on trajectory to, to upend the legal industry. 290 00:20:13,380 --> 00:20:17,190 And so if you're a lawyer, if you're a leader at a law firm and you hear this 291 00:20:18,300 --> 00:20:27,284 and I. Perfectly reasonable and expected response to start thinking like, okay, 292 00:20:27,284 --> 00:20:31,304 how can, how can we mitigate this risk? 293 00:20:31,304 --> 00:20:32,504 What steps should we take? 294 00:20:32,865 --> 00:20:37,034 Even if it is, you know, some small percentage that this type of 295 00:20:37,034 --> 00:20:43,845 transformation really happens, uh, how can we make sure that we're in a position 296 00:20:43,845 --> 00:20:49,485 to, to handle the transformation well and. 297 00:20:50,250 --> 00:20:55,050 Law firms aren't, they're not in the business of assessing 298 00:20:55,530 --> 00:21:00,840 the viability of particular startups, legal or AI strategies. 299 00:21:01,320 --> 00:21:01,649 Right. 300 00:21:01,860 --> 00:21:08,490 Um, so from a risk management perspective, the, the strategy. 301 00:21:09,270 --> 00:21:15,120 Makes the most sense is adopt whatever the leading player is, because then, 302 00:21:15,270 --> 00:21:17,879 you know, if the transformation happens, at least you're in the same 303 00:21:17,879 --> 00:21:19,919 boat as all of your competitors. 304 00:21:21,060 --> 00:21:25,889 Yeah, and I mean, historically, more broadly in normal moving markets 305 00:21:25,889 --> 00:21:32,399 and in, in more broadly, you have the garters and the foresters of 306 00:21:32,399 --> 00:21:34,710 the world that they're the ones who. 307 00:21:35,685 --> 00:21:40,665 Evaluate the strategy and the ability to execute on that strategy and put you 308 00:21:40,665 --> 00:21:46,785 on a magic quadrant that this dynamic happened before Gartner had a chance. 309 00:21:46,785 --> 00:21:52,875 I don't know if they've, if they even have, um, a magic quadrant for legal 310 00:21:52,875 --> 00:21:56,475 tools, but they're, they're the ones who typically do that, not lawyers. 311 00:21:57,075 --> 00:21:57,435 Yeah. 312 00:21:57,675 --> 00:21:57,915 Yeah. 313 00:21:57,915 --> 00:21:58,545 Absolutely. 314 00:21:58,635 --> 00:21:59,685 And I, I would argue that. 315 00:22:01,455 --> 00:22:08,325 The fear and market leadership dynamic was, was entering into 316 00:22:08,325 --> 00:22:10,035 place before anyone had a chance. 317 00:22:10,305 --> 00:22:14,835 Um, I mean, for a long time, Harvey's website was, was just a, a sign up 318 00:22:14,835 --> 00:22:17,145 and a wait list for like a year. 319 00:22:17,505 --> 00:22:23,055 And, but somehow in that era they managed to, to build that brand prestige. 320 00:22:23,925 --> 00:22:24,345 Yeah. 321 00:22:25,005 --> 00:22:25,305 And um. 322 00:22:26,669 --> 00:22:27,540 I've seen their product. 323 00:22:27,540 --> 00:22:28,740 It's, it's a good product. 324 00:22:28,860 --> 00:22:30,210 Um, so is Lara's. 325 00:22:30,210 --> 00:22:32,010 I've seen both of them in action. 326 00:22:32,610 --> 00:22:38,490 Um, the one point I wanted to make too, that, that, that you touched 327 00:22:38,490 --> 00:22:44,429 on is like how legal tech was not a good fit historically for, for vc. 328 00:22:45,000 --> 00:22:52,560 And I think there was so much excitement when the AI light bulb went off in. 329 00:22:53,400 --> 00:22:58,560 Investors heads on, wow, what could this do to legal is because, you know, 330 00:22:58,560 --> 00:23:06,180 legal is one of the last holdouts of industries that have successfully 331 00:23:06,180 --> 00:23:09,750 resisted a tech transformation, right? 332 00:23:09,750 --> 00:23:13,440 If you look at almost every other industry, not all, but the 333 00:23:13,440 --> 00:23:17,610 vast majority, there has been a major tech transformation. 334 00:23:17,760 --> 00:23:18,990 Look at financial services. 335 00:23:18,990 --> 00:23:20,220 So that's a world I came from. 336 00:23:20,220 --> 00:23:21,900 I spent 10 years at Bank of America. 337 00:23:22,800 --> 00:23:26,190 They are literally shutting down branches right now because 338 00:23:26,190 --> 00:23:33,120 of the digital experience and services that they can deliver. 339 00:23:33,720 --> 00:23:37,320 Um, that is a massive shift. 340 00:23:37,380 --> 00:23:42,540 They've got, you know, bank of America's in all 50 states and internationally, 341 00:23:42,540 --> 00:23:46,890 and being able to reduce their physical footprint because of technology. 342 00:23:47,160 --> 00:23:48,930 You don't even have to go into a branch. 343 00:23:49,290 --> 00:23:56,669 I had a. Six figure and not a low six figure wire that I went into a branch 344 00:23:56,669 --> 00:24:02,790 to initiate and the guy told me to go home and do it online, uh, because I 345 00:24:02,790 --> 00:24:05,189 didn't set an appointment on the app. 346 00:24:05,189 --> 00:24:06,450 He's like, you can do this for your phone. 347 00:24:06,450 --> 00:24:07,169 I was blown away. 348 00:24:07,169 --> 00:24:09,699 I was like, I can wire hundreds of thousands of dollars on. 349 00:24:10,620 --> 00:24:11,310 Through the app. 350 00:24:11,310 --> 00:24:12,300 He's like, absolutely. 351 00:24:12,629 --> 00:24:13,980 And he was right and I did it. 352 00:24:14,370 --> 00:24:22,409 But legal has been, has successfully held the, the transformation, um, God's away. 353 00:24:22,919 --> 00:24:29,820 And now it's like YY that um, that dynamic, that ability to 354 00:24:29,850 --> 00:24:32,250 resist is, is no longer there. 355 00:24:32,250 --> 00:24:37,050 So I think that there's a ton of pen up demand for efficiency on the buy side 356 00:24:37,439 --> 00:24:39,990 in law firms and VCs recognize that. 357 00:24:40,815 --> 00:24:41,835 Yeah, I think that's right. 358 00:24:42,945 --> 00:24:50,055 Um, well, tell us a little bit about your take on like the, the a GI piece, 359 00:24:50,055 --> 00:24:57,885 because it sounds like you're skeptical of these models and their ability to 360 00:24:58,425 --> 00:25:01,515 deliver on the legal reasoning piece. 361 00:25:01,635 --> 00:25:06,315 Um, am I, am I reading your position correctly? 362 00:25:07,185 --> 00:25:07,635 Um. 363 00:25:09,899 --> 00:25:10,530 Somewhat. 364 00:25:10,530 --> 00:25:16,500 I think I, I'm, I definitely believe in the value of AI and 365 00:25:16,500 --> 00:25:19,260 LLMs in a number of domains. 366 00:25:19,260 --> 00:25:26,040 Like I certainly think that AI can and will be really important to the law. 367 00:25:26,550 --> 00:25:31,199 Uh, to the extent I'm skeptical, I think I'm skeptical of the, 368 00:25:31,530 --> 00:25:38,189 the claims that AI will render lawyers obsolete, that they will. 369 00:25:39,149 --> 00:25:49,110 Um, yeah, the, the more the, the, the a GI claims that there, there isn't 370 00:25:49,110 --> 00:25:52,530 a need for a human lawyer anymore. 371 00:25:53,460 --> 00:25:59,250 And I'm not going to say that I'm, you know, a hundred percent confident that 372 00:25:59,490 --> 00:26:02,129 a GI isn't is theoretically impossible. 373 00:26:02,790 --> 00:26:06,570 Uh, I think from my experience of using LLMs. 374 00:26:07,530 --> 00:26:09,360 Every day for hours a day coding. 375 00:26:10,350 --> 00:26:12,390 I, I have reasons for skepticism. 376 00:26:12,570 --> 00:26:16,050 Um, if, if a DI does happen, then we're all in the same boat. 377 00:26:16,110 --> 00:26:25,065 But setting that aside, um, I think there is reason to think that lawyers 378 00:26:25,205 --> 00:26:32,010 in particular are the, some of the least likely people to be automated from ai. 379 00:26:33,360 --> 00:26:35,610 So one pretty critical difference. 380 00:26:37,199 --> 00:26:45,330 Being a in coding is that with coding and, and, and the difference in, in 381 00:26:45,330 --> 00:26:49,679 respect to how AI can augment your workflow is that with coding you have 382 00:26:50,520 --> 00:26:53,580 built in systems of verifiability. 383 00:26:54,360 --> 00:27:03,060 So if in LLM writes a bunch of code for you, you can test your code and verify. 384 00:27:03,465 --> 00:27:08,145 Whether or not the feature works or passes a suite of tests very quickly. 385 00:27:09,254 --> 00:27:16,845 When Ai, uh, drafts a contract for you, there isn't a system of 386 00:27:16,845 --> 00:27:21,915 verification that's akin to like running a unit test or testing a feature. 387 00:27:22,215 --> 00:27:28,545 Aside from you, the lawyer assiduously reviewing each and every 388 00:27:28,545 --> 00:27:30,795 line that was introduced by the. 389 00:27:33,345 --> 00:27:40,365 So that is not to say that AI drafting can't add value to the workflow. 390 00:27:40,785 --> 00:27:41,895 Uh, I think it can. 391 00:27:41,895 --> 00:27:51,945 I think it can, especially when say a lawyer is, uh, practicing, uh, or 392 00:27:54,105 --> 00:27:57,435 working in a field of law that they don't expertise. 393 00:28:02,940 --> 00:28:11,910 Some knowledge from the LM can be useful or I think it can also be really helpful 394 00:28:11,910 --> 00:28:20,370 for lower stakes agreements as well, where the, it's more acceptable to 395 00:28:21,240 --> 00:28:26,160 incur some risk of LMS making mistake. 396 00:28:26,310 --> 00:28:28,710 Um, but I think in any case. 397 00:28:29,580 --> 00:28:37,350 There will, so long as LLMs aren't capable of fully replacing all people, 398 00:28:38,040 --> 00:28:46,410 then there will be a need for a human lawyer to review its work to make 399 00:28:46,410 --> 00:28:48,210 sure it's not introducing mistakes. 400 00:28:48,210 --> 00:28:56,280 Uh, and so for all of those reasons, I am skeptical of that LLMs. 401 00:28:57,705 --> 00:29:01,755 Pose a, a risk to lawyers' livelihood. 402 00:29:02,355 --> 00:29:03,105 If anything. 403 00:29:03,165 --> 00:29:04,245 I think that 404 00:29:06,855 --> 00:29:12,795 the AI era that we are entering into in the broader economy is likely going to 405 00:29:12,855 --> 00:29:18,945 increase the demand for legal services more broadly across a variety of domains. 406 00:29:20,325 --> 00:29:24,405 So, um, yeah, and I don't, I don't necessarily disagree with you. 407 00:29:24,405 --> 00:29:25,425 Um, I think that. 408 00:29:26,595 --> 00:29:29,685 There are some inherent limitations in the LLM 409 00:29:31,755 --> 00:29:37,485 architecture, um, and its ability to generate novel ideas. 410 00:29:37,545 --> 00:29:40,905 Um, I think there, there's been some creative engineering until, you 411 00:29:40,905 --> 00:29:46,545 know, when we, when inference time compute, um, kind of saved the day 412 00:29:46,755 --> 00:29:50,415 a little bit with LLMs because we'd really started to hit our head on 413 00:29:50,415 --> 00:29:53,985 the scaling laws and, you know, um. 414 00:29:54,600 --> 00:29:58,950 Reasoning or, you know, inference, time compute really extended 415 00:29:59,850 --> 00:30:03,179 the, um, progress trajectory. 416 00:30:03,780 --> 00:30:04,740 But, um. 417 00:30:05,610 --> 00:30:10,320 Switching gears a little bit, I'm curious about, so valuations, I gave a, I, uh, 418 00:30:10,830 --> 00:30:15,810 I had the privilege of moderating the keynote panel at the Legal Tech Connect 419 00:30:15,810 --> 00:30:18,000 conference in New York a few months ago. 420 00:30:18,659 --> 00:30:19,950 And as part of my. 421 00:30:20,754 --> 00:30:21,774 Opening remarks. 422 00:30:21,774 --> 00:30:27,055 I, I used an analogy to, this was right after the Harvey and Lara 423 00:30:27,055 --> 00:30:30,055 a DX, uh, rounds were announced. 424 00:30:30,415 --> 00:30:33,655 So for those that don't know, there were investments in both 425 00:30:33,655 --> 00:30:35,090 Harvey and Lara and they were. 426 00:30:35,940 --> 00:30:38,250 At an 80 times revenue valuation. 427 00:30:38,700 --> 00:30:42,240 So to put this in context, I, I gave a little, a little 428 00:30:42,240 --> 00:30:43,530 metaphor that I'll share here. 429 00:30:43,530 --> 00:30:46,680 I haven't shared it on the podcast before, but it's, it's really interesting. 430 00:30:46,980 --> 00:30:51,480 So my wife and I own five gyms in St. Louis, and, um, they 431 00:30:51,480 --> 00:30:53,640 operate on about a 20% net margin. 432 00:30:54,390 --> 00:31:00,120 If we were to sell those businesses, it would be three to four times ebitda, 433 00:31:00,300 --> 00:31:02,100 which is basically your net margin. 434 00:31:03,150 --> 00:31:03,660 So. 435 00:31:04,334 --> 00:31:09,915 Let's assume for just a minute that, um, that was the 436 00:31:10,574 --> 00:31:13,814 ultimate outcome for investors. 437 00:31:14,145 --> 00:31:18,615 Like if these businesses, whatever they stalled, they, they grew, they 438 00:31:18,615 --> 00:31:24,885 matured and hit a ceiling, and you ended up selling for a multiple of ebitda. 439 00:31:25,635 --> 00:31:32,145 So at a 20% multiple of ebitda, it would take. 440 00:31:32,985 --> 00:31:37,245 400 years to break even on your investment. 441 00:31:38,715 --> 00:31:40,695 400 years. 442 00:31:41,264 --> 00:31:47,985 If George Washington made that, wrote that check to invest in that 443 00:31:47,985 --> 00:31:53,475 round, he'd still have another 150 years until he broke even. 444 00:31:53,745 --> 00:31:55,965 Now, again, lots of caveats there. 445 00:31:55,965 --> 00:32:00,824 First of all, that's not the bet that investors are placing on a on a 20%. 446 00:32:01,785 --> 00:32:07,605 EBITDA multiple, like, but it's a good lens as just a sanity check. 447 00:32:07,605 --> 00:32:12,375 Like, okay, this is how, and it's not just gyms, that's, that's how restaurants, 448 00:32:12,375 --> 00:32:16,185 that's how pretty much every business sells is for a multiple of ebitda, except 449 00:32:16,185 --> 00:32:20,745 in the software world, the technology world where things get, you know. 450 00:32:21,465 --> 00:32:24,795 Typically it's, it's a, it's a multiple of recurring revenue. 451 00:32:25,395 --> 00:32:30,705 But, um, man, and I made a joke and said, you know, 400 years ago, you know, what 452 00:32:30,705 --> 00:32:35,385 was going on 400 years ago, the Nina, the Pinta, and the Santa Maria, right? 453 00:32:35,385 --> 00:32:42,195 Just to put some, put a visual on what time span we're sitting in New York City, 454 00:32:42,255 --> 00:32:45,735 which was, um, didn't even really exist. 455 00:32:46,245 --> 00:32:46,695 Uh. 456 00:32:47,100 --> 00:32:49,650 So what are your thoughts on these valuations? 457 00:32:49,650 --> 00:32:51,510 Like, um, do they make sense? 458 00:32:51,510 --> 00:32:55,230 Can you make sense of them or is it r hard for you to get your head around as well? 459 00:32:55,800 --> 00:32:57,570 It's a little hard for me to get my head around. 460 00:32:58,020 --> 00:33:05,490 Um, I mean, I try to think of like, what is a comparable company in legal 461 00:33:05,490 --> 00:33:12,870 tech that has a comparable value that I could imagine them growing into? 462 00:33:14,324 --> 00:33:17,205 And I'm just not really sure what that is. 463 00:33:17,205 --> 00:33:22,215 I mean, the, these numbers are private, but I believe according to estimates 464 00:33:22,215 --> 00:33:26,715 that Harvey's latest valuation is greater than that of iManage and net docs and 465 00:33:26,715 --> 00:33:30,254 let combined, um, don't quote me on that. 466 00:33:30,495 --> 00:33:32,419 I, that's based off of estimates. 467 00:33:32,564 --> 00:33:39,585 Um, but that seems at least, I think that's like ballpark about right. 468 00:33:40,725 --> 00:33:41,655 And so. 469 00:33:44,100 --> 00:33:50,220 Yeah, it, it, it's a bit, it is challenging for me to understand 470 00:33:52,290 --> 00:33:58,740 how they grow into evaluation that is, you know, as great or larger 471 00:33:58,740 --> 00:34:02,970 than the most critical pieces of lawyers' existing workflows. 472 00:34:04,410 --> 00:34:05,460 It's a great point. 473 00:34:05,460 --> 00:34:08,130 So, yeah, it's been a little bit, um. 474 00:34:08,520 --> 00:34:11,250 Since I've looked at this, but I just kind of did a quick search. 475 00:34:11,909 --> 00:34:18,510 So, um, Thomson Reuters has a market cap of around 60 billion. 476 00:34:19,350 --> 00:34:23,880 And what investors, so there's a little bit of debate. 477 00:34:23,880 --> 00:34:30,450 One of the in, in early VC rounds, investors are thinking to themselves, 478 00:34:30,450 --> 00:34:35,250 what has to be true in order for me to get 10 x on this investment? 479 00:34:36,465 --> 00:34:41,895 Now you could, you could argue that, well, this wasn't a series A or a series B. 480 00:34:42,195 --> 00:34:47,355 This was, I don't know what, how, how many rounds in they were, but you could 481 00:34:47,355 --> 00:34:52,215 argue that, okay, that number, that multiple goes down in later rounds. 482 00:34:52,545 --> 00:34:59,625 But I would counter back with the reason that the expectations are 10 x in the 483 00:34:59,625 --> 00:35:01,815 earlier rounds is because of risk. 484 00:35:02,445 --> 00:35:05,985 Right They to, to get the right risk, reward balance you need to 485 00:35:05,985 --> 00:35:10,665 10 x you know, one or two of your, to your use, your example of your 486 00:35:10,665 --> 00:35:12,765 20 port codes, half the 10 x. 487 00:35:13,395 --> 00:35:18,885 I would argue that based on that valuation, you're at a similar. 488 00:35:19,740 --> 00:35:25,260 Risk as somebody who's investing in a series A in a startup, right? 489 00:35:25,260 --> 00:35:28,560 That's kind of the risk profile, and you need a commensurate reward. 490 00:35:28,800 --> 00:35:35,760 So if I'm writing that check, I'm thinking 10 x. So 10 x would 491 00:35:35,760 --> 00:35:38,910 require an $80 billion exit. 492 00:35:40,169 --> 00:35:44,910 For Harvey, if they were to IPO, they're actually too big to be bought right now. 493 00:35:45,120 --> 00:35:47,459 There's not a legal tech company maybe other than Thomson 494 00:35:47,459 --> 00:35:49,109 Reuters, which is a hard no. 495 00:35:49,680 --> 00:35:51,930 Uh, given that, you know, their investment in co-counsel, they're 496 00:35:51,930 --> 00:35:53,490 not, they're not gonna buy Harvey. 497 00:35:54,060 --> 00:35:56,850 So how do you get to an $80 billion exit? 498 00:35:56,910 --> 00:36:04,649 The largest, the the largest IPO in legal tech history is, um, LegalZoom. 499 00:36:05,505 --> 00:36:12,585 And it was, it iPod during the, uh, 21, you know, craziness. 500 00:36:12,615 --> 00:36:19,995 2021 when the fed dumped $7 trillion of capital into the economy and that all that 501 00:36:19,995 --> 00:36:21,495 capital had to find a productive home. 502 00:36:22,065 --> 00:36:28,185 And it iPod for, I think the number, it was in the high single digits, seven 503 00:36:28,785 --> 00:36:33,765 8 billion Today, it's worth about 1.5. 504 00:36:34,620 --> 00:36:34,980 Right. 505 00:36:34,980 --> 00:36:40,890 So think about what has to happen in order for an IPO of that size. 506 00:36:41,220 --> 00:36:45,810 Again, you would eclipse Thomson Reuters, which goes way beyond just legal tech. 507 00:36:46,500 --> 00:36:49,230 I mean, Thomson Reuters is a massive company. 508 00:36:49,230 --> 00:36:53,069 So yeah, it's hard for me to, to, to do the metal math to get there. 509 00:36:53,100 --> 00:36:53,970 It's, it really is. 510 00:36:54,509 --> 00:36:54,750 Yeah. 511 00:36:55,140 --> 00:36:55,319 Yeah. 512 00:36:55,319 --> 00:36:55,740 I agree. 513 00:36:56,819 --> 00:37:00,240 So, um, let's talk a little bit about, um. 514 00:37:01,020 --> 00:37:07,529 The open ais and the Anthropics, and for that matter, GRS and Geminis, 515 00:37:08,009 --> 00:37:10,620 their ability to compete with Harvey. 516 00:37:10,620 --> 00:37:12,900 Even though OpenAI was an early investor. 517 00:37:13,470 --> 00:37:17,759 Um, I've heard, I'm pretty sure I've heard or read like some of 518 00:37:17,759 --> 00:37:21,840 the founders of these companies say that that's their biggest threat. 519 00:37:22,380 --> 00:37:25,380 How do you see like the frontier models. 520 00:37:26,370 --> 00:37:30,360 Competing and their ability to compete with these legal specific solutions? 521 00:37:33,780 --> 00:37:44,130 Well, I think that the, the intelligence component of legal AI product, really 522 00:37:44,160 --> 00:37:45,900 any AI product for that matter, 523 00:37:47,940 --> 00:37:51,150 it, it comes straight from the model providers at this point. 524 00:37:51,690 --> 00:37:52,770 So, um. 525 00:37:53,505 --> 00:37:58,605 For as a quick definition, foundation model is the term that's used to refer 526 00:37:58,605 --> 00:38:03,525 to the sort of out of the box models that OpenAI or Entropic or Google Produce, 527 00:38:04,965 --> 00:38:12,555 and there no industry specific provider is going to be the foundation models. 528 00:38:14,025 --> 00:38:17,565 Early on when CHATT was just released. 529 00:38:19,920 --> 00:38:23,549 GPT-3 five really wasn't that useful yet. 530 00:38:23,970 --> 00:38:25,529 The context window wasn't big enough. 531 00:38:25,620 --> 00:38:26,549 It wasn't that smart. 532 00:38:27,750 --> 00:38:33,150 There were gains to be made by fine tuning it, so sort of retraining the 533 00:38:33,150 --> 00:38:36,150 model on context specific information. 534 00:38:37,230 --> 00:38:42,720 But since then, the subsequent model iterations have gotten, 535 00:38:43,410 --> 00:38:48,990 have gotten so big with so much training data that any marginal. 536 00:38:50,085 --> 00:38:58,095 Data that you could fine tune it with from, you know, customer's legal documents 537 00:38:58,875 --> 00:39:04,845 is the tiniest drop in the bucket, and it really is not going to make a 538 00:39:04,845 --> 00:39:06,165 meaningful difference at this point. 539 00:39:07,245 --> 00:39:10,275 Now, the thing that does make a difference relative to just CHATT 540 00:39:10,395 --> 00:39:14,715 PT Outta box is referred to as in context learning, which you can think 541 00:39:14,715 --> 00:39:16,995 of basically as just how well you. 542 00:39:19,050 --> 00:39:21,510 Having it a well-designed, 543 00:39:23,850 --> 00:39:25,260 yeah, well-crafted prompt. 544 00:39:25,260 --> 00:39:28,920 Pulling in the right context does meaningfully affect the quality of the 545 00:39:28,920 --> 00:39:34,650 response, and so I think a strategy that I've noticed with a lot of AI 546 00:39:34,650 --> 00:39:41,640 startups at the moment is positioning themself as a domain specific product. 547 00:39:44,819 --> 00:39:52,230 Repackaging well written prompts on top of foundation models, and I think a lot of 548 00:39:52,649 --> 00:40:00,000 users of Chat GPT, um, don't realize ways in which you can modify it with your own 549 00:40:00,000 --> 00:40:02,310 custom prompts to achieve similar results. 550 00:40:03,629 --> 00:40:08,609 All of which is to say that I don't think that any. 551 00:40:10,305 --> 00:40:16,815 AI startup, legal or otherwise is, can build any kind of meaningful note 552 00:40:17,265 --> 00:40:22,785 on the quality of the intelligence to the extent that there's going 553 00:40:22,785 --> 00:40:24,435 to be a moat, a product note. 554 00:40:24,555 --> 00:40:28,755 It needs to come from the, the product that you build on top of 555 00:40:28,755 --> 00:40:34,395 it, the workflows that you build on top of it, uh, how meaningfully 556 00:40:34,395 --> 00:40:38,325 different your solution is to what? 557 00:40:39,270 --> 00:40:42,660 Somebody would be capable of achieving by using a chat bott or 558 00:40:42,660 --> 00:40:44,520 by using cloud code, et cetera. 559 00:40:45,840 --> 00:40:55,290 Yeah, and speaking of which, it's um, it is astounding at the rate at which, like, 560 00:40:55,290 --> 00:40:56,850 you know, cloud code's a good example. 561 00:40:57,240 --> 00:41:00,400 Again, I'm not sure when this will will come out, but, um, you know. 562 00:41:01,274 --> 00:41:05,504 Claude Code has historically been a little bit challenging to use. 563 00:41:05,504 --> 00:41:09,375 Like you have to, you have to use a terminal, like, you 564 00:41:09,375 --> 00:41:11,294 know, visual Studio Code. 565 00:41:11,745 --> 00:41:15,734 You have to wire up GitHub, you gotta run some bash scripts. 566 00:41:15,765 --> 00:41:16,455 It's um. 567 00:41:17,190 --> 00:41:20,850 It's been a little bit painful, candidly, and nobody really loves 568 00:41:20,850 --> 00:41:22,529 the whole terminal interface. 569 00:41:23,220 --> 00:41:28,860 Um, you know, they are rapidly innovating and Claude Cowork just got released 570 00:41:28,860 --> 00:41:30,240 today, as we talked about earlier. 571 00:41:30,870 --> 00:41:33,840 And, um, they're making it easier and easier. 572 00:41:34,290 --> 00:41:38,190 And what I have noticed, I don't know if this is true with the Harvey's and 573 00:41:38,190 --> 00:41:44,549 Leggos of the world, but like it, um, companies who, Microsoft's a good example, 574 00:41:44,549 --> 00:41:46,590 who have taken and built on top of. 575 00:41:47,115 --> 00:41:52,335 The foundation models, it's usually watered down, at least to some extent. 576 00:41:52,634 --> 00:41:57,585 And the innovations get there later, right? 577 00:41:57,674 --> 00:42:00,134 So co-pilot's a great example. 578 00:42:00,765 --> 00:42:04,605 Uh, my listeners and people who follow me on LinkedIn know 579 00:42:04,605 --> 00:42:06,615 I am not a fan of co-pilot. 580 00:42:06,705 --> 00:42:09,315 It's, um, it's, it's just not good. 581 00:42:09,944 --> 00:42:12,915 And frankly, Microsoft should be embarrassed. 582 00:42:13,275 --> 00:42:19,575 Um, uh, Satya has been such a amazing CEOI was at, I was at Microsoft in the 583 00:42:19,575 --> 00:42:24,975 Gates Balmer era, and Satya has just changed Microsoft in so many positive 584 00:42:24,975 --> 00:42:29,805 ways, but I've actually lost some respect for him through, um, through 585 00:42:29,805 --> 00:42:37,005 this copilot fiasco and how, how far away the marketing is from reality. 586 00:42:37,545 --> 00:42:38,505 Like any, it definitely works. 587 00:42:39,180 --> 00:42:40,830 It just doesn't work. 588 00:42:40,890 --> 00:42:44,670 Like, it doesn't work, like it, it like try using it in Excel 589 00:42:44,670 --> 00:42:46,440 and do anything meaningful. 590 00:42:46,710 --> 00:42:48,600 It just doesn't work. 591 00:42:49,080 --> 00:42:51,060 And, um, yeah, I don't know. 592 00:42:51,060 --> 00:42:54,360 Is it your take too that not specifically with copilot? 593 00:42:54,990 --> 00:42:59,580 Uh, I really want, by the way, I'm like, I'm, I want copilot to be better. 594 00:42:59,610 --> 00:43:00,630 We're a Microsoft shop. 595 00:43:01,230 --> 00:43:01,320 Sure. 596 00:43:01,320 --> 00:43:01,380 Yeah. 597 00:43:01,380 --> 00:43:02,820 I want it, I want it to be good. 598 00:43:02,940 --> 00:43:03,690 It's just not. 599 00:43:04,200 --> 00:43:05,070 Um, but. 600 00:43:05,895 --> 00:43:10,275 Like, is it your perspective as well that like the foundation, anybody, 601 00:43:10,275 --> 00:43:12,105 anything built on the foundation models? 602 00:43:13,395 --> 00:43:14,985 It's not a one-to-one. 603 00:43:14,985 --> 00:43:19,155 Like there's, you get features sometimes later and there's some translate. 604 00:43:19,155 --> 00:43:22,755 I don't know if something happens in translation, but the, the, the 605 00:43:22,755 --> 00:43:24,225 frontier model seem to be better. 606 00:43:26,390 --> 00:43:27,285 Oh, interesting. 607 00:43:27,285 --> 00:43:30,495 So you, you, you're asking if the frontier models. 608 00:43:31,365 --> 00:43:33,405 Better than the models that you get when you're actually 609 00:43:33,405 --> 00:43:34,875 using a third party application. 610 00:43:34,965 --> 00:43:35,625 Exactly. 611 00:43:36,675 --> 00:43:43,965 Um, you know, I guess I, I, I've heard people speculate something similar 612 00:43:43,965 --> 00:43:49,665 like, is Chat, is OpenAI holding out the best version of 5.2 for chat? 613 00:43:49,665 --> 00:43:51,165 TPT and I? 614 00:43:51,165 --> 00:43:52,395 I can't say for certain. 615 00:43:53,685 --> 00:43:56,445 Uh, I would say that I think, I think. 616 00:43:57,870 --> 00:44:03,360 The APIs that the foundation models that the model providers expose to, 617 00:44:04,350 --> 00:44:07,320 to everyone else are pretty good. 618 00:44:08,190 --> 00:44:11,700 Uh, I haven't been disappointed with them. 619 00:44:11,760 --> 00:44:17,850 I think a lot of, I mean, I think a lot of the execution of a product comes down to 620 00:44:20,130 --> 00:44:23,160 integrating it well into the workflow, providing it with the 621 00:44:23,160 --> 00:44:26,190 right context, being able to. 622 00:44:26,625 --> 00:44:31,335 Um, make modifications meaningfully and correctly to your 623 00:44:31,335 --> 00:44:33,225 documents, that sort of thing. 624 00:44:33,285 --> 00:44:38,984 And then as I mentioned a moment ago, I also think prompting, uh, there's 625 00:44:38,984 --> 00:44:45,585 quite a bit of alpha and prompting and as well, um, yeah, so in terms of 626 00:44:45,585 --> 00:44:49,815 like the pure model quality, perhaps they're, you know, they're holding 627 00:44:49,815 --> 00:44:51,404 out the better one for themselves. 628 00:44:52,455 --> 00:44:57,254 That to me doesn't seem like a, a major impediment for startups. 629 00:44:57,765 --> 00:45:01,575 Yeah, and you know, I think with Microsoft, I ha I speculate, I 630 00:45:01,575 --> 00:45:04,695 think what Microsoft is doing is they're doing some token throttling. 631 00:45:05,174 --> 00:45:10,095 'cause you can take the exact same prompt in copilot and run it. 632 00:45:10,095 --> 00:45:11,475 And I did side by side. 633 00:45:11,475 --> 00:45:12,884 I actually PO posted on LinkedIn. 634 00:45:12,884 --> 00:45:13,695 Examples of it. 635 00:45:14,055 --> 00:45:19,125 Just a really quick one is I asked for the last I all time 636 00:45:19,515 --> 00:45:22,424 NCAA men's basketball winners. 637 00:45:23,010 --> 00:45:24,480 And the opponents. 638 00:45:24,990 --> 00:45:29,190 And when I ran it in chat pt, it gave me things I didn't even ask for, like the 639 00:45:29,190 --> 00:45:34,170 venue, the score, and it went back all the way to the beginning, which is like in 640 00:45:34,170 --> 00:45:40,710 the 1930s when I ran that same prompt in copilot, it only gave me the last 20 years 641 00:45:41,010 --> 00:45:44,880 and it just gave me, um, like the winner. 642 00:45:45,120 --> 00:45:45,600 I see. 643 00:45:45,600 --> 00:45:46,350 Um, yeah. 644 00:45:46,620 --> 00:45:51,360 So if I had to speculate, I would guess that. 645 00:45:53,070 --> 00:45:57,990 Copilot is probably not using the, whatever the current 646 00:45:57,990 --> 00:46:00,480 top tier OpenAI model is. 647 00:46:00,480 --> 00:46:03,060 Five two reasoning. 648 00:46:03,630 --> 00:46:06,270 Uh, they're probably using a faster, cheaper model. 649 00:46:06,450 --> 00:46:06,540 Mm-hmm. 650 00:46:06,780 --> 00:46:13,050 And I think the, the incentives are a bit different for OpenAI versus Microsoft. 651 00:46:13,380 --> 00:46:17,550 So like OpenAI is competing with Anthropic and Gemini right now. 652 00:46:17,880 --> 00:46:21,300 They wanna make sure that consumers who go on their app are getting like the best. 653 00:46:21,885 --> 00:46:28,725 Most intelligent response, whereas when you're in the context of using a Microsoft 654 00:46:28,725 --> 00:46:37,005 product copilot, there's not really like an alternative to copilot that you can 655 00:46:37,395 --> 00:46:42,135 super easily switch to in the way that you can open up a cloud window if you're 656 00:46:42,135 --> 00:46:44,085 not satisfied with the chat response. 657 00:46:45,060 --> 00:46:45,330 Yeah. 658 00:46:45,425 --> 00:46:47,370 That, that would be my guess is what's going on there. 659 00:46:47,640 --> 00:46:48,750 Yeah, that's a good point. 660 00:46:49,350 --> 00:46:53,040 Um, okay, we're almost outta time, but I did wanna ask you one, one more question. 661 00:46:53,040 --> 00:46:53,130 Sure. 662 00:46:53,310 --> 00:46:59,730 And that was about vibe coding and um, I have seen it is on fire right now in 663 00:46:59,730 --> 00:47:02,040 legal and I think it's so cool to see. 664 00:47:02,340 --> 00:47:02,550 Yeah. 665 00:47:02,610 --> 00:47:07,650 Um, I have a strong bias towards it's great for prototyping and 666 00:47:07,740 --> 00:47:12,300 shortening the software development lifecycle by compressing requirements, 667 00:47:12,300 --> 00:47:13,260 gathering and refinement. 668 00:47:14,295 --> 00:47:18,075 Um, but in that capacity, yeah. 669 00:47:18,735 --> 00:47:23,325 But in no way are these vibe coded apps ready to go be deployed. 670 00:47:23,325 --> 00:47:28,155 You, you literally can't deploy a vibe coded app in your enterprise and maintain 671 00:47:28,155 --> 00:47:31,395 soc two, type two, and, um, not right now. 672 00:47:31,400 --> 00:47:31,420 No. 673 00:47:31,725 --> 00:47:32,355 No, not right. 674 00:47:32,355 --> 00:47:32,595 Yeah. 675 00:47:32,865 --> 00:47:34,935 There's, there's requirements associated with that. 676 00:47:34,935 --> 00:47:39,165 But like, I'm curious just, uh, what's your, what's your take is on, like, 677 00:47:39,195 --> 00:47:43,125 I, there, there's people vibe, coding, like tabular review, and a lot of these. 678 00:47:43,725 --> 00:47:47,444 Features that, um, some of these vertical solutions hang their hat on. 679 00:47:47,444 --> 00:47:49,785 What's your, what's your take on the whole vibe coding movement? 680 00:47:51,435 --> 00:47:52,604 I think it's the future. 681 00:47:52,694 --> 00:47:58,785 I mean, the first time I used cloud code, like March of last 682 00:47:58,785 --> 00:48:01,484 year, it completely blew me away. 683 00:48:01,634 --> 00:48:07,095 I, before that I was pretty skeptical of claims that, you know, a AI was 684 00:48:07,095 --> 00:48:09,854 going to write all my code for me. 685 00:48:09,854 --> 00:48:13,035 'cause it wa like, it just, up until that point it just did not do that. 686 00:48:13,605 --> 00:48:15,645 But cloud code really changed the game. 687 00:48:16,215 --> 00:48:19,005 And I mean, even since then it's gotten so much better. 688 00:48:19,665 --> 00:48:22,725 Um, and so, I mean, I do think it's the future. 689 00:48:22,785 --> 00:48:25,905 How do I expect that future to play out? 690 00:48:26,595 --> 00:48:32,505 So for one thing, it wasn't until maybe six weeks ago when Gemini three was 691 00:48:32,620 --> 00:48:39,135 released and I used Google AI Studio for the first time that, so AI Studio is like. 692 00:48:39,930 --> 00:48:41,520 It takes vibe coding a step further. 693 00:48:41,520 --> 00:48:43,590 It does the full application end-to-end. 694 00:48:43,590 --> 00:48:46,770 You don't even look at a line of code and you can deploy it with a click of button. 695 00:48:46,860 --> 00:48:49,050 It's what like companies like Rep do. 696 00:48:50,250 --> 00:48:57,240 And I was able to build myself fairly non-trivial personal applications 697 00:48:57,240 --> 00:49:01,080 for my development workflow that I know use on a day-to-day basis. 698 00:49:01,590 --> 00:49:03,450 And like, I don't need to hook them up to the internet. 699 00:49:03,450 --> 00:49:07,800 Like I'm not, like, I'm not doing anything too sensitive, so I'm not concerned. 700 00:49:08,610 --> 00:49:12,030 Ultimately a ton about their reliability and security. 701 00:49:12,660 --> 00:49:14,610 Uh, but they are genuinely useful. 702 00:49:15,270 --> 00:49:20,250 And so, I mean, I think it starts with, yeah, like you can build your own tabular 703 00:49:20,250 --> 00:49:25,235 review 'cause it's, it's not a very complicated product and use it yourself. 704 00:49:26,640 --> 00:49:32,010 If we're going to be talking about hosting though, then it 705 00:49:32,010 --> 00:49:33,360 certainly gets more complicated. 706 00:49:33,360 --> 00:49:35,250 No, you can't just throw a coated. 707 00:49:35,685 --> 00:49:42,134 Project out into the world or, or on your firm's, um, you know, private cloud. 708 00:49:43,035 --> 00:49:47,384 The, the way I think about it is, I can imagine there being a few different 709 00:49:47,384 --> 00:49:49,935 phases towards it rolling out. 710 00:49:49,995 --> 00:49:55,455 So for one, a lot of firms have practice innovation teams that have already 711 00:49:55,455 --> 00:49:59,595 been in the practice of developing tools and uh, you know, when they 712 00:49:59,595 --> 00:50:02,895 decide to build or buy, they would be the people who are doing the building. 713 00:50:03,689 --> 00:50:08,310 So I think the bill of buy come calculus can change a lot. 714 00:50:08,310 --> 00:50:11,160 I think those product, those teams can become way more productive. 715 00:50:11,609 --> 00:50:15,540 And of course the things that they produce are still going to be subject to the same 716 00:50:15,540 --> 00:50:19,799 code review processes, the same security review processes, it, review, et cetera. 717 00:50:20,790 --> 00:50:23,399 Um, so that's like short term effect. 718 00:50:23,640 --> 00:50:24,029 Yeah. 719 00:50:24,029 --> 00:50:27,330 I don't see any reason why a law firm couldn't build their own tabular review. 720 00:50:27,419 --> 00:50:28,589 It's not that hard. 721 00:50:29,279 --> 00:50:32,399 Um, you know, as assuming they have that type of process in place. 722 00:50:33,270 --> 00:50:40,560 Long term, I would imagine that the private cloud infrastructure of law firms, 723 00:50:40,589 --> 00:50:46,950 but then enterprises more broadly, will probably adapt to the possibilities of 724 00:50:46,950 --> 00:50:52,680 five coding, like one can imagine in individual lawyer or employee building 725 00:50:52,680 --> 00:50:59,430 their own application and uh, having some kind of streamlined review process that. 726 00:51:00,000 --> 00:51:03,810 It checks up on with automated security checks and then it gets 727 00:51:03,810 --> 00:51:08,670 deployed into like, uh, an isolated container within their private cloud. 728 00:51:09,330 --> 00:51:19,440 Um, I would be shocked if 10 years from now there aren't avenues for 729 00:51:19,500 --> 00:51:25,440 individual employees to be making meaningful contribute contributions to. 730 00:51:27,405 --> 00:51:28,905 Building software with ai. 731 00:51:29,445 --> 00:51:31,035 Yeah, yeah, I agree with that. 732 00:51:31,335 --> 00:51:34,275 Just I'm gonna plug info dash for a second. 733 00:51:34,275 --> 00:51:37,215 We have a integration layer, it's called Integration hub. 734 00:51:38,100 --> 00:51:43,680 We, we, we build it because it was, um, when we were, it hydrates all 735 00:51:43,680 --> 00:51:47,760 of our web parts and SharePoint for internet and extranet scenarios. 736 00:51:48,240 --> 00:51:51,030 But it's an Azure appliance and it has tentacles into 737 00:51:51,030 --> 00:51:52,500 all the back office systems. 738 00:51:53,040 --> 00:51:56,340 And it presents a unified API that security trimmed. 739 00:51:56,820 --> 00:52:01,170 And if you have your walls integrated, it will respect ethical wall boundaries. 740 00:52:01,440 --> 00:52:04,650 So we think it's like the perfect foundation because that's the 741 00:52:04,650 --> 00:52:06,360 hardest part is getting to the data. 742 00:52:07,470 --> 00:52:08,880 You know, you've got different APIs. 743 00:52:08,880 --> 00:52:10,080 Those APIs change. 744 00:52:10,080 --> 00:52:11,580 Some of them aren't well documented. 745 00:52:12,060 --> 00:52:17,370 So if you can tap into our API, if there's any firms out there, we're looking 746 00:52:17,370 --> 00:52:19,860 for firms who want to explore this. 747 00:52:19,860 --> 00:52:23,520 The reality is most of them aren't ready for that. 748 00:52:23,550 --> 00:52:28,920 'cause I mean, we have currently, we have like 25% of the AM law as clients already. 749 00:52:28,950 --> 00:52:34,020 So this is already just one out 4:00 AM law firms already have this ready to go. 750 00:52:35,130 --> 00:52:40,530 Most firms just aren't willing to let their people do this yet, but, 751 00:52:40,620 --> 00:52:44,520 um, we think they will, hopefully in the future, give the time. 752 00:52:44,910 --> 00:52:45,270 Yeah. 753 00:52:45,780 --> 00:52:48,060 So, uh, well this has been a great conversation, man. 754 00:52:48,090 --> 00:52:52,740 Um, before we go, how do people find out more about you and your product? 755 00:52:52,860 --> 00:52:54,480 Um, or, or get in touch? 756 00:52:55,380 --> 00:52:56,190 Yeah, sure. 757 00:52:56,190 --> 00:52:59,759 So, uh, again, our product is version story. 758 00:53:00,270 --> 00:53:04,020 So you can check us out@versionstory.com or on LinkedIn. 759 00:53:04,620 --> 00:53:11,520 Uh, additionally, I also write a blog on Substack called The Redline. 760 00:53:12,120 --> 00:53:15,960 Um, I believe the domain is the redline.com or the 761 00:53:15,960 --> 00:53:17,730 redline dot version story.com. 762 00:53:18,270 --> 00:53:21,090 If you wanna draw me a line, send me an email free. 763 00:53:21,210 --> 00:53:22,110 Feel free to reach out. 764 00:53:22,110 --> 00:53:24,900 My email isJordan@versionstory.com. 765 00:53:25,695 --> 00:53:26,205 Awesome. 766 00:53:26,625 --> 00:53:27,915 Well, best of luck to you, man. 767 00:53:27,975 --> 00:53:30,915 Um, I'm, you know, I'm pulling for everybody in this space. 768 00:53:30,975 --> 00:53:34,875 Uh, I, I don't think everybody's gonna survive because there's been, 769 00:53:34,875 --> 00:53:37,485 this money has, has flowed so freely. 770 00:53:37,905 --> 00:53:42,525 I think there are gonna be some, um, uh, there's gonna be some 771 00:53:42,525 --> 00:53:44,175 people who don't make the cut. 772 00:53:44,175 --> 00:53:48,255 But, um, yeah, I'm really excited to see where things go and thanks 773 00:53:48,255 --> 00:53:50,925 so much for, for coming and spending some time with me today. 774 00:53:50,925 --> 00:53:51,825 I really appreciate it. 775 00:53:52,350 --> 00:53:52,830 Of course. 776 00:53:52,890 --> 00:53:53,880 Uh, thank you, Ted. 777 00:53:53,880 --> 00:53:55,290 This has been a lot of fun. 778 00:53:55,710 --> 00:53:56,220 Awesome. 779 00:53:56,280 --> 00:53:57,960 All right, we'll, uh, we'll chat soon. 780 00:53:58,440 --> 00:53:58,950 Sounds good. 781 00:53:59,130 --> 00:53:59,460 All right. 782 00:53:59,460 --> 00:53:59,910 Take care. 783 00:54:00,180 --> 00:54:02,430 Thanks for listening to Legal Innovation Spotlight. 784 00:54:02,970 --> 00:54:06,450 If you found value in this chat, hit the subscribe button to be notified 785 00:54:06,450 --> 00:54:07,950 when we release new episodes. 786 00:54:08,460 --> 00:54:11,160 We'd also really appreciate it if you could take a moment to rate 787 00:54:11,160 --> 00:54:13,770 us and leave us a review wherever you're listening right now. 788 00:54:14,370 --> 00:54:17,070 Your feedback helps us provide you with top-notch content. 00:00:02,880 Jordan, thanks for joining us this afternoon. 2 00:00:03,090 --> 00:00:04,019 It's good to have you on. 3 00:00:04,350 --> 00:00:05,460 It's great to be here, Ted. 4 00:00:05,670 --> 00:00:09,420 Thank you for inviting me to the preeminent legal Tech podcast. 5 00:00:09,930 --> 00:00:11,550 Yes, thank you very much. 6 00:00:11,640 --> 00:00:16,590 Uh, I think we're one of the few, if not maybe even the only, that is 7 00:00:16,590 --> 00:00:18,450 solely focused on legal innovation. 8 00:00:18,930 --> 00:00:23,820 So, um, which is a hot topic these days with the way AI 9 00:00:23,820 --> 00:00:25,020 is kind of shaking things up. 10 00:00:25,620 --> 00:00:25,950 Yeah. 11 00:00:28,275 --> 00:00:34,575 Um, yeah, I, I've been in the legal tech space for five years now, and I think 12 00:00:34,725 --> 00:00:38,985 it's really exciting to see that there's actually appetite for the industry now, 13 00:00:39,315 --> 00:00:43,935 because when we started in 2021, there, there weren't like legal tech podcasts. 14 00:00:43,935 --> 00:00:48,375 There wasn't, there wasn't an audience for, for that type of thing. 15 00:00:48,585 --> 00:00:52,695 So, yeah, it, you know, it's, it's funny, legal innovation. 16 00:00:53,385 --> 00:00:57,915 I joke and say, really used to be an oxymoron, like jumbo shrimp 17 00:00:57,915 --> 00:01:03,074 or military intelligence, but it, the market's coming around it. 18 00:01:03,074 --> 00:01:03,135 Um. 19 00:01:04,290 --> 00:01:09,300 You know, we, we jokingly would call the CINO role, which is Chief 20 00:01:09,300 --> 00:01:14,520 Innovation Officers Chief in name only, because so many in the early 21 00:01:14,520 --> 00:01:16,650 days were just not properly empowered. 22 00:01:16,710 --> 00:01:20,100 They had good people they would put in those roles, but they just 23 00:01:20,100 --> 00:01:22,470 wouldn't properly empower 'em. 24 00:01:22,710 --> 00:01:23,280 Um. 25 00:01:24,045 --> 00:01:26,835 You and I got connected because you, you wrote an article that I thought 26 00:01:26,835 --> 00:01:29,775 was, was really good and provocative. 27 00:01:30,345 --> 00:01:34,185 Uh, and I like when people push the boundaries a little bit 28 00:01:34,215 --> 00:01:35,865 and challenge the status quo. 29 00:01:36,405 --> 00:01:40,425 And your article was called Why Can't, 4.3 billion. 30 00:01:40,899 --> 00:01:45,640 In legal AI investment, outcompete $20 a month for chat CPT. 31 00:01:46,119 --> 00:01:49,780 So, um, we're gonna get into that, but before we do, why don't 32 00:01:49,780 --> 00:01:53,440 you tell us kind of who you are, what you do, and where you do it? 33 00:01:54,190 --> 00:01:54,670 Sure thing. 34 00:01:55,360 --> 00:01:56,289 So my name is Jordan. 35 00:01:56,350 --> 00:02:01,630 I'm a co-founder and CTO of version story and we are a startup 36 00:02:01,720 --> 00:02:03,610 building get for documents. 37 00:02:04,060 --> 00:02:06,729 So we're trying to solve legal version control. 38 00:02:07,125 --> 00:02:11,055 We're trying to make it so that you as a lawyer, always have confidence about 39 00:02:11,055 --> 00:02:12,405 which version you're working off of. 40 00:02:12,825 --> 00:02:16,515 You are never fearful that you're missing changes when you send a draft 41 00:02:16,515 --> 00:02:18,105 to a counterparty or to your client. 42 00:02:18,885 --> 00:02:25,755 And, uh, the inspiration for our product is Git, which is the defacto 43 00:02:25,755 --> 00:02:28,935 version control for how any software engineer in the world works. 44 00:02:29,715 --> 00:02:33,045 Uh, it's kind of the backbone upon which. 45 00:02:34,049 --> 00:02:38,160 Vibe coding, for example, can be built, is having reliable version control. 46 00:02:38,850 --> 00:02:41,100 Anyway, so that's what we work on. 47 00:02:41,190 --> 00:02:43,799 Uh, I am, like I said, I am the CTO. 48 00:02:44,310 --> 00:02:48,359 We've been working on version story for, uh, a little over five years now. 49 00:02:49,560 --> 00:02:54,329 And, um, we're based out of Brooklyn, New York, and yeah. 50 00:02:56,489 --> 00:02:57,059 Good stuff. 51 00:02:57,059 --> 00:02:57,299 Yeah. 52 00:02:57,299 --> 00:03:01,920 You know, it's interesting, GitHub, I, you know, I'm, I own a software 53 00:03:01,920 --> 00:03:03,690 business, so I've known about gi. 54 00:03:03,690 --> 00:03:08,790 I used to work for Microsoft 26 years ago, so I'm very familiar. 55 00:03:09,930 --> 00:03:14,250 With GitHub, but I feel like GitHub has really gone mainstream. 56 00:03:14,520 --> 00:03:18,269 You know, to use Claude Code, you need GitHub. 57 00:03:18,660 --> 00:03:23,430 Um, right, they, they've now released Cowork, which is 58 00:03:23,430 --> 00:03:25,290 literally like two days old. 59 00:03:25,290 --> 00:03:27,870 By the time this airs, it will have been out for a little while. 60 00:03:27,870 --> 00:03:29,490 But, yeah, it's funny, man. 61 00:03:29,490 --> 00:03:33,450 I see non-technical people now with repos on GitHub because 62 00:03:33,720 --> 00:03:35,070 they want to use cloud code. 63 00:03:35,070 --> 00:03:37,050 So GitHub is now. 64 00:03:37,350 --> 00:03:39,990 Kind of a household name, which is pretty cool. 65 00:03:40,710 --> 00:03:41,460 It is, yeah. 66 00:03:41,550 --> 00:03:48,240 Uh, and it's really powerful and it is ki it is essential to, to vibe coding. 67 00:03:48,330 --> 00:03:53,190 I you one kind of beginner mistake that I've seen from a few times 68 00:03:53,190 --> 00:03:57,240 from non-professional software engineers who dip their toes in. 69 00:03:57,885 --> 00:04:01,785 They don't know that they need to set up Git first and they end up 70 00:04:02,115 --> 00:04:06,225 having cloud code go in there and rewrite their whole code base, delete 71 00:04:06,225 --> 00:04:07,605 files, and now they can't go back. 72 00:04:08,115 --> 00:04:13,725 Um, and so it really is imperative to the vibe coding workflow. 73 00:04:14,445 --> 00:04:21,615 And so I think what we're building, uh, is going to similarly be imperative for truly 74 00:04:21,615 --> 00:04:24,885 unlocking AI empowered legal drafting. 75 00:04:25,680 --> 00:04:26,580 Very cool man. 76 00:04:27,270 --> 00:04:29,909 Well, let's talk a little bit about, uh, about the article. 77 00:04:29,909 --> 00:04:34,229 So again, the title was WA Can't, 4.3 billion in Legal AI Investment, 78 00:04:34,229 --> 00:04:37,320 outcompete Chachi PT for 20 bucks a month. 79 00:04:37,919 --> 00:04:42,240 And, um, you know, that's a very attention grabbing headline and I think 80 00:04:42,240 --> 00:04:50,159 it's one, it touches on a nerve that a lot of people have been, you know, uh, 81 00:04:50,190 --> 00:04:54,000 pontificating on, which is, you know, how. 82 00:04:54,405 --> 00:04:55,455 Where's the moat? 83 00:04:56,474 --> 00:04:59,805 Um, you're too young to remember, but in the eighties, may maybe 84 00:04:59,805 --> 00:05:03,735 you've, you've seen reruns of the commercial, uh, for Wendy's. 85 00:05:03,735 --> 00:05:06,224 There used to be an old lady and she'd say, where's the beef? 86 00:05:06,974 --> 00:05:10,995 Um, well, investors are going, where's the moat? 87 00:05:11,505 --> 00:05:14,835 Um, but they still keep, they keep writing checks at, right. 88 00:05:14,835 --> 00:05:18,585 You know, a DX valuations so clearly. 89 00:05:18,990 --> 00:05:22,620 They're willing to look past the absence of a moat, which I 90 00:05:22,830 --> 00:05:24,750 don't think really exists today. 91 00:05:25,110 --> 00:05:30,630 I think the moat is really distribution and install base, right? 92 00:05:31,170 --> 00:05:38,005 And, um, that's, and it's not a bad strategy because I think when you're 93 00:05:38,005 --> 00:05:41,100 first to market and market, if you can, if you can generate enough 94 00:05:41,100 --> 00:05:45,480 momentum and create a large enough install base, it gives you time. 95 00:05:46,140 --> 00:05:47,310 To figure out your mode. 96 00:05:47,460 --> 00:05:51,150 And a lot of businesses, some businesses do that before they 97 00:05:51,150 --> 00:05:52,650 even figure out how to make money. 98 00:05:53,040 --> 00:05:54,150 I mean, look at Facebook. 99 00:05:54,390 --> 00:05:57,150 Facebook didn't have a revenue model for years. 100 00:05:57,599 --> 00:06:01,409 Git like GitHub didn't have a revenue model. 101 00:06:01,890 --> 00:06:07,050 Um, it was really catering to a lot of open source, um, development projects. 102 00:06:07,080 --> 00:06:09,030 And, you know, over time. 103 00:06:09,570 --> 00:06:11,250 You, you can kind of figure that out. 104 00:06:11,250 --> 00:06:16,920 But tell us just a high level what the, what the premise and the 105 00:06:16,920 --> 00:06:19,350 motivation for, for the article was. 106 00:06:20,250 --> 00:06:20,790 Sure. 107 00:06:20,880 --> 00:06:27,630 So I'll start with the motivation and kind of what led me to write it. 108 00:06:28,830 --> 00:06:34,920 Uh, I think a lot of the ideas in the piece were stewing for quite a while, 109 00:06:35,100 --> 00:06:36,780 uh, but it kind of clicked for me. 110 00:06:37,575 --> 00:06:42,165 After having a few conversations with some friends of mine, these friends of 111 00:06:42,165 --> 00:06:45,075 mine are software engineers and founders. 112 00:06:45,344 --> 00:06:49,995 I would characterize the way they think as being fairly reflective of like the 113 00:06:49,995 --> 00:06:52,215 Silicon Valley mindset, if you will. 114 00:06:52,724 --> 00:06:54,344 I mean, they're, they're on tech Twitter. 115 00:06:54,795 --> 00:06:59,230 Um, so like, I kind of think of them as, um. 116 00:07:01,484 --> 00:07:05,354 Yeah, kind of a reflection of what, what Silicon Valley is, 117 00:07:05,414 --> 00:07:06,734 is thinking at any given time. 118 00:07:07,455 --> 00:07:13,094 And I noticed that over the course of like a couple months this past summer, 119 00:07:13,484 --> 00:07:16,664 I, I felt like I was having the same conversation over with a few of them 120 00:07:17,234 --> 00:07:24,075 where they were asking me like, how worried are you about like these, 121 00:07:24,554 --> 00:07:26,895 these big name legal AI companies are. 122 00:07:30,315 --> 00:07:32,085 Don't you need to be doing the same thing? 123 00:07:32,955 --> 00:07:41,085 Um, and I, I kind of felt like, well, I don't exactly think we 124 00:07:41,085 --> 00:07:43,515 are approaching the same problems. 125 00:07:43,664 --> 00:07:46,455 We're, we're not really trying to do the same thing. 126 00:07:47,145 --> 00:07:52,424 Uh, but the, the point that these friends kept reemphasizing is 127 00:07:52,424 --> 00:07:57,554 that in, in the AI world, all that matters is distribution and that. 128 00:07:58,350 --> 00:08:05,640 It's really a winner take all market where you need to have maximum 129 00:08:05,640 --> 00:08:08,430 distribution and product is secondary. 130 00:08:08,730 --> 00:08:11,880 That was like kind of the, the repeated idea. 131 00:08:11,880 --> 00:08:17,250 And so I, I kind of, um, was curious about this and I, and I was pushing 132 00:08:17,250 --> 00:08:22,410 back and trying to understand where they were coming from because disagreed. 133 00:08:24,825 --> 00:08:27,794 The way we built our company is starting with building product first, 134 00:08:27,794 --> 00:08:31,245 making sure it's really solid, and then figuring out distribution later. 135 00:08:32,595 --> 00:08:40,485 And the idea that kept resurfacing is that, um, it is a land grab 136 00:08:40,485 --> 00:08:46,694 opportunity because all AI companies are betting on the future 137 00:08:46,694 --> 00:08:48,525 improvements of foundation models. 138 00:08:49,215 --> 00:08:54,915 So the move is you capture distribution as quickly as possible so that once 139 00:08:55,185 --> 00:09:00,675 OpenAI releases whatever, you know, the model is that unlocks a GI or 140 00:09:00,675 --> 00:09:02,415 something similar to a TPT seven. 141 00:09:02,805 --> 00:09:07,965 You have, you are in the, the market position to be ushering in like the 142 00:09:07,965 --> 00:09:10,275 total revolution of the industry. 143 00:09:10,965 --> 00:09:13,455 Um, and yeah, once. 144 00:09:14,760 --> 00:09:15,810 That clicked with me. 145 00:09:15,810 --> 00:09:24,510 I think a lot of other dimensions of the big player strategies made sense and, 146 00:09:24,720 --> 00:09:29,790 um, aligned with my understanding of, of how the venture capital world worked. 147 00:09:30,270 --> 00:09:33,325 Uh, my experiences with venture capitalists over the years, et cetera. 148 00:09:35,055 --> 00:09:37,334 Yeah, so well, let's talk a little bit about that. 149 00:09:37,334 --> 00:09:40,665 So we are largely bootstrapped here at info dash. 150 00:09:40,665 --> 00:09:45,375 We've taken a little bit of, I'll call it, uh, we've done a couple of small seed 151 00:09:45,375 --> 00:09:53,805 rounds with the legal tech fund, and we're mostly, um, founder, founder backed, and. 152 00:09:55,110 --> 00:10:00,690 I have a natural aversion to especially the West coast kind 153 00:10:00,690 --> 00:10:07,230 of VC mindset, which is tends to lean towards growth at all costs. 154 00:10:07,740 --> 00:10:12,420 And there's a very limited focus on capital efficiency. 155 00:10:12,930 --> 00:10:18,150 And as an older founder that you know somebody who's lived through, um. 156 00:10:19,335 --> 00:10:20,880 Thin times, we'll call 'em. 157 00:10:21,105 --> 00:10:27,735 Um, I, I know how valuable resources are and I I'm not willing to trade, 158 00:10:28,905 --> 00:10:35,205 you know, my cap table for a hope and a prayer that I can achieve the growth. 159 00:10:35,385 --> 00:10:40,785 I wanna be able to demonstrate and systematize my mechanisms for growth. 160 00:10:40,785 --> 00:10:44,055 And then if I can raise money and I've got a predictable and 161 00:10:44,055 --> 00:10:46,965 demonstratable go to market machine. 162 00:10:47,625 --> 00:10:55,545 Then I feel good about A, taking the investment and, and b, my willingness to, 163 00:10:55,635 --> 00:10:59,235 or my ability to deliver to expectations. 164 00:10:59,895 --> 00:11:06,944 And, um, but the VC mindset oftentimes is like, you know, we. 165 00:11:07,335 --> 00:11:10,155 We get routinely solicited. 166 00:11:10,185 --> 00:11:12,314 When I say routinely, I've got a spreadsheet. 167 00:11:12,705 --> 00:11:14,445 My listeners have heard me talk about this. 168 00:11:14,445 --> 00:11:16,605 I think there are, there are over 40 bcs. 169 00:11:16,665 --> 00:11:19,905 I started not, not like touches by bcs. 170 00:11:20,025 --> 00:11:22,845 40 different funds have reached out to us. 171 00:11:22,875 --> 00:11:26,415 I started tracking it right before ilta, so that would've been like 172 00:11:26,415 --> 00:11:28,485 late summer, like so mid August. 173 00:11:28,995 --> 00:11:30,675 So in four months. 174 00:11:30,765 --> 00:11:31,365 Um. 175 00:11:33,180 --> 00:11:40,650 Roughly 10 different funds a month and they keep, and it's, uh, it's, it's 176 00:11:40,650 --> 00:11:44,730 really indicative of how aggressive. 177 00:11:45,330 --> 00:11:45,570 Yeah. 178 00:11:45,810 --> 00:11:50,070 You know, and it's just like, but I think you had a good point, which is like the 179 00:11:50,070 --> 00:11:54,840 VCs really favor, like moonshots over building a viable, sustainable business. 180 00:11:54,840 --> 00:11:56,370 Do you, is that your take as well? 181 00:11:56,775 --> 00:11:59,895 It is, and I think it's worth noting that it didn't use to be that way. 182 00:12:00,225 --> 00:12:04,785 Uh, VCs reaching out to legal tech startups wanting to invest Oh, yeah. 183 00:12:05,925 --> 00:12:06,585 In particular. 184 00:12:07,605 --> 00:12:13,965 And so, um, the reason for that is I think downstream of the incentives 185 00:12:13,995 --> 00:12:16,065 of the two industries respectively. 186 00:12:17,790 --> 00:12:22,155 So VCs are incentivized to make moonshots. 187 00:12:23,235 --> 00:12:28,694 Uh, the way that their portfolio map works for those who are listening who perhaps 188 00:12:28,694 --> 00:12:33,585 aren't as familiar with venture capital, is that a venture capitalist, a venture 189 00:12:33,585 --> 00:12:39,495 capital firm, will raise a fund and they will make a limited number of bets on, 190 00:12:39,495 --> 00:12:47,655 say, 20 startups with the expectation that 18 of them will go to zero, and 191 00:12:47,655 --> 00:12:52,545 one or two of them will be worth tens or hundreds of billions of dollars, and. 192 00:12:53,175 --> 00:12:57,315 They're carry that the percentage of the profits that the firm, 193 00:12:57,585 --> 00:13:00,735 uh, earns will return the fund. 194 00:13:02,175 --> 00:13:06,555 And so because of that, VCs structurally aren't interested 195 00:13:06,555 --> 00:13:16,695 in viable businesses that are sub billion dollar outcomes, and they are. 196 00:13:18,420 --> 00:13:23,100 Willing to make, to take a lot of risk if it has some percentage 197 00:13:23,100 --> 00:13:24,840 chance of being the big outcome. 198 00:13:26,220 --> 00:13:31,800 I'll, I'll, I'll note that another dynamic that's worth considering is that VCs 199 00:13:32,250 --> 00:13:39,120 receive a compensation of assets under management as well, and they also will 200 00:13:39,689 --> 00:13:46,199 usually raise their next fund while a fund is, their current fund is still active. 201 00:13:47,250 --> 00:13:52,319 The way that they will raise that fund is by reporting to their investors, the 202 00:13:52,319 --> 00:13:58,290 limited partners, what the returns of the fund are, and so they have a, a short term 203 00:13:58,290 --> 00:14:03,720 incentive for their portfolio companies to go on and raise a new round, 18 months 204 00:14:03,720 --> 00:14:06,209 after investment at a five x markup. 205 00:14:06,615 --> 00:14:08,840 That helps them raise their new fund fund. 206 00:14:08,845 --> 00:14:14,445 So VCs have a short term incentive as well as a big risk taking incentive, 207 00:14:14,715 --> 00:14:20,025 and that combination is diametrically opposed to the way lawyers work as 208 00:14:20,355 --> 00:14:22,485 sure most people listening understand. 209 00:14:22,995 --> 00:14:30,105 Uh, lawyers are naturally risk averse as a profession, which completely makes sense. 210 00:14:30,165 --> 00:14:32,200 Their upside is capped by. 211 00:14:32,895 --> 00:14:39,405 Whatever fees they've negotiated, uh, for a deal, um, by, you know, hourly 212 00:14:39,405 --> 00:14:42,915 part, hourly billing rates, et cetera. 213 00:14:43,155 --> 00:14:46,635 Um, they have uncapped downside. 214 00:14:46,665 --> 00:14:50,475 However, if they make a mistake that exposes their client to, you 215 00:14:50,475 --> 00:14:51,795 know, hundreds of millions dollars. 216 00:14:53,535 --> 00:14:57,975 Uh, so traditionally the. 217 00:14:58,410 --> 00:15:02,819 And then another component about the legal industry is that it, it takes a while 218 00:15:02,819 --> 00:15:04,949 to build a good product in, in legal. 219 00:15:05,520 --> 00:15:08,790 Uh, working with Doc XFiles is really technically challenging. 220 00:15:09,209 --> 00:15:12,599 Gaining the trust of lawyers, uh, takes a while. 221 00:15:12,930 --> 00:15:19,170 So traditionally, the legal industry has not been particularly 222 00:15:19,170 --> 00:15:21,089 favored by venture capital. 223 00:15:21,120 --> 00:15:23,615 When we went through Y Combinator, winner of 21. 224 00:15:25,755 --> 00:15:30,255 Pretty prominent venture capitalist was, was speaking with us and advised 225 00:15:30,255 --> 00:15:34,005 all of the legal tech companies in are bashed to pivot to a different industry. 226 00:15:35,145 --> 00:15:37,785 Um, so the times have changed. 227 00:15:38,475 --> 00:15:42,345 Yeah, it's, and you know what, it wasn't bad advice in 2021. 228 00:15:42,645 --> 00:15:47,355 I can tell you, as somebody who sold and sold technology into law firms 229 00:15:47,355 --> 00:15:51,195 for many years, it's a, it's a slog. 230 00:15:52,935 --> 00:15:56,324 Has been, yeah, it's difficult and you know, started. 231 00:15:59,715 --> 00:16:01,065 Had some difficult lessons. 232 00:16:01,065 --> 00:16:01,905 We had to learn. 233 00:16:02,145 --> 00:16:06,975 You have to eat table scraps for years, potentially to build the 234 00:16:06,975 --> 00:16:11,535 resume, to just even get the at bats at the top of the AM law. 235 00:16:11,925 --> 00:16:12,075 Yep. 236 00:16:12,105 --> 00:16:17,595 And you know, um, the Harveys and LARAs of the world and others who've, 237 00:16:17,835 --> 00:16:23,365 um, demonstrated incredible growth out of the gate have really kind of sh. 238 00:16:23,880 --> 00:16:24,750 Took a shortcut. 239 00:16:25,020 --> 00:16:28,980 They've shortcutted that that process, that historically has been in place. 240 00:16:29,400 --> 00:16:33,420 And my theory on how they've been able to do that is really twofold. 241 00:16:33,600 --> 00:16:34,980 Two main drivers. 242 00:16:35,490 --> 00:16:38,580 One is the VCs in their cap table. 243 00:16:39,360 --> 00:16:43,770 So if you look at Harvey, for example, I'm more familiar with their cap table. 244 00:16:43,860 --> 00:16:51,960 OpenAI, Sequoia, A 16 Z, I mean the who's who of Silicon Valley vc. 245 00:16:52,590 --> 00:16:53,160 And. 246 00:16:54,840 --> 00:17:00,330 A, those funds are big buyers of legal services, right? 247 00:17:00,360 --> 00:17:07,650 All their Port cos and, um, all their m and a work, and they have pull 248 00:17:07,950 --> 00:17:13,470 in the legal market that where they can facilitate, um, introductions. 249 00:17:13,890 --> 00:17:18,270 It's also a very strong signal of credibility when 250 00:17:18,270 --> 00:17:19,650 you come to the table with. 251 00:17:20,040 --> 00:17:24,930 Open AI and Google Ventures and Sequoia and a 16 Z in your cap table. 252 00:17:24,930 --> 00:17:25,020 Sure. 253 00:17:25,290 --> 00:17:30,360 It gives you in instant credibility and lawyers are very much focused 254 00:17:30,360 --> 00:17:37,770 on, um, precedent and brand and if you come and prestige. 255 00:17:37,950 --> 00:17:40,290 That's how the, that's how the law firm market works. 256 00:17:40,290 --> 00:17:43,110 Like the, the, the firms at the top of the AM law. 257 00:17:43,560 --> 00:17:47,430 How they, how they're able to get those rates are, it's, it's really 258 00:17:47,430 --> 00:17:49,050 their prestige and their brand. 259 00:17:50,264 --> 00:17:54,254 So that that translates really well into legal. 260 00:17:54,554 --> 00:17:57,615 And then the other dynamic that has allowed them to 261 00:17:57,615 --> 00:17:59,985 shortcut that process is fomo. 262 00:18:00,554 --> 00:18:02,865 There is a tremendous amount of fomo. 263 00:18:02,865 --> 00:18:06,705 There's this existential threat out there that law firm leadership 264 00:18:06,705 --> 00:18:09,705 looks at AI and is implement. 265 00:18:10,365 --> 00:18:14,865 A tremendous amount of change and transformation within the industry, and 266 00:18:14,865 --> 00:18:18,435 we don't really know how it's gonna shake out, but one thing we do know, as law firm 267 00:18:18,435 --> 00:18:20,235 leaders, we better be doing something. 268 00:18:20,595 --> 00:18:24,495 So let's just go buy something regardless if it's a fit. 269 00:18:24,500 --> 00:18:27,465 Do, do you agree with those dynamics or do you see others at play? 270 00:18:28,065 --> 00:18:29,295 Yeah, I completely agree. 271 00:18:29,295 --> 00:18:34,965 I think you, it's hard to underestimate the power of fear as a motivator, 272 00:18:35,535 --> 00:18:38,025 uh, in, in this conversation. 273 00:18:39,570 --> 00:18:46,410 Uh, the narrative that lawyers have been seeing really since the 274 00:18:46,410 --> 00:18:52,800 release of chat GPT, is that AI is marching on a trajectory to pose an 275 00:18:52,800 --> 00:18:54,570 existential threat to their livelihood. 276 00:18:55,590 --> 00:19:01,890 That there are claims that, uh, AI is going to automate legal reasoning, 277 00:19:02,400 --> 00:19:05,550 uh, in a number of years and. 278 00:19:08,490 --> 00:19:16,200 Personally don't agree with those claims, but I think that one can see how tempting 279 00:19:16,200 --> 00:19:23,010 they are for, say somebody like a venture capitalist who is regularly entering into 280 00:19:23,610 --> 00:19:29,100 legal agreements and they are reviewing term sheets on a regular basis, and 281 00:19:29,100 --> 00:19:34,440 then they drop a term sheet into JTPT and you know, it can do a, a pretty 282 00:19:34,440 --> 00:19:36,240 reasonable job of translating legalese. 283 00:19:39,675 --> 00:19:40,545 In a lot of cases. 284 00:19:41,175 --> 00:19:44,745 Now, obviously there are plenty of issues of reliability, of 285 00:19:44,835 --> 00:19:49,095 hallucinating of it, not having proper context to do an adequate job. 286 00:19:50,385 --> 00:19:59,145 But, um, I think it would, it, an idea arose amongst, uh, VCs amongst 287 00:19:59,145 --> 00:20:00,765 the tech community, which then. 288 00:20:02,670 --> 00:20:07,230 Uh, amplified out to the right, wider world and to lawyers in particular, 289 00:20:07,590 --> 00:20:12,780 that AI was, was on trajectory to, to upend the legal industry. 290 00:20:13,380 --> 00:20:17,190 And so if you're a lawyer, if you're a leader at a law firm and you hear this 291 00:20:18,300 --> 00:20:27,284 and I. Perfectly reasonable and expected response to start thinking like, okay, 292 00:20:27,284 --> 00:20:31,304 how can, how can we mitigate this risk? 293 00:20:31,304 --> 00:20:32,504 What steps should we take? 294 00:20:32,865 --> 00:20:37,034 Even if it is, you know, some small percentage that this type of 295 00:20:37,034 --> 00:20:43,845 transformation really happens, uh, how can we make sure that we're in a position 296 00:20:43,845 --> 00:20:49,485 to, to handle the transformation well and. 297 00:20:50,250 --> 00:20:55,050 Law firms aren't, they're not in the business of assessing 298 00:20:55,530 --> 00:21:00,840 the viability of particular startups, legal or AI strategies. 299 00:21:01,320 --> 00:21:01,649 Right. 300 00:21:01,860 --> 00:21:08,490 Um, so from a risk management perspective, the, the strategy. 301 00:21:09,270 --> 00:21:15,120 Makes the most sense is adopt whatever the leading player is, because then, 302 00:21:15,270 --> 00:21:17,879 you know, if the transformation happens, at least you're in the same 303 00:21:17,879 --> 00:21:19,919 boat as all of your competitors. 304 00:21:21,060 --> 00:21:25,889 Yeah, and I mean, historically, more broadly in normal moving markets 305 00:21:25,889 --> 00:21:32,399 and in, in more broadly, you have the garters and the foresters of 306 00:21:32,399 --> 00:21:34,710 the world that they're the ones who. 307 00:21:35,685 --> 00:21:40,665 Evaluate the strategy and the ability to execute on that strategy and put you 308 00:21:40,665 --> 00:21:46,785 on a magic quadrant that this dynamic happened before Gartner had a chance. 309 00:21:46,785 --> 00:21:52,875 I don't know if they've, if they even have, um, a magic quadrant for legal 310 00:21:52,875 --> 00:21:56,475 tools, but they're, they're the ones who typically do that, not lawyers. 311 00:21:57,075 --> 00:21:57,435 Yeah. 312 00:21:57,675 --> 00:21:57,915 Yeah. 313 00:21:57,915 --> 00:21:58,545 Absolutely. 314 00:21:58,635 --> 00:21:59,685 And I, I would argue that. 315 00:22:01,455 --> 00:22:08,325 The fear and market leadership dynamic was, was entering into 316 00:22:08,325 --> 00:22:10,035 place before anyone had a chance. 317 00:22:10,305 --> 00:22:14,835 Um, I mean, for a long time, Harvey's website was, was just a, a sign up 318 00:22:14,835 --> 00:22:17,145 and a wait list for like a year. 319 00:22:17,505 --> 00:22:23,055 And, but somehow in that era they managed to, to build that brand prestige. 320 00:22:23,925 --> 00:22:24,345 Yeah. 321 00:22:25,005 --> 00:22:25,305 And um. 322 00:22:26,669 --> 00:22:27,540 I've seen their product. 323 00:22:27,540 --> 00:22:28,740 It's, it's a good product. 324 00:22:28,860 --> 00:22:30,210 Um, so is Lara's. 325 00:22:30,210 --> 00:22:32,010 I've seen both of them in action. 326 00:22:32,610 --> 00:22:38,490 Um, the one point I wanted to make too, that, that, that you touched 327 00:22:38,490 --> 00:22:44,429 on is like how legal tech was not a good fit historically for, for vc. 328 00:22:45,000 --> 00:22:52,560 And I think there was so much excitement when the AI light bulb went off in. 329 00:22:53,400 --> 00:22:58,560 Investors heads on, wow, what could this do to legal is because, you know, 330 00:22:58,560 --> 00:23:06,180 legal is one of the last holdouts of industries that have successfully 331 00:23:06,180 --> 00:23:09,750 resisted a tech transformation, right? 332 00:23:09,750 --> 00:23:13,440 If you look at almost every other industry, not all, but the 333 00:23:13,440 --> 00:23:17,610 vast majority, there has been a major tech transformation. 334 00:23:17,760 --> 00:23:18,990 Look at financial services. 335 00:23:18,990 --> 00:23:20,220 So that's a world I came from. 336 00:23:20,220 --> 00:23:21,900 I spent 10 years at Bank of America. 337 00:23:22,800 --> 00:23:26,190 They are literally shutting down branches right now because 338 00:23:26,190 --> 00:23:33,120 of the digital experience and services that they can deliver. 339 00:23:33,720 --> 00:23:37,320 Um, that is a massive shift. 340 00:23:37,380 --> 00:23:42,540 They've got, you know, bank of America's in all 50 states and internationally, 341 00:23:42,540 --> 00:23:46,890 and being able to reduce their physical footprint because of technology. 342 00:23:47,160 --> 00:23:48,930 You don't even have to go into a branch. 343 00:23:49,290 --> 00:23:56,669 I had a. Six figure and not a low six figure wire that I went into a branch 344 00:23:56,669 --> 00:24:02,790 to initiate and the guy told me to go home and do it online, uh, because I 345 00:24:02,790 --> 00:24:05,189 didn't set an appointment on the app. 346 00:24:05,189 --> 00:24:06,450 He's like, you can do this for your phone. 347 00:24:06,450 --> 00:24:07,169 I was blown away. 348 00:24:07,169 --> 00:24:09,699 I was like, I can wire hundreds of thousands of dollars on. 349 00:24:10,620 --> 00:24:11,310 Through the app. 350 00:24:11,310 --> 00:24:12,300 He's like, absolutely. 351 00:24:12,629 --> 00:24:13,980 And he was right and I did it. 352 00:24:14,370 --> 00:24:22,409 But legal has been, has successfully held the, the transformation, um, God's away. 353 00:24:22,919 --> 00:24:29,820 And now it's like YY that um, that dynamic, that ability to 354 00:24:29,850 --> 00:24:32,250 resist is, is no longer there. 355 00:24:32,250 --> 00:24:37,050 So I think that there's a ton of pen up demand for efficiency on the buy side 356 00:24:37,439 --> 00:24:39,990 in law firms and VCs recognize that. 357 00:24:40,815 --> 00:24:41,835 Yeah, I think that's right. 358 00:24:42,945 --> 00:24:50,055 Um, well, tell us a little bit about your take on like the, the a GI piece, 359 00:24:50,055 --> 00:24:57,885 because it sounds like you're skeptical of these models and their ability to 360 00:24:58,425 --> 00:25:01,515 deliver on the legal reasoning piece. 361 00:25:01,635 --> 00:25:06,315 Um, am I, am I reading your position correctly? 362 00:25:07,185 --> 00:25:07,635 Um. 363 00:25:09,899 --> 00:25:10,530 Somewhat. 364 00:25:10,530 --> 00:25:16,500 I think I, I'm, I definitely believe in the value of AI and 365 00:25:16,500 --> 00:25:19,260 LLMs in a number of domains. 366 00:25:19,260 --> 00:25:26,040 Like I certainly think that AI can and will be really important to the law. 367 00:25:26,550 --> 00:25:31,199 Uh, to the extent I'm skeptical, I think I'm skeptical of the, 368 00:25:31,530 --> 00:25:38,189 the claims that AI will render lawyers obsolete, that they will. 369 00:25:39,149 --> 00:25:49,110 Um, yeah, the, the more the, the, the a GI claims that there, there isn't 370 00:25:49,110 --> 00:25:52,530 a need for a human lawyer anymore. 371 00:25:53,460 --> 00:25:59,250 And I'm not going to say that I'm, you know, a hundred percent confident that 372 00:25:59,490 --> 00:26:02,129 a GI isn't is theoretically impossible. 373 00:26:02,790 --> 00:26:06,570 Uh, I think from my experience of using LLMs. 374 00:26:07,530 --> 00:26:09,360 Every day for hours a day coding. 375 00:26:10,350 --> 00:26:12,390 I, I have reasons for skepticism. 376 00:26:12,570 --> 00:26:16,050 Um, if, if a DI does happen, then we're all in the same boat. 377 00:26:16,110 --> 00:26:25,065 But setting that aside, um, I think there is reason to think that lawyers 378 00:26:25,205 --> 00:26:32,010 in particular are the, some of the least likely people to be automated from ai. 379 00:26:33,360 --> 00:26:35,610 So one pretty critical difference. 380 00:26:37,199 --> 00:26:45,330 Being a in coding is that with coding and, and, and the difference in, in 381 00:26:45,330 --> 00:26:49,679 respect to how AI can augment your workflow is that with coding you have 382 00:26:50,520 --> 00:26:53,580 built in systems of verifiability. 383 00:26:54,360 --> 00:27:03,060 So if in LLM writes a bunch of code for you, you can test your code and verify. 384 00:27:03,465 --> 00:27:08,145 Whether or not the feature works or passes a suite of tests very quickly. 385 00:27:09,254 --> 00:27:16,845 When Ai, uh, drafts a contract for you, there isn't a system of 386 00:27:16,845 --> 00:27:21,915 verification that's akin to like running a unit test or testing a feature. 387 00:27:22,215 --> 00:27:28,545 Aside from you, the lawyer assiduously reviewing each and every 388 00:27:28,545 --> 00:27:30,795 line that was introduced by the. 389 00:27:33,345 --> 00:27:40,365 So that is not to say that AI drafting can't add value to the workflow. 390 00:27:40,785 --> 00:27:41,895 Uh, I think it can. 391 00:27:41,895 --> 00:27:51,945 I think it can, especially when say a lawyer is, uh, practicing, uh, or 392 00:27:54,105 --> 00:27:57,435 working in a field of law that they don't expertise. 393 00:28:02,940 --> 00:28:11,910 Some knowledge from the LM can be useful or I think it can also be really helpful 394 00:28:11,910 --> 00:28:20,370 for lower stakes agreements as well, where the, it's more acceptable to 395 00:28:21,240 --> 00:28:26,160 incur some risk of LMS making mistake. 396 00:28:26,310 --> 00:28:28,710 Um, but I think in any case. 397 00:28:29,580 --> 00:28:37,350 There will, so long as LLMs aren't capable of fully replacing all people, 398 00:28:38,040 --> 00:28:46,410 then there will be a need for a human lawyer to review its work to make 399 00:28:46,410 --> 00:28:48,210 sure it's not introducing mistakes. 400 00:28:48,210 --> 00:28:56,280 Uh, and so for all of those reasons, I am skeptical of that LLMs. 401 00:28:57,705 --> 00:29:01,755 Pose a, a risk to lawyers' livelihood. 402 00:29:02,355 --> 00:29:03,105 If anything. 403 00:29:03,165 --> 00:29:04,245 I think that 404 00:29:06,855 --> 00:29:12,795 the AI era that we are entering into in the broader economy is likely going to 405 00:29:12,855 --> 00:29:18,945 increase the demand for legal services more broadly across a variety of domains. 406 00:29:20,325 --> 00:29:24,405 So, um, yeah, and I don't, I don't necessarily disagree with you. 407 00:29:24,405 --> 00:29:25,425 Um, I think that. 408 00:29:26,595 --> 00:29:29,685 There are some inherent limitations in the LLM 409 00:29:31,755 --> 00:29:37,485 architecture, um, and its ability to generate novel ideas. 410 00:29:37,545 --> 00:29:40,905 Um, I think there, there's been some creative engineering until, you 411 00:29:40,905 --> 00:29:46,545 know, when we, when inference time compute, um, kind of saved the day 412 00:29:46,755 --> 00:29:50,415 a little bit with LLMs because we'd really started to hit our head on 413 00:29:50,415 --> 00:29:53,985 the scaling laws and, you know, um. 414 00:29:54,600 --> 00:29:58,950 Reasoning or, you know, inference, time compute really extended 415 00:29:59,850 --> 00:30:03,179 the, um, progress trajectory. 416 00:30:03,780 --> 00:30:04,740 But, um. 417 00:30:05,610 --> 00:30:10,320 Switching gears a little bit, I'm curious about, so valuations, I gave a, I, uh, 418 00:30:10,830 --> 00:30:15,810 I had the privilege of moderating the keynote panel at the Legal Tech Connect 419 00:30:15,810 --> 00:30:18,000 conference in New York a few months ago. 420 00:30:18,659 --> 00:30:19,950 And as part of my. 421 00:30:20,754 --> 00:30:21,774 Opening remarks. 422 00:30:21,774 --> 00:30:27,055 I, I used an analogy to, this was right after the Harvey and Lara 423 00:30:27,055 --> 00:30:30,055 a DX, uh, rounds were announced. 424 00:30:30,415 --> 00:30:33,655 So for those that don't know, there were investments in both 425 00:30:33,655 --> 00:30:35,090 Harvey and Lara and they were. 426 00:30:35,940 --> 00:30:38,250 At an 80 times revenue valuation. 427 00:30:38,700 --> 00:30:42,240 So to put this in context, I, I gave a little, a little 428 00:30:42,240 --> 00:30:43,530 metaphor that I'll share here. 429 00:30:43,530 --> 00:30:46,680 I haven't shared it on the podcast before, but it's, it's really interesting. 430 00:30:46,980 --> 00:30:51,480 So my wife and I own five gyms in St. Louis, and, um, they 431 00:30:51,480 --> 00:30:53,640 operate on about a 20% net margin. 432 00:30:54,390 --> 00:31:00,120 If we were to sell those businesses, it would be three to four times ebitda, 433 00:31:00,300 --> 00:31:02,100 which is basically your net margin. 434 00:31:03,150 --> 00:31:03,660 So. 435 00:31:04,334 --> 00:31:09,915 Let's assume for just a minute that, um, that was the 436 00:31:10,574 --> 00:31:13,814 ultimate outcome for investors. 437 00:31:14,145 --> 00:31:18,615 Like if these businesses, whatever they stalled, they, they grew, they 438 00:31:18,615 --> 00:31:24,885 matured and hit a ceiling, and you ended up selling for a multiple of ebitda. 439 00:31:25,635 --> 00:31:32,145 So at a 20% multiple of ebitda, it would take. 440 00:31:32,985 --> 00:31:37,245 400 years to break even on your investment. 441 00:31:38,715 --> 00:31:40,695 400 years. 442 00:31:41,264 --> 00:31:47,985 If George Washington made that, wrote that check to invest in that 443 00:31:47,985 --> 00:31:53,475 round, he'd still have another 150 years until he broke even. 444 00:31:53,745 --> 00:31:55,965 Now, again, lots of caveats there. 445 00:31:55,965 --> 00:32:00,824 First of all, that's not the bet that investors are placing on a on a 20%. 446 00:32:01,785 --> 00:32:07,605 EBITDA multiple, like, but it's a good lens as just a sanity check. 447 00:32:07,605 --> 00:32:12,375 Like, okay, this is how, and it's not just gyms, that's, that's how restaurants, 448 00:32:12,375 --> 00:32:16,185 that's how pretty much every business sells is for a multiple of ebitda, except 449 00:32:16,185 --> 00:32:20,745 in the software world, the technology world where things get, you know. 450 00:32:21,465 --> 00:32:24,795 Typically it's, it's a, it's a multiple of recurring revenue. 451 00:32:25,395 --> 00:32:30,705 But, um, man, and I made a joke and said, you know, 400 years ago, you know, what 452 00:32:30,705 --> 00:32:35,385 was going on 400 years ago, the Nina, the Pinta, and the Santa Maria, right? 453 00:32:35,385 --> 00:32:42,195 Just to put some, put a visual on what time span we're sitting in New York City, 454 00:32:42,255 --> 00:32:45,735 which was, um, didn't even really exist. 455 00:32:46,245 --> 00:32:46,695 Uh. 456 00:32:47,100 --> 00:32:49,650 So what are your thoughts on these valuations? 457 00:32:49,650 --> 00:32:51,510 Like, um, do they make sense? 458 00:32:51,510 --> 00:32:55,230 Can you make sense of them or is it r hard for you to get your head around as well? 459 00:32:55,800 --> 00:32:57,570 It's a little hard for me to get my head around. 460 00:32:58,020 --> 00:33:05,490 Um, I mean, I try to think of like, what is a comparable company in legal 461 00:33:05,490 --> 00:33:12,870 tech that has a comparable value that I could imagine them growing into? 462 00:33:14,324 --> 00:33:17,205 And I'm just not really sure what that is. 463 00:33:17,205 --> 00:33:22,215 I mean, the, these numbers are private, but I believe according to estimates 464 00:33:22,215 --> 00:33:26,715 that Harvey's latest valuation is greater than that of iManage and net docs and 465 00:33:26,715 --> 00:33:30,254 let combined, um, don't quote me on that. 466 00:33:30,495 --> 00:33:32,419 I, that's based off of estimates. 467 00:33:32,564 --> 00:33:39,585 Um, but that seems at least, I think that's like ballpark about right. 468 00:33:40,725 --> 00:33:41,655 And so. 469 00:33:44,100 --> 00:33:50,220 Yeah, it, it, it's a bit, it is challenging for me to understand 470 00:33:52,290 --> 00:33:58,740 how they grow into evaluation that is, you know, as great or larger 471 00:33:58,740 --> 00:34:02,970 than the most critical pieces of lawyers' existing workflows. 472 00:34:04,410 --> 00:34:05,460 It's a great point. 473 00:34:05,460 --> 00:34:08,130 So, yeah, it's been a little bit, um. 474 00:34:08,520 --> 00:34:11,250 Since I've looked at this, but I just kind of did a quick search. 475 00:34:11,909 --> 00:34:18,510 So, um, Thomson Reuters has a market cap of around 60 billion. 476 00:34:19,350 --> 00:34:23,880 And what investors, so there's a little bit of debate. 477 00:34:23,880 --> 00:34:30,450 One of the in, in early VC rounds, investors are thinking to themselves, 478 00:34:30,450 --> 00:34:35,250 what has to be true in order for me to get 10 x on this investment? 479 00:34:36,465 --> 00:34:41,895 Now you could, you could argue that, well, this wasn't a series A or a series B. 480 00:34:42,195 --> 00:34:47,355 This was, I don't know what, how, how many rounds in they were, but you could 481 00:34:47,355 --> 00:34:52,215 argue that, okay, that number, that multiple goes down in later rounds. 482 00:34:52,545 --> 00:34:59,625 But I would counter back with the reason that the expectations are 10 x in the 483 00:34:59,625 --> 00:35:01,815 earlier rounds is because of risk. 484 00:35:02,445 --> 00:35:05,985 Right They to, to get the right risk, reward balance you need to 485 00:35:05,985 --> 00:35:10,665 10 x you know, one or two of your, to your use, your example of your 486 00:35:10,665 --> 00:35:12,765 20 port codes, half the 10 x. 487 00:35:13,395 --> 00:35:18,885 I would argue that based on that valuation, you're at a similar. 488 00:35:19,740 --> 00:35:25,260 Risk as somebody who's investing in a series A in a startup, right? 489 00:35:25,260 --> 00:35:28,560 That's kind of the risk profile, and you need a commensurate reward. 490 00:35:28,800 --> 00:35:35,760 So if I'm writing that check, I'm thinking 10 x. So 10 x would 491 00:35:35,760 --> 00:35:38,910 require an $80 billion exit. 492 00:35:40,169 --> 00:35:44,910 For Harvey, if they were to IPO, they're actually too big to be bought right now. 493 00:35:45,120 --> 00:35:47,459 There's not a legal tech company maybe other than Thomson 494 00:35:47,459 --> 00:35:49,109 Reuters, which is a hard no. 495 00:35:49,680 --> 00:35:51,930 Uh, given that, you know, their investment in co-counsel, they're 496 00:35:51,930 --> 00:35:53,490 not, they're not gonna buy Harvey. 497 00:35:54,060 --> 00:35:56,850 So how do you get to an $80 billion exit? 498 00:35:56,910 --> 00:36:04,649 The largest, the the largest IPO in legal tech history is, um, LegalZoom. 499 00:36:05,505 --> 00:36:12,585 And it was, it iPod during the, uh, 21, you know, craziness. 500 00:36:12,615 --> 00:36:19,995 2021 when the fed dumped $7 trillion of capital into the economy and that all that 501 00:36:19,995 --> 00:36:21,495 capital had to find a productive home. 502 00:36:22,065 --> 00:36:28,185 And it iPod for, I think the number, it was in the high single digits, seven 503 00:36:28,785 --> 00:36:33,765 8 billion Today, it's worth about 1.5. 504 00:36:34,620 --> 00:36:34,980 Right. 505 00:36:34,980 --> 00:36:40,890 So think about what has to happen in order for an IPO of that size. 506 00:36:41,220 --> 00:36:45,810 Again, you would eclipse Thomson Reuters, which goes way beyond just legal tech. 507 00:36:46,500 --> 00:36:49,230 I mean, Thomson Reuters is a massive company. 508 00:36:49,230 --> 00:36:53,069 So yeah, it's hard for me to, to, to do the metal math to get there. 509 00:36:53,100 --> 00:36:53,970 It's, it really is. 510 00:36:54,509 --> 00:36:54,750 Yeah. 511 00:36:55,140 --> 00:36:55,319 Yeah. 512 00:36:55,319 --> 00:36:55,740 I agree. 513 00:36:56,819 --> 00:37:00,240 So, um, let's talk a little bit about, um. 514 00:37:01,020 --> 00:37:07,529 The open ais and the Anthropics, and for that matter, GRS and Geminis, 515 00:37:08,009 --> 00:37:10,620 their ability to compete with Harvey. 516 00:37:10,620 --> 00:37:12,900 Even though OpenAI was an early investor. 517 00:37:13,470 --> 00:37:17,759 Um, I've heard, I'm pretty sure I've heard or read like some of 518 00:37:17,759 --> 00:37:21,840 the founders of these companies say that that's their biggest threat. 519 00:37:22,380 --> 00:37:25,380 How do you see like the frontier models. 520 00:37:26,370 --> 00:37:30,360 Competing and their ability to compete with these legal specific solutions? 521 00:37:33,780 --> 00:37:44,130 Well, I think that the, the intelligence component of legal AI product, really 522 00:37:44,160 --> 00:37:45,900 any AI product for that matter, 523 00:37:47,940 --> 00:37:51,150 it, it comes straight from the model providers at this point. 524 00:37:51,690 --> 00:37:52,770 So, um. 525 00:37:53,505 --> 00:37:58,605 For as a quick definition, foundation model is the term that's used to refer 526 00:37:58,605 --> 00:38:03,525 to the sort of out of the box models that OpenAI or Entropic or Google Produce, 527 00:38:04,965 --> 00:38:12,555 and there no industry specific provider is going to be the foundation models. 528 00:38:14,025 --> 00:38:17,565 Early on when CHATT was just released. 529 00:38:19,920 --> 00:38:23,549 GPT-3 five really wasn't that useful yet. 530 00:38:23,970 --> 00:38:25,529 The context window wasn't big enough. 531 00:38:25,620 --> 00:38:26,549 It wasn't that smart. 532 00:38:27,750 --> 00:38:33,150 There were gains to be made by fine tuning it, so sort of retraining the 533 00:38:33,150 --> 00:38:36,150 model on context specific information. 534 00:38:37,230 --> 00:38:42,720 But since then, the subsequent model iterations have gotten, 535 00:38:43,410 --> 00:38:48,990 have gotten so big with so much training data that any marginal. 536 00:38:50,085 --> 00:38:58,095 Data that you could fine tune it with from, you know, customer's legal documents 537 00:38:58,875 --> 00:39:04,845 is the tiniest drop in the bucket, and it really is not going to make a 538 00:39:04,845 --> 00:39:06,165 meaningful difference at this point. 539 00:39:07,245 --> 00:39:10,275 Now, the thing that does make a difference relative to just CHATT 540 00:39:10,395 --> 00:39:14,715 PT Outta box is referred to as in context learning, which you can think 541 00:39:14,715 --> 00:39:16,995 of basically as just how well you. 542 00:39:19,050 --> 00:39:21,510 Having it a well-designed, 543 00:39:23,850 --> 00:39:25,260 yeah, well-crafted prompt. 544 00:39:25,260 --> 00:39:28,920 Pulling in the right context does meaningfully affect the quality of the 545 00:39:28,920 --> 00:39:34,650 response, and so I think a strategy that I've noticed with a lot of AI 546 00:39:34,650 --> 00:39:41,640 startups at the moment is positioning themself as a domain specific product. 547 00:39:44,819 --> 00:39:52,230 Repackaging well written prompts on top of foundation models, and I think a lot of 548 00:39:52,649 --> 00:40:00,000 users of Chat GPT, um, don't realize ways in which you can modify it with your own 549 00:40:00,000 --> 00:40:02,310 custom prompts to achieve similar results. 550 00:40:03,629 --> 00:40:08,609 All of which is to say that I don't think that any. 551 00:40:10,305 --> 00:40:16,815 AI startup, legal or otherwise is, can build any kind of meaningful note 552 00:40:17,265 --> 00:40:22,785 on the quality of the intelligence to the extent that there's going 553 00:40:22,785 --> 00:40:24,435 to be a moat, a product note. 554 00:40:24,555 --> 00:40:28,755 It needs to come from the, the product that you build on top of 555 00:40:28,755 --> 00:40:34,395 it, the workflows that you build on top of it, uh, how meaningfully 556 00:40:34,395 --> 00:40:38,325 different your solution is to what? 557 00:40:39,270 --> 00:40:42,660 Somebody would be capable of achieving by using a chat bott or 558 00:40:42,660 --> 00:40:44,520 by using cloud code, et cetera. 559 00:40:45,840 --> 00:40:55,290 Yeah, and speaking of which, it's um, it is astounding at the rate at which, like, 560 00:40:55,290 --> 00:40:56,850 you know, cloud code's a good example. 561 00:40:57,240 --> 00:41:00,400 Again, I'm not sure when this will will come out, but, um, you know. 562 00:41:01,274 --> 00:41:05,504 Claude Code has historically been a little bit challenging to use. 563 00:41:05,504 --> 00:41:09,375 Like you have to, you have to use a terminal, like, you 564 00:41:09,375 --> 00:41:11,294 know, visual Studio Code. 565 00:41:11,745 --> 00:41:15,734 You have to wire up GitHub, you gotta run some bash scripts. 566 00:41:15,765 --> 00:41:16,455 It's um. 567 00:41:17,190 --> 00:41:20,850 It's been a little bit painful, candidly, and nobody really loves 568 00:41:20,850 --> 00:41:22,529 the whole terminal interface. 569 00:41:23,220 --> 00:41:28,860 Um, you know, they are rapidly innovating and Claude Cowork just got released 570 00:41:28,860 --> 00:41:30,240 today, as we talked about earlier. 571 00:41:30,870 --> 00:41:33,840 And, um, they're making it easier and easier. 572 00:41:34,290 --> 00:41:38,190 And what I have noticed, I don't know if this is true with the Harvey's and 573 00:41:38,190 --> 00:41:44,549 Leggos of the world, but like it, um, companies who, Microsoft's a good example, 574 00:41:44,549 --> 00:41:46,590 who have taken and built on top of. 575 00:41:47,115 --> 00:41:52,335 The foundation models, it's usually watered down, at least to some extent. 576 00:41:52,634 --> 00:41:57,585 And the innovations get there later, right? 577 00:41:57,674 --> 00:42:00,134 So co-pilot's a great example. 578 00:42:00,765 --> 00:42:04,605 Uh, my listeners and people who follow me on LinkedIn know 579 00:42:04,605 --> 00:42:06,615 I am not a fan of co-pilot. 580 00:42:06,705 --> 00:42:09,315 It's, um, it's, it's just not good. 581 00:42:09,944 --> 00:42:12,915 And frankly, Microsoft should be embarrassed. 582 00:42:13,275 --> 00:42:19,575 Um, uh, Satya has been such a amazing CEOI was at, I was at Microsoft in the 583 00:42:19,575 --> 00:42:24,975 Gates Balmer era, and Satya has just changed Microsoft in so many positive 584 00:42:24,975 --> 00:42:29,805 ways, but I've actually lost some respect for him through, um, through 585 00:42:29,805 --> 00:42:37,005 this copilot fiasco and how, how far away the marketing is from reality. 586 00:42:37,545 --> 00:42:38,505 Like any, it definitely works. 587 00:42:39,180 --> 00:42:40,830 It just doesn't work. 588 00:42:40,890 --> 00:42:44,670 Like, it doesn't work, like it, it like try using it in Excel 589 00:42:44,670 --> 00:42:46,440 and do anything meaningful. 590 00:42:46,710 --> 00:42:48,600 It just doesn't work. 591 00:42:49,080 --> 00:42:51,060 And, um, yeah, I don't know. 592 00:42:51,060 --> 00:42:54,360 Is it your take too that not specifically with copilot? 593 00:42:54,990 --> 00:42:59,580 Uh, I really want, by the way, I'm like, I'm, I want copilot to be better. 594 00:42:59,610 --> 00:43:00,630 We're a Microsoft shop. 595 00:43:01,230 --> 00:43:01,320 Sure. 596 00:43:01,320 --> 00:43:01,380 Yeah. 597 00:43:01,380 --> 00:43:02,820 I want it, I want it to be good. 598 00:43:02,940 --> 00:43:03,690 It's just not. 599 00:43:04,200 --> 00:43:05,070 Um, but. 600 00:43:05,895 --> 00:43:10,275 Like, is it your perspective as well that like the foundation, anybody, 601 00:43:10,275 --> 00:43:12,105 anything built on the foundation models? 602 00:43:13,395 --> 00:43:14,985 It's not a one-to-one. 603 00:43:14,985 --> 00:43:19,155 Like there's, you get features sometimes later and there's some translate. 604 00:43:19,155 --> 00:43:22,755 I don't know if something happens in translation, but the, the, the 605 00:43:22,755 --> 00:43:24,225 frontier model seem to be better. 606 00:43:26,390 --> 00:43:27,285 Oh, interesting. 607 00:43:27,285 --> 00:43:30,495 So you, you, you're asking if the frontier models. 608 00:43:31,365 --> 00:43:33,405 Better than the models that you get when you're actually 609 00:43:33,405 --> 00:43:34,875 using a third party application. 610 00:43:34,965 --> 00:43:35,625 Exactly. 611 00:43:36,675 --> 00:43:43,965 Um, you know, I guess I, I, I've heard people speculate something similar 612 00:43:43,965 --> 00:43:49,665 like, is Chat, is OpenAI holding out the best version of 5.2 for chat? 613 00:43:49,665 --> 00:43:51,165 TPT and I? 614 00:43:51,165 --> 00:43:52,395 I can't say for certain. 615 00:43:53,685 --> 00:43:56,445 Uh, I would say that I think, I think. 616 00:43:57,870 --> 00:44:03,360 The APIs that the foundation models that the model providers expose to, 617 00:44:04,350 --> 00:44:07,320 to everyone else are pretty good. 618 00:44:08,190 --> 00:44:11,700 Uh, I haven't been disappointed with them. 619 00:44:11,760 --> 00:44:17,850 I think a lot of, I mean, I think a lot of the execution of a product comes down to 620 00:44:20,130 --> 00:44:23,160 integrating it well into the workflow, providing it with the 621 00:44:23,160 --> 00:44:26,190 right context, being able to. 622 00:44:26,625 --> 00:44:31,335 Um, make modifications meaningfully and correctly to your 623 00:44:31,335 --> 00:44:33,225 documents, that sort of thing. 624 00:44:33,285 --> 00:44:38,984 And then as I mentioned a moment ago, I also think prompting, uh, there's 625 00:44:38,984 --> 00:44:45,585 quite a bit of alpha and prompting and as well, um, yeah, so in terms of 626 00:44:45,585 --> 00:44:49,815 like the pure model quality, perhaps they're, you know, they're holding 627 00:44:49,815 --> 00:44:51,404 out the better one for themselves. 628 00:44:52,455 --> 00:44:57,254 That to me doesn't seem like a, a major impediment for startups. 629 00:44:57,765 --> 00:45:01,575 Yeah, and you know, I think with Microsoft, I ha I speculate, I 630 00:45:01,575 --> 00:45:04,695 think what Microsoft is doing is they're doing some token throttling. 631 00:45:05,174 --> 00:45:10,095 'cause you can take the exact same prompt in copilot and run it. 632 00:45:10,095 --> 00:45:11,475 And I did side by side. 633 00:45:11,475 --> 00:45:12,884 I actually PO posted on LinkedIn. 634 00:45:12,884 --> 00:45:13,695 Examples of it. 635 00:45:14,055 --> 00:45:19,125 Just a really quick one is I asked for the last I all time 636 00:45:19,515 --> 00:45:22,424 NCAA men's basketball winners. 637 00:45:23,010 --> 00:45:24,480 And the opponents. 638 00:45:24,990 --> 00:45:29,190 And when I ran it in chat pt, it gave me things I didn't even ask for, like the 639 00:45:29,190 --> 00:45:34,170 venue, the score, and it went back all the way to the beginning, which is like in 640 00:45:34,170 --> 00:45:40,710 the 1930s when I ran that same prompt in copilot, it only gave me the last 20 years 641 00:45:41,010 --> 00:45:44,880 and it just gave me, um, like the winner. 642 00:45:45,120 --> 00:45:45,600 I see. 643 00:45:45,600 --> 00:45:46,350 Um, yeah. 644 00:45:46,620 --> 00:45:51,360 So if I had to speculate, I would guess that. 645 00:45:53,070 --> 00:45:57,990 Copilot is probably not using the, whatever the current 646 00:45:57,990 --> 00:46:00,480 top tier OpenAI model is. 647 00:46:00,480 --> 00:46:03,060 Five two reasoning. 648 00:46:03,630 --> 00:46:06,270 Uh, they're probably using a faster, cheaper model. 649 00:46:06,450 --> 00:46:06,540 Mm-hmm. 650 00:46:06,780 --> 00:46:13,050 And I think the, the incentives are a bit different for OpenAI versus Microsoft. 651 00:46:13,380 --> 00:46:17,550 So like OpenAI is competing with Anthropic and Gemini right now. 652 00:46:17,880 --> 00:46:21,300 They wanna make sure that consumers who go on their app are getting like the best. 653 00:46:21,885 --> 00:46:28,725 Most intelligent response, whereas when you're in the context of using a Microsoft 654 00:46:28,725 --> 00:46:37,005 product copilot, there's not really like an alternative to copilot that you can 655 00:46:37,395 --> 00:46:42,135 super easily switch to in the way that you can open up a cloud window if you're 656 00:46:42,135 --> 00:46:44,085 not satisfied with the chat response. 657 00:46:45,060 --> 00:46:45,330 Yeah. 658 00:46:45,425 --> 00:46:47,370 That, that would be my guess is what's going on there. 659 00:46:47,640 --> 00:46:48,750 Yeah, that's a good point. 660 00:46:49,350 --> 00:46:53,040 Um, okay, we're almost outta time, but I did wanna ask you one, one more question. 661 00:46:53,040 --> 00:46:53,130 Sure. 662 00:46:53,310 --> 00:46:59,730 And that was about vibe coding and um, I have seen it is on fire right now in 663 00:46:59,730 --> 00:47:02,040 legal and I think it's so cool to see. 664 00:47:02,340 --> 00:47:02,550 Yeah. 665 00:47:02,610 --> 00:47:07,650 Um, I have a strong bias towards it's great for prototyping and 666 00:47:07,740 --> 00:47:12,300 shortening the software development lifecycle by compressing requirements, 667 00:47:12,300 --> 00:47:13,260 gathering and refinement. 668 00:47:14,295 --> 00:47:18,075 Um, but in that capacity, yeah. 669 00:47:18,735 --> 00:47:23,325 But in no way are these vibe coded apps ready to go be deployed. 670 00:47:23,325 --> 00:47:28,155 You, you literally can't deploy a vibe coded app in your enterprise and maintain 671 00:47:28,155 --> 00:47:31,395 soc two, type two, and, um, not right now. 672 00:47:31,400 --> 00:47:31,420 No. 673 00:47:31,725 --> 00:47:32,355 No, not right. 674 00:47:32,355 --> 00:47:32,595 Yeah. 675 00:47:32,865 --> 00:47:34,935 There's, there's requirements associated with that. 676 00:47:34,935 --> 00:47:39,165 But like, I'm curious just, uh, what's your, what's your take is on, like, 677 00:47:39,195 --> 00:47:43,125 I, there, there's people vibe, coding, like tabular review, and a lot of these. 678 00:47:43,725 --> 00:47:47,444 Features that, um, some of these vertical solutions hang their hat on. 679 00:47:47,444 --> 00:47:49,785 What's your, what's your take on the whole vibe coding movement? 680 00:47:51,435 --> 00:47:52,604 I think it's the future. 681 00:47:52,694 --> 00:47:58,785 I mean, the first time I used cloud code, like March of last 682 00:47:58,785 --> 00:48:01,484 year, it completely blew me away. 683 00:48:01,634 --> 00:48:07,095 I, before that I was pretty skeptical of claims that, you know, a AI was 684 00:48:07,095 --> 00:48:09,854 going to write all my code for me. 685 00:48:09,854 --> 00:48:13,035 'cause it wa like, it just, up until that point it just did not do that. 686 00:48:13,605 --> 00:48:15,645 But cloud code really changed the game. 687 00:48:16,215 --> 00:48:19,005 And I mean, even since then it's gotten so much better. 688 00:48:19,665 --> 00:48:22,725 Um, and so, I mean, I do think it's the future. 689 00:48:22,785 --> 00:48:25,905 How do I expect that future to play out? 690 00:48:26,595 --> 00:48:32,505 So for one thing, it wasn't until maybe six weeks ago when Gemini three was 691 00:48:32,620 --> 00:48:39,135 released and I used Google AI Studio for the first time that, so AI Studio is like. 692 00:48:39,930 --> 00:48:41,520 It takes vibe coding a step further. 693 00:48:41,520 --> 00:48:43,590 It does the full application end-to-end. 694 00:48:43,590 --> 00:48:46,770 You don't even look at a line of code and you can deploy it with a click of button. 695 00:48:46,860 --> 00:48:49,050 It's what like companies like Rep do. 696 00:48:50,250 --> 00:48:57,240 And I was able to build myself fairly non-trivial personal applications 697 00:48:57,240 --> 00:49:01,080 for my development workflow that I know use on a day-to-day basis. 698 00:49:01,590 --> 00:49:03,450 And like, I don't need to hook them up to the internet. 699 00:49:03,450 --> 00:49:07,800 Like I'm not, like, I'm not doing anything too sensitive, so I'm not concerned. 700 00:49:08,610 --> 00:49:12,030 Ultimately a ton about their reliability and security. 701 00:49:12,660 --> 00:49:14,610 Uh, but they are genuinely useful. 702 00:49:15,270 --> 00:49:20,250 And so, I mean, I think it starts with, yeah, like you can build your own tabular 703 00:49:20,250 --> 00:49:25,235 review 'cause it's, it's not a very complicated product and use it yourself. 704 00:49:26,640 --> 00:49:32,010 If we're going to be talking about hosting though, then it 705 00:49:32,010 --> 00:49:33,360 certainly gets more complicated. 706 00:49:33,360 --> 00:49:35,250 No, you can't just throw a coated. 707 00:49:35,685 --> 00:49:42,134 Project out into the world or, or on your firm's, um, you know, private cloud. 708 00:49:43,035 --> 00:49:47,384 The, the way I think about it is, I can imagine there being a few different 709 00:49:47,384 --> 00:49:49,935 phases towards it rolling out. 710 00:49:49,995 --> 00:49:55,455 So for one, a lot of firms have practice innovation teams that have already 711 00:49:55,455 --> 00:49:59,595 been in the practice of developing tools and uh, you know, when they 712 00:49:59,595 --> 00:50:02,895 decide to build or buy, they would be the people who are doing the building. 713 00:50:03,689 --> 00:50:08,310 So I think the bill of buy come calculus can change a lot. 714 00:50:08,310 --> 00:50:11,160 I think those product, those teams can become way more productive. 715 00:50:11,609 --> 00:50:15,540 And of course the things that they produce are still going to be subject to the same 716 00:50:15,540 --> 00:50:19,799 code review processes, the same security review processes, it, review, et cetera. 717 00:50:20,790 --> 00:50:23,399 Um, so that's like short term effect. 718 00:50:23,640 --> 00:50:24,029 Yeah. 719 00:50:24,029 --> 00:50:27,330 I don't see any reason why a law firm couldn't build their own tabular review. 720 00:50:27,419 --> 00:50:28,589 It's not that hard. 721 00:50:29,279 --> 00:50:32,399 Um, you know, as assuming they have that type of process in place. 722 00:50:33,270 --> 00:50:40,560 Long term, I would imagine that the private cloud infrastructure of law firms, 723 00:50:40,589 --> 00:50:46,950 but then enterprises more broadly, will probably adapt to the possibilities of 724 00:50:46,950 --> 00:50:52,680 five coding, like one can imagine in individual lawyer or employee building 725 00:50:52,680 --> 00:50:59,430 their own application and uh, having some kind of streamlined review process that. 726 00:51:00,000 --> 00:51:03,810 It checks up on with automated security checks and then it gets 727 00:51:03,810 --> 00:51:08,670 deployed into like, uh, an isolated container within their private cloud. 728 00:51:09,330 --> 00:51:19,440 Um, I would be shocked if 10 years from now there aren't avenues for 729 00:51:19,500 --> 00:51:25,440 individual employees to be making meaningful contribute contributions to. 730 00:51:27,405 --> 00:51:28,905 Building software with ai. 731 00:51:29,445 --> 00:51:31,035 Yeah, yeah, I agree with that. 732 00:51:31,335 --> 00:51:34,275 Just I'm gonna plug info dash for a second. 733 00:51:34,275 --> 00:51:37,215 We have a integration layer, it's called Integration hub. 734 00:51:38,100 --> 00:51:43,680 We, we, we build it because it was, um, when we were, it hydrates all 735 00:51:43,680 --> 00:51:47,760 of our web parts and SharePoint for internet and extranet scenarios. 736 00:51:48,240 --> 00:51:51,030 But it's an Azure appliance and it has tentacles into 737 00:51:51,030 --> 00:51:52,500 all the back office systems. 738 00:51:53,040 --> 00:51:56,340 And it presents a unified API that security trimmed. 739 00:51:56,820 --> 00:52:01,170 And if you have your walls integrated, it will respect ethical wall boundaries. 740 00:52:01,440 --> 00:52:04,650 So we think it's like the perfect foundation because that's the 741 00:52:04,650 --> 00:52:06,360 hardest part is getting to the data. 742 00:52:07,470 --> 00:52:08,880 You know, you've got different APIs. 743 00:52:08,880 --> 00:52:10,080 Those APIs change. 744 00:52:10,080 --> 00:52:11,580 Some of them aren't well documented. 745 00:52:12,060 --> 00:52:17,370 So if you can tap into our API, if there's any firms out there, we're looking 746 00:52:17,370 --> 00:52:19,860 for firms who want to explore this. 747 00:52:19,860 --> 00:52:23,520 The reality is most of them aren't ready for that. 748 00:52:23,550 --> 00:52:28,920 'cause I mean, we have currently, we have like 25% of the AM law as clients already. 749 00:52:28,950 --> 00:52:34,020 So this is already just one out 4:00 AM law firms already have this ready to go. 750 00:52:35,130 --> 00:52:40,530 Most firms just aren't willing to let their people do this yet, but, 751 00:52:40,620 --> 00:52:44,520 um, we think they will, hopefully in the future, give the time. 752 00:52:44,910 --> 00:52:45,270 Yeah. 753 00:52:45,780 --> 00:52:48,060 So, uh, well this has been a great conversation, man. 754 00:52:48,090 --> 00:52:52,740 Um, before we go, how do people find out more about you and your product? 755 00:52:52,860 --> 00:52:54,480 Um, or, or get in touch? 756 00:52:55,380 --> 00:52:56,190 Yeah, sure. 757 00:52:56,190 --> 00:52:59,759 So, uh, again, our product is version story. 758 00:53:00,270 --> 00:53:04,020 So you can check us out@versionstory.com or on LinkedIn. 759 00:53:04,620 --> 00:53:11,520 Uh, additionally, I also write a blog on Substack called The Redline. 760 00:53:12,120 --> 00:53:15,960 Um, I believe the domain is the redline.com or the 761 00:53:15,960 --> 00:53:17,730 redline dot version story.com. 762 00:53:18,270 --> 00:53:21,090 If you wanna draw me a line, send me an email free. 763 00:53:21,210 --> 00:53:22,110 Feel free to reach out. 764 00:53:22,110 --> 00:53:24,900 My email isJordan@versionstory.com. 765 00:53:25,695 --> 00:53:26,205 Awesome. 766 00:53:26,625 --> 00:53:27,915 Well, best of luck to you, man. 767 00:53:27,975 --> 00:53:30,915 Um, I'm, you know, I'm pulling for everybody in this space. 768 00:53:30,975 --> 00:53:34,875 Uh, I, I don't think everybody's gonna survive because there's been, 769 00:53:34,875 --> 00:53:37,485 this money has, has flowed so freely. 770 00:53:37,905 --> 00:53:42,525 I think there are gonna be some, um, uh, there's gonna be some 771 00:53:42,525 --> 00:53:44,175 people who don't make the cut. 772 00:53:44,175 --> 00:53:48,255 But, um, yeah, I'm really excited to see where things go and thanks 773 00:53:48,255 --> 00:53:50,925 so much for, for coming and spending some time with me today. 774 00:53:50,925 --> 00:53:51,825 I really appreciate it. 775 00:53:52,350 --> 00:53:52,830 Of course. 776 00:53:52,890 --> 00:53:53,880 Uh, thank you, Ted. 777 00:53:53,880 --> 00:53:55,290 This has been a lot of fun. 778 00:53:55,710 --> 00:53:56,220 Awesome. 779 00:53:56,280 --> 00:53:57,960 All right, we'll, uh, we'll chat soon. 780 00:53:58,440 --> 00:53:58,950 Sounds good. 781 00:53:59,130 --> 00:53:59,460 All right. 782 00:53:59,460 --> 00:53:59,910 Take care. 783 00:54:00,180 --> 00:54:02,430 Thanks for listening to Legal Innovation Spotlight. 784 00:54:02,970 --> 00:54:06,450 If you found value in this chat, hit the subscribe button to be notified 785 00:54:06,450 --> 00:54:07,950 when we release new episodes. 786 00:54:08,460 --> 00:54:11,160 We'd also really appreciate it if you could take a moment to rate 787 00:54:11,160 --> 00:54:13,770 us and leave us a review wherever you're listening right now. 788 00:54:14,370 --> 00:54:17,070 Your feedback helps us provide you with top-notch content. -->