Too Much Money Is Not Enough

Horace and Damien record an unedited and unfiltered episode, opening with Horace’s stressful overnight apartment move in New York while planning a two-month summer trip to Australia. 

We discuss legal tech execution versus announcements, discussing a link about Kirkland & Ellis’s reported $500 million AI commitment and whether it reflects real technology spend or largely lawyer hours, while noting firms and Fortune 100 legal departments are building internal AI teams. They explore AI raising the floor and ceiling of legal work, GPU scarcity, consumption/token pricing versus seat pricing, and proposals like taxing tokens. They discuss the Pope’s AI-focused writing as a warning about tools that can help or diminish humanity, then cover market shifts including Claude for Legal, Microsoft’s legal Word add-in, open source momentum, and agentic “law firm” projects like Lavern AI, emphasizing that data, KM, and practical execution matter more than hype.

00:00 Welcome and Moving Trauma
03:42 Kirkland AI Spend Debate
09:12 AI Floor vs Ceiling
12:00 Universal Basic Services
13:44 Nvidia Bubble and Pricing
16:54 Taxing Tokens and Models
18:21 Pope on AI and Humanity
21:45 Legal Tech Market Check
22:20 Open Source Vibe Coding
23:31 Quality Security and Stitching
25:21 Linux and Court Tools
26:42 Open Source Custom UX
27:38 Users Don’t Know Yet
28:59 Designers vs Atoms
31:12 Agentic Legal Open Source
32:25 Humans Still Sell
34:41 Jevons and GPU Scarcity
38:11 Symbolic AI Saves Tokens
42:10 Three Paths to Efficiency
46:17 AI Literacy Dunning Kruger
49:06 Precision Recall and KM Power
52:28 Royalties for Know How

Transcript

0:01

Hello, stranger! Uh, man, I, I, is that Horace Wu I see? My, my, my eyes deceive me. Oh, I've had a haircut, so, you know, might not recognize me. Well, it's been so long too. It has been. You know what keeps these recordings from, from taking so long and, or taking such a long time? It's because we have to go and edit stuff afterwards. And so I'm thinking for this one, we edit nothing. We just Oh, that'd let it out as is Oh, uh, that's, uh, that's r- walking without a, a net. Uh, tightrope without a net. Let's do it. I saw this Cirque du Soleil, um, act where it showed like, um, I think it was a Pirates of the Caribbean theme, like the, the trapeze fellows, like swinging and missing and dropping in the water and swinging and missing, all these rounds of practice. And you realize from watching that how, how hard it is for that one perfect swing during that evening show. And you don't even go, "Oh, that, that's just a swing. It's not that hard." When you realize all the effort that goes into it, your appreciation goes way up. Exactly. So as you listen to this unedited episode, know that our lifeless bodies at the end are, are reflecting how poorly we did Oh, the casualties, the casualties. Um, but, but Damien, look, I, I was gonna sh- I was gonna share my trauma, um, before this call, but I figured I'll record it, um, so everyone can hear I always love hearing your trauma, especially if it's being recorded for posterity Oh, well. So, so listeners who, uh, s- get, get to catch up with me will know that, um, I am planning to spend the summer in Australia. So Damien, I hadn't told you this yet. Um, going there for two months, uh, to, to have the kids spend time with their grandparents, uh, July and August. Now, all of that sounds easy until we learned a couple of months ago that our lease for our apartment came up yesterday. Yesterday was the last day, and then there was a month gap, and we're like,"What are we gonna do in that month gap? We can't just pull the kids out of school." So we had to find a temporary apartment inside of New York while still also looking for a lease that started three months from now when we come back, and here's what happened. We found an apartment from the 25th floor o- of our building to go to the 6th floor, so it's like, this is easy. This is fine, right? And we'll put, like, two-thirds of our stuff into storage, the other, you know, one-third in our apartment, and then we'll, we'll be fine. The movers came. They brought a truck that was too small despite having a full inventory. We had to discard our bed and our couch in order to make the move happen.

When we moved in, it was about 4: 2:40

00 AM in the morning because the movers

didn't arrive till 5: 2:45

00 PM that day. Oh no And so this is the trauma, Damien. You're talking to me from, from the other side of having lived through an adrenaline rush for 12 hours Are, are you sleep-deprived and, uh, otherwise, uh, taking copious amounts of caffeine and other, uh, other substances? Look at Yes. Coke. Coca-Cola. yeah, I think that Co- shouldn't say Coke for the listeners who can't see YouTube Exactly. That was indeed a Coca-Cola. The, the, the, I'd like to teach the world to sing and not have them do cocaine, but instead, uh, do the Coca-Cola Yes, 2000-- vintage 2026, not vintage 1926 Exactly. Well, that, that is indeed traumatizing. Uh, that's… I, I wonder if there's a legal tech metaphor in there somewhere. That, uh, you can, you can order, uh, the, the movers that you want, but, uh, of course, when the rubber hits the road, uh, n- n- you know, anything can happen.

So this morning at 7: 3:42

00 AM, uh, Stephanie Gal- Stephanie Gautos, uh, who's now head of AI at, um, Littler instead of… Yeah. Yeah. So it's awesome for her to move. She messages me. Um, I, I hadn't woken up yet, and, uh, she messages me the link to the Kirkland $500 million commitment to do internal AI spend. I sat up in my bed and I was like, "What?" Um, anyway, there was, there was a point I raised this. It was to answer your question, um, which I've now forgotten about, um, 'cause there's been so much excitement in my life. Um, and the story here, and, oh, that's it. The story here and what she and I ended up saying was, um, it's not about the commitment. It's not about the, the, the description and the list of furniture you're gonna move. It's the execution, and I think that's the lesson here for legal tech. I, I think that's right. Uh, you know, when I worked at Thomson Reuters, um, uh, somebody, uh, said,"You know, it's any, any jerk can have an idea." Except they didn't say jerk, they said a word that starts with an A, right? Uh, jerk can have an idea, but really it takes professionals to execute that idea. somebody said, "Boy, you at Thomson Reuters should build X," or, "You at vLex should build Y," or, "You at Clio should build Z," um, you know, it's been, yeah, I had that idea back in 2015 too, right? Hmm But it takes a long time, uh, and, uh, a lot of execution, uh, you know, uh, personnel to be able to actually build the thing that you have this really great idea about. Um, so the, the build has always been the hard part. Mm-hmm. the, um… Right now I, I guess the build is maybe easier than ever, to be able to do. That maybe you don't need dozens of people and a year. Uh, maybe you just need, uh, you know, a vibe coder on Coca-Cola, uh, that can be able to, you know, use Claude Code in the way that, uh, God intended, uh, to be able to get things out the door more quickly. And I, I guess the, the real question is, you know, K&E is $500 billion. How… What are they gonna spend it on? Uh, and, and, you know, is this, is this gonna be on a bunch of vibe coders with Claude Code? Are they gonna spend it on data? Uh, like, what… Where's this gonna go? I thought the article said they, they've committed 250 lawyers on this, including 100 partners. Um, and someone, I f- I forget who, um, on LinkedIn did the math and went,"Hang on, 250, uh, lawyers at Kirkland's blended rate of, say, $2,000 an hour, that's your $100 million spend per year. You're not spending any money on tech. It's all people hours." Um, so, you know, maybe that's it I, and it's funny, like when I worked at a law firm, I would look around the table and see how many lawyers are there, and I would look at the collective billable hours for this one one-hour meeting, and I would say,"Wow, this is a $10,000 meeting," or,"This is a $12,000 meeting," right? Um, so I wonder if they're using that funny math to be able to say, "Yeah, the, what's the value of the thing that we're doing?" And just counting billable hours. I think that's, that's possibly it. Like, it, to me, and I'm sorry Kirkland for saying this, but it felt a little bit l- more like a marketing play than it is actually like a, a, a real kind of directional thing that they've taken steps towards. Um, and, and now as I say that, Kirkland's gonna strike any possible contract with me. That's it. You're gone forever, Horace But I, I would, I would-- I have no doubt because I've talked to people that are working with Kirkland. They're, they're spending real dollars on real solutions with real architects and building real things. So I would say that I-- at least some of that five hundred thousand dollars is real. Um, and I would say that also, um, you know, of course, uh, Kirkland& Ellis and all the Am Law are doing this, the same type of thing. Uh, so are internal counsel. Uh, so I, I mentioned maybe, uh, you know, I mentioned before the call that I spent a lovely morning with, uh, two Fortune 100 companies, uh, who are internal counsel that are doing similar things that Kirkland & Ellis are doing. Uh, and, uh, you know, they're, they're setting forth teams of lawyers building things. Uh, and, uh, you know, I guess the question is, you know, do these Fortune one hundred companies, uh, put the hourly rate of their lawyers and, uh, and say, "Okay, we're spending a hundred million dollars within this law, within this law department, uh, building these cool things," right? Uh, so if, if we're, uh, if we're talking, you know, dollars to dollars, uh, probably good that K&E is previ- putting the effort. Uh, you know, and kudos to them, whether they're spending money, real dollars on external people bringing them in or real, uh, da-days, weeks, months from their internal lawyers, uh, building things. Uh, either way, I think kudos to them for actually seeing the writing on the wall and let's make things better. Mm-hmm. that the clients are doing the same things. So I, I think that we need to ensure that both sides of, uh, the relationship, uh, need to step it up, and I'm glad that K&E is stepping it up, and I'm glad that the Fortune 100 companies that I've talked to with are also, uh, stepping it up. Two comments come to mind as you say that. The first is in the article itself, I forget who at Kirkland said this, but one of the partners, I think partner in charge perhaps, said, "Nobody pays us for the floor." Like, you, you don't pay Kirkland to get the floor of what AI can do. It has to go higher than that. It has to be top, you know, top of the market. So I g- I guess that makes a lot of sense why they'll be investing their own money to build something that goes beyond, like, buying Harvey, Legora, or whatever else is off the shelf. Um, that said, there's a lot of value in combining these technologies in a way that actually makes sense. So the old buy v- buy versus build thing comes along. But second comment, which is something we've raised before, and I forget who said this. I think, I think you would know better than I do, but it might have been Darth at Ford who said, "Hey, hey guys, you got a new competitor in town, and, and it's me. It's the in-house team." Darth Vaughn that's it. So, so, like, I think law firms can see the writing on the wall. Um, and the way I see it is there's, like, these two curves, right? There's, there's, uh, um, uh, time at the, the horizontal axis, and then there's, um, quality on the vertical axis. And as time goes on, the floor of what AI can deliver as a quality goes up. Um, and the ceiling also goes up, but the floor is how much can humans actually adopt? How much can humans take advantage of what the ceiling of AI is? And I think that gap is getting larger. And I think that's the, the sort of thing we have to resolve as a community I, I a hundred percent agree. And, and what you've just said is reminiscent of, you know, I, I listen to a lot of the, uh, the Valley, uh, podcast, whether it's A16Z or Moonshots or others, uh, and they've used that floor and ceiling analogy a lot. Uh, and that, that really AI raises the floor for everyone. So now everyone can be a mediocre lawyer, right? Even if they haven't-- don't have a law degree, right? Uh, and so that floor has now been raised from, uh, you know, a pro se litigant, uh, put scratching something out on a, a yellow legal pad in the old world, uh, to the new world of they can actually do a pretty good, uh, pretty good first draft and something that maybe will get past a motion to dismiss. Um, is it as bad-- is it as good as an Am Law 20 law firm? No way. Uh, but the floor has been raised. Uh, and so I would say that, uh, yeah, I think AI has raised the floor, but it also raises the ceiling for those who actually are, uh, wielding AI like it should be. Um, and you know, Elon has said this, is that, you know, there's, there's the idea of universal basic income, uh, which has been unattainable. Uh, but there's also the idea of universal basic services, uh, whereas more of the goods and services are being tied to AI. That is, you know, if you're doing legal services using GPUs, maybe you get your universal basic services of GPUs for legal things or universal basic services for accounting things. Uh, so the services you get your GPUs and maybe even products you get your GPUs. You know, as robots are building things, you get your allotment as a citizen of your universal basic services and products. Um, so anyway. So I, I-- in that way, he said that we're raising the floor for everyone, but then people like him, he said they're-- we're also raising the ceiling for him. Now he can make a trillion dollars, right? And so the ceiling goes to the moon, right? While the floor also gets raised. Um, whether that future comes to pass or not, who knows? Uh, but at least it's an interesting, uh, thought experiment to say, okay, if we are truly raising the floor, uh, for humanity, um, and we're truly raising the floor for all the legal law firms and all the law departments, um, then how high can those, uh, people that are using AI like it should be? Th-- maybe there is no ceiling. You know what's funny? Like, segue, complete segue. I, I just set up new internet at my house, and I went and looked at what are plans that we're on, and we're like, "Wow." Boiling it down, it's like, you know, $2, $3 per day. And I started counting the units, uh, and, and, and the cost that we have to pay to survive, right? You know, nothing's free. It costs, like, $15 per day for food minimum. It costs $2 per day for internet. It costs extra blah, blah, blah. And as you kind of break it down, it sounds like this universal ability to access GPUs is another one of these units that then contributes to what a human needs to then go and be productive and, and do whatever they choose to do. Um, it's a very interesting way of breaking down what is available as a resource to humans. Um, I don't know, however, if it is as straightforward as, you know, Damien, your 10 units of GPU is gonna be equivalent to my 10 units of GPU or equivalent to like a 10-year-old at, at school's GPU, right? They're, they're very, very different things. And so yes, Dario Amadei probably has 10 GPUs or maybe more, but I think what they can do with it, with the education and, and, and all of that background understanding is very different to someone who doesn't have the education. Which brings us to a societal question of is there universal basic education No. Yes. And, and for what? To what ends do we need this education? Uh, and I, I would also say, too, um, I've-- you know, continue this thread. There's N-Nvidia, of course, uh, we talked a lot about whether it's a bubble, uh, for Nvidia. Well-- And that was, you know, the, the, you know, talk of the town six months ago, and I think everyone has agreed now there's, there's probably no bubble. Ooh, are-- ah. not. So, so let's, let's… I, I wanna hear how, how you might disagree with me. But, um, you could imagine, you know, there's a question like, is all of the mathematics for, uh, Nvidia circular? Uh, that is, they're giving, you know, stuff, uh, to, to OpenAI, which is then going stuff, uh, back to Nvidia. And so there's a, a real question is, you know, is this just funny math? Uh, but you could imagine with universal bas-basic services, if one path of the multiverse is that we now, um, take all of the human value, human labor, and get rid of the humans and go to every-- whoever has the GPUs and the electricity now has all of the human value. What used to be human labor goes to, uh, GPUs and electricity. If that's true, then who is j- uh, Microsoft and V-Nvidia and Google gonna sell to? Because nobody's gonna be able to buy the things. Uh, like Henry Ford said, "I want everyone on my assembly line to be able to afford the cars that they're, uh, they're building," uh, because without them, then there's no market for the cars. So, um, I wonder if this universal basic services is a way to be able to help that, to be able to say, Yes you may not be working anymore, but, uh, you're gonna be using the Nvidia GPUs that has value, uh, and then we're gonna keep this circle going because we now, uh, and there's, uh, actually a consumer of the GPUs rather than everyone's out of work." Th- this, this goes to a, a really topical thing which a lot of people are discussing these days, which is do we tax AI? Because if the, the companies are not being taxed, how's that circular sort of funneling of the, the, the money gonna go around? And I, I don't think there's a neat answer to that right now. I think this is a really, really deep conversation, um, which I don't have an answer for, so I'm gonna move us back to the bubble conversation. And, and the bubble conversation I think comes down to, I think people now see that AI has value. I think, I think s- six months, 12 months ago, there was a debate around, okay, is this a technology bubble or is this a financial bubble or, you know, what, what kind of bubble is it? And, and I think now everybody agrees there's no technology bubble here. This technology is real and it's helpful even if it's not AGI, which it may be in 12 months, it may be today depending on who you ask. But there's still a question around a financial bubble I think, and, and I think most of that comes down to if you look at all the financial forecasts, if you look at kind of where the revenue is coming from, people are assuming that everyone's gonna be happy to switch over to a consumption token-based pricing model Yeah, I, I think that's probably the way it's gonna go because you're-- then you're actually paying for, uh, the true cost, the, the cost in electricity and the cost of GPUs that everyone has to cover. Um, and so, um, I, I think that that's probably where we're gonna head, and that's gonna butt heads with the seat-based, uh, pricing that legal tech has done for a bunch of years. However, right? yeah, yeah. And everybody, yeah, Salesforce, everybody. Um, so, um, and then, you know, you talked about taxing, uh, uh, AI. Uh, and one Mark Cuban you may have heard is, uh, talked about maybe taxing tokens. And this might be a way to be able to, uh, both, you know, uh, accommodate, uh, maybe it'll maybe dampen demand, uh, for, uh, for, uh, tokens, uh, therefore, you know, maybe cool the market a little bit so society can catch up with it, and also raise much needed revenue to be able to, uh, make people think def-differently about whether they're gonna, you know, token max or not. Uh, is it really worth the billion dollars in tokens or billions of tokens, uh, to be able to do this stupid thing, right? Or, or should I think twice because my tokens are gonna get taxed? Anyway, Or how many agents do I use in my orchestrated workflows, right? Let's throw 100 agents at it Yeah. totally. And yeah, because, you know, and let God sort it out, right? So that's, uh, that's maybe the, um, the Mark Cuban saying, "Hey, let's tax this, let's tax this to slow it down." Uh, but then I've heard the other side of the argument to say,"Cool, like if you tax the number of tokens, the world will go to a tokenless AI." So for example, transformers use tokens, but diffusion models do not. so, so there's a real, a real question, okay, do we just move over to the diffusion models which are untaxed, uh, and therefore Mark Cuban's, uh, idea just goes away? Um, so I, I think that, uh, yeah, there's, there's an idea that, uh, do we tax, uh, you know, t- AI to be able to cool down the market? I guess maybe, but the, the details matter and, uh, and you know, maybe tokens ain't it. Uh, but then the question is what is the way to tax it? To even make it broader than that, theologically, you might have seen also the Pope has come out and Uh, you might remember that I'm, I'm on a too Catholic user group. Uh, yeah, so there's a, a, Monday, my, my Ta- WhatsApp group for the, my Catholic group just, uh, blew up, and there's lots of thoughts on that. Yeah Yeah. Well, I, I wanna hear your thoughts on this i- in a minute. But for those who didn't catch up on the news, the Pope has come out with, uh, I think a few hundred pages, um, on, on the value of AI and, and humanity and, and what all this means. And I think it essentially boils down to, um, a, a critique of the economic model, um, and, and a critique of today's generation of AI, which doesn't actually have any empathy, uh, and, and, and kind of the, the humanness, the, the emotions. Um, and, and therefore it's fine for some exercises, but when it comes to stuff that matters, you're talking to a soulless machine masquerading as intelligence. So I think that's the gist of it. But Damien, what, what was the actual, you know, to the extent you can share, conversations in your chat? Uh, so I, I have to admit that it is four hundred some pages that I've not made it my way through the entire four hundred plus pages. And, uh, I got J- I got GPT to summarize it for me and, and it's funny because you're using GPT to summarize it, and there's a real speculation that maybe the Pope used GPT to or maybe Claude to be able to build the thing. So anyway, so it, it goes back That's your four bullet points in there, and four… bullet point to, you know, something to the bullet point, right? Um, so, uh, so yeah. So, um, I haven't read the whole thing and, uh, but I, I will say that, um, you know, we did, uh, get, uh, a US mathematician, Chicago raised, uh, uh, pope, uh, who as a ma-mathematician, uh, and, and he named himself Pope Leo, uh, after the Leo from the Industrial Revolution. So the whole point of this pope's name is that he knows that he's coming with the second Industrial Revo-Revolution. Uh, that is, that is AI. Uh, he named himself after, uh, the Industrial Revolution. So it's not a surprise that he's, his first encyclical, uh, is focused on, uh, on AI. So, uh, so that's thing number one. Thing number two is that, yes, um, we can use any tool for good and for bad, and this is what the encyclical says. You know, a hammer, uh, I can build things. I can also hit somebody over the head and kill them. Uh, right? So these are tools, uh, that we need to be able to wield wisely. And he warns against, uh, you know, there's, uh, us using it to diminish humanity us using it to raise humanity, uh, to be able to further human, uh, flourishing. Uh, and so that's, uh, I, I think that that's what everyone should be thinking about, is that how, how did social media, diminish humanity? And maybe how, how did we hope that it would raise humanity? Uh, you know, that, that say that social media, gosh, now everyone can talk to everyone. It's democ-- Twitter's democratized, you know, journalism. Isn't this gonna be great? It could have raised humanity, but instead it ultimately diminished it. Um, and so I think what this pope is saying, "Hey, let's learn the lessons from social media," and let's say that, yes, uh, there-- we can cure cancer with AI. Uh, we could be able to maybe solve the energy crisis with AI. Uh, but we could also do really bad things like, you know, um, prompt people to commit suicide and, uh, also be able to, uh, you know, uh, people more lonely rather than less lonely. Um, and so I, I think that, uh, that's, that's ultimately the gist is let's use it to further humanity rather than diminish it. Yeah I, I like that positive framing. Um, and, and I think, like, let's bring it back to legal tech for a moment, right? Uh, I think the talk of the town in the last couple of weeks has been Claude coming in, uh, also Microsoft creating a Word add-in that's targeted to legal, and then the open source stuff which Will Chan has done with Mike OS, uh, and Harvey and Legora both making announcements and the, the, the usual jazz. Um, uh, we haven't covered a market in a very long time, but Damien, what's your kind of feel right now? Has anything changed in May that you did not expect in March? I, I would say that open source is having a moment, and, uh, God bless open source. Uh, and I, I think that, um, you and I talked about this as we are-- we're vibe coding. And, and by the way, the vibe coding episode is, uh… Listeners, w- it's coming out. It's gonna be it's sitting there. It's sitting there waiting You're gonna love it. Uh, but during the vibe coding, you know, if, if, uh, you know, in the past you had to have a dozen, uh, coders spending a year on a thing, that's sunk cost of maybe millions of dollars. Uh, but I vibe coded it on a weekend. Uh, I don't have that sunk cost, therefore, I've made it free and open source because why not, right? And so I didn't just do that for one software program. I've done it now for four software programs, and I'll just keep cranking them out and making them open source. So really, the sunk cost of the old world, you spend a lot of money on coders and personnel to build a thing over a year, you have to recoup your cost. If there's no sunk cost anymore, is open source going to rule the day? And I think the answer is maybe.

Like, like then the question is: 23:14

What do we as for-profit companies, software, what can we do atop what the open sourcers can do? uh, that comes back to data and what, uh, from the application layer can I provide on top of what the open sourcers can? I'm gonna challenge a little bit of that in terms of the, the vibe coded open source movement. A-and, and I speak to a lot of engineers who tell me that the quality of what's out there today has gone down a lot, both from, uh, you know, like just the beauty of the code perspective, um, 'cause vibe coded stuff has a lot of redundancies in it, and, and also from a security perspective. Um, so, so I-I'm, I'm… Hello Mythos is coming out very soon. Um, and, and so I'm not so sure, like, if open source is having a moment because I'm not sure if this is a good moment. Um, um, it, it's definitely gaining popularity, but I'm, I'm, I'm not sure the stuff that's being shared is of the same caliber as, oh, I don't know, Linux, right? Like, um, obviously that's a setup. Linux is like the best there is. Um, so, so I think that's, that's the first thing that comes to mind. And, and the second i-is about this idea of like, you know, w-we… Even if we have access to open source stuff, and let's say it is really high quality stuff that's out there and the volume of it keeps increasing, so no AI slop, high quality open source stuff, we still have this issue as you raised of data and also of assembling the stuff in a way that meets the needs of the user. So even if you have 100 perfect open source projects, someone has to stitch the stack together for a particular law firm and plug it into the data, et cetera, et cetera, et cetera, which is a service that currently this market is not set up to meet That's-- I, I agree with that one hundred percent. Uh, and so to pull on a bit of this, a few of those threads, uh, thing number one is you mentioned Linux, and of course, Linus Torvalds, uh, started Linux a bunch of years ago, and he said, uh, "More eyes make for fewer bugs." Yes and so, so you could imagine that, uh, as you, you could have, you know, option number one is a closed-source proprietary system, uh, where the engineers for that closed-source, uh, proprietary system have done, you know, debugging and security reviews. option number one. Option number two is have an open source where everybody can do security reviews, and everybody could be able to do a p-a, a pull request to say, "Hey, let's fix this security bug, and let's run Mythos over this open source thing," uh, to be able to maybe, uh, batten down the hatches. So between option one, a proprietary system, and option two, open source, there's a pretty decent argument that the open source would be maybe safer because it'll have more eyes that will make fewer bugs and se- fewer security, uh, vulnerabilities. So that's thing number one. Thing number two, to your point about, uh, you know, we need to assemble these open source things to meet the users' needs, I a hundred percent agree. Um, and you can imagine that maybe where we're headed is that, um, here's an example. Mike Bommarito, building something for a court, uh, Yep. It's gonna help the court do its work. Uh, and so the court, we said, "Hey, can you make this output free and open source?" And they said, "Yeah, of course." so then you can imagine that now being used by, say, Texas courts, uh, and then maybe used by Arizona courts and maybe used by Colorado courts. And then as the personnel say, "Oh, we found this bug," they contribute it back to the open source, uh, be able to say now, you know, say if Arizona s-says, "Here's a bug that's now fixed," now that's the, the beneficiaries are Texas and, uh, and Colorado, et cetera. So a-as, as the open source ecosystem gets better, I think that the assembling to meet the users' needs is gonna be far easier 'cause I can say to Claude Code,"Hey, take this open source repo and tweak it to what I want to do. I don't want twelve buttons. I want three buttons." And then Claude Code will make my version of the thing have three buttons. Uh, and so I think we're maybe if one of our paths, uh, of the, of the multiverse is that maybe we get customized software that is suited for me, if it's open source, maybe I can do that customization more easily. I like that universe. Um, I, I, I think that means there'll be more jobs for engineers than lawyers going forward, um, which is not a bad thing. Um, I do find it interesting though because it, it presupposes that users know what they want. Um, and, and Damien, I think you and I have both been in the product game for long enough to say that users often do not know what they want. They think they know something, and then when, when you deliver it to them, they're like, "Oh, no, that's, that's not what I meant. When I say three buttons, I mean three big buttons and three sub buttons." Yes. Yes, and you're missing the eight buttons that you took off that I, I thought I didn't want, but it turns out I kind of need. Um, and, and that's, you know, you and I have been building products for long enough and are product people to know that, um, even, even good product people have that same problem. Uh, Yes I'm, as I'm vibe coding with Claude, I say, "Give me the three buttons." And then once I see it on the screen, I'm like, "Oh, no, that doesn't work at all." Uh, so I, I wonder if the world is gonna become product people. Where, um, that if I'm just a, a judicial clerk, uh, or if I'm just a, in the, in the clerk's office, a clerk of court, I speak to Claude Code to say, "You know, I thought I only wanted three buttons, but I wanted those three buttons with the other sub buttons. Claude, can you please do that for me?" And then 20 minutes later, I have the thing I want. Uh, and so I wonder if the world is gonna be like you and I are, to be able to say, "I thought that's what I needed, but turns out I don't need that." And maybe we can just have customized things that are, you know, software for one Anyway, whatever happened to Jony Ive? What has, what, what has, what has he been producing? Because, like, if… Imagine, imagine Steve Jobs was alive and he got access to Claude Code, and he would just vibe out everything himself, right? It'll be the best UX for everybody. Jony Ive is the, the spiritual successor to, to Steve Jobs, at least his design ethos. He's been bought out by OpenAI, and I've heard not a whimper for the last year Well, it's easy to fling bits. It's hard to sling atoms. Uh, so he's, he's an atom slinger, right? He's builds, uh, physical products. And so I, I wonder if, uh, he's, you know, he can't vibe code, uh, this physical thing with Claude. Uh, and I, I was just, uh, there was a, um, a Lenny's Podcast. I don't know if you know Lenny's No, I don't Valley-- He's a, a Silicon Valley guy. He, um, interviewed the, a woman who worked for, um, Apple's, um… She worked for O- Oculus in product. She also worked for Apple's, uh, Vision Pro, and she's doing, um, I think now Google's glasses. Anyway, so she works in, in physical products, not software. And she said the difference between, um, software is software you can iterate a million times, right? And there's, there's really, uh… And you could t-take it to customers and then iterate again. But with products, you got like three bites at the apple, and then it ships. Um, and so you can't recompile and recompile like you can in software. Like, you only got a few bites at the apple, so you gotta think really hard as to, you know, before it goes to the manufacturing facility, what this design is going to be. Um, so I wonder if Jony Ive is kind of, uh, stuck with, uh, you know, everyone's running fast with Claude Code 'cause you can iterate literally millions of times. Yeah you can't do that with atoms. You can't do that with physical products. I thought 3D printing solved that, or at least partially solved that a few years ago Maybe, but, uh, uh, you can't do the circuits with 3D printing, right? I App. as, as he, as he does electronics, like he has to say, "Okay, this circuit board has to fit in this physical constraint," right? And, uh, this-- these wires have to go this place and, uh, you know, you can't 3D print those things Yeah. Oh man, I, I'm looking forward to what he produces, 'cause I think, you know, there, there's a, there's a handful of people in the universe who are very, very good at design, and I think Johnny's earned his sort of reputation and place among that very esteemed crowd. So, um, I, I'm, I'm looking forward to seeing that. But speaking of esteemed people, uh, let- let's talk about esteemed people in legal tech, um, and, and the big movers and shakers. Um, in the last couple of months… I guess I'll throw the question to you first, and I'll, I'll add my thoughts to this. Um, what have you seen that's really impressed you? I have been really impressed with, uh, agentic, uh, you know, agents, whether it be Open Claw or otherwise, and getting b-back to the open source. There's Lavern AI, um, has open Yes their associate agent and their partner agent and their, you know, uh, data privacy agent, and their, you know, they have different agents for different skills that, and essentially replicates the types of agents, uh, types of lawyers that would be in a law firm. Uh, so that, uh, so, um, things like that. Uh, you know, I've been saying that somebody's gonna build this. And I, you know, uh, and I've, I've actually toyed with building it myself, and it turns out Lavern AI did this and made it free and open source. Uh, and So, Go yeah, I, I was gonna say, so, uh, uh, Anti, I'm, I'm, I hope I'm pronouncing his name correctly, I think that's right um, he, he open sourced this, and I remember during the month of April, he didn't open source it because he was trying to figure out, like, okay, can he do anything with this commercially? What does it look like? And I think ultimately, like, uh, and this is kind of my read of the room, ultimately, I think consumers of legal services want to pay for a human on the other end rather than a pure agentic law firm. Um, and I wonder if that will ever change Uh, uh, it won't, uh, because we humans like buying from humans. Uh, and, you know, uh, I think we've talked about this in the past with the law firm that, um, you know, if you are an excellent associate, um, that's not gonna cut it if you're not bringing in business. Uh, you're not gonna make partner unless you've got a book of business that you can now say, "Okay, now we're gonna elevate you to the equity partnership because Yeah bring in business, because you're a rainmaker." And another word for rainmaker is salesperson. And good salespeople sell business, and people like to buy from humans. They like to buy from good salespeople. And so, um, you could imagine a law firm today could use something like Lavern AI. And using all of these agent, you know, the swarm of two dozen different agents putting together something, packaging it for me, the human, that serves up this output with my human face and my human, um, uh, you know, mellifluous voice that, uh, goes out to this, uh, this really, uh, grateful client that really is happy to have a human, uh, meat puppet, uh, provide this output that the machine had actually created. Um, and so I think that the rainmakers, otherwise known as salespeople, um, that might be all that's left, uh, as we go forward and do things. That I'm gonna say that I, as a meat puppet rainmaker, um, have my little bits of taste, uh, in, in the way that, uh, you know, Rick Rubin has taste as a producer. That I've put my, uh, my, you know, thousand dollar an hour, two thousand dollar, ten thousand dollar an hour law firm brain and taste into this AI output in a way that you as a client will find delightful. Uh, so I think that you're right, that we're gonna continue to have humans selling to humans even if the input, uh, and the output was all done by machines. But a f- a far fewer number of humans, right? Like, and, Or more. Oh, okay. Okay Uh, Jevons Paradox, right? 92% of legal needs are unmet because we lawyers are too expensive. So maybe instead of 10 clients in a given month, I have 100 or 1,000 clients. So maybe we need more lawyers You know, th- this is a part that I have no idea how it's gonna play out. Like, there, there is a, you know, talking a multiverse analogy. There, there is, there is a universe where I can see it's gonna be constrained by spend. It's not as if the bottom 98% or 92% of legal needs are satisfied, it's just they don't have the resources to spend on humans in the first place. Um, and so yeah, okay, Jevons paradox says there'll be more, you know, potential to serve those legal needs, but those legal needs couldn't afford it in the first place. So that's kind of like, you know, thought number one, right? But on the other hand, you're right. Like, you know, there is a universe where they can just, well, pay for more, and there'll be more services and, and so on. So I don't, I don't know how it's gonna play out. I think, I think it's, uh, a, a really interesting kind of like bifurcation depending on the cost of legal services A-and the cost is gonna be constrained by who has the GPUs and who has the electricity. Uh, because right now, if I want to token max, um, I can with my two hundred dollar a month personal, uh, Claude, uh, Max account, uh, but I cannot token max on, you know, Clio gives me, uh, my Claude too, um, and that's, that's more constrained than my two hundred dollars a month. So I can do much less as a, uh, a Clio employee than I can as a personal person paying two hundred dollars a month for Claude, uh, Claude Max. And so, um, I think that constraint, uh, is going to be, uh, what you just said, because as we say, you know, if GPU and compute relates to how many legal services you can consume, uh, that is as, as legal services are created by those GPUs, and if there's only so many GPUs to go around, um, do the rich people get the legal services, also known as GPUs, and the poor people can't afford the GPUs, therefore they don't get the legal services? Um, I think that's, that's one of the multiverse to say that, you know, this is how we separate the rich from the poor, that the people who have GPUs can get the legal services. Those who don't have the GPUs, the poor people, uh, don't get the legal services. So and, and I think that, that, um, that kind of GPU scarcity, uh, that we're in right now and maybe will continue to be as the demand for GPUs keeps going up and up and up, um, makes Dario, uh, and Sam and, uh, Elon, I guess, no, Elon's out of the game now. He's just supplying GPUs. He's not actually b-building, uh, foundation models. But, um, Sam and Dario are, are saying,"Okay, how can I maximize the value per token? Um, how can I maximize if we are token constrained?" And because, you know, everyone's searching for GPUs, um, then we want to maximize our value per token. And, uh, Dario has said that the highest value per token is right now coding. That's why he's built Claude, uh, you know, uh, Claude Code. Um, the second highest value is legal Legal and so, and so that's why Claude for Legal has come about. And so because Dario is not selling software, is selling GPUs, uh, that is a foundational model wrapped around GPUs, um, he's happy to give away free and open source Claude for Legal because that's not his business. not-- His business is not the workflow that you do. His business is maximizing the value per token, which legal has a high value per token. uh, uh, so you can imagine, um, what does that mean for lawyers? Um, well, we also have to maximize our value per token, uh, that we, we wanna be able to build systems that are efficient, uh, and not token maxing and being, uh, you know, us-using an, uh, you know, an Open Claw thing to spin all night on something that a symbolic system could have done in twenty minutes, uh, with CPUs and not GPUs. And so we've talked a lot on this pod about symbolic AI versus, uh, versus neural nets and large language models, and symbolic AI is just good old-fashioned AI. So I wonder if this GPU constraint is gonna push us more towards let's have the Claude codes create symbolic AI to be able to use CPUs that are not GPU intensive, to be able to do a lot of this legal work a- and make it deterministic, uh, that is reliable on the, on the, uh, symbolic AI side, and then let the large language models fill in the gaps with a probabilistic, uh, therefore unreliable but more flexible large language model. So, uh, that's all a way of saying that let's maximize our value per token and don't use GPUs for LLMs where you can use CPUs for large language models. So for, symbolic AI we cannot agree more violently on this point. Like, like if, if, if we had hammers and nails, we'll be both just, you know, nailing all the coffin shuts, all the coffins shut right now. Um, a- and, and, and like the simple economics of this for the large language model providers is easy. It's revenue equals the cost per token multiplied by the number of tokens. So they want you to max out both the number of tokens and the cost per token, whether it is for read or write. Like, that's, that's their simple economic model. Um, every economic model is simple until you actually, you know, try and do stuff with it. But for, for us as the consumers of these tokens, what we want to do is we want to pick the cheapest token that would do the job, i.e., the less intelligent models, and use the minimum number of tokens possible to get the job done. And I think this is where the tension's gonna be. How do we as consumers, and, and, and when I say consumers, I mean both legal tech builders who build this technology and the law firms and legal services providers who use the technology to consume the tokens, how do we reduce the number of tokens and also, uh, use cheaper models where possible to constrain and reduce the costs that we have to bear in order to deliver the services? And I think this is a question that the market hasn't started to answer yet, but there are a few people, I'll put myself in that category, working on how to achieve that. Um, and I think this is gonna be the interesting stuff as we move out of the VC-subsidized era, which is already coming to a close because I think Claude recently announced that everything now is on a consumption model. There's no longer all-you-can-eat plans Yeah. So, uh, l- let's, uh, let's put a real, um economics hat on. And a friend of mine is, uh, uh, used to be the chief economist of Spotify. Uh, so he's a, he Ooh. who, who, uh, fi-figured out how the, you know, Spotify was and the music industry is gonna make money. Um, but he always talks about, um, uh, uh, people, uh, that he worked with, one of the business leaders as he, uh, my friend had just finished explaining, uh, an economic principle, uh, the business person said, "You know, I don't, I-- What I want," uh, the business person said, "is a one-handed economist." Uh, because, uh, Okay. because, uh, uh, Yes. Yes, go on, the one hand, it's this, but on the other hand, it's that." So I, I just want a one-handed economist. I don't want a two-handed economist. Um, anyway, or three-handed, as the case may be. Uh, but when we think about the economist, you say, "Okay, I as, um, a legal tech company or I as a law firm or I as a law department, um, have to think about what is my revenue and what is my cost, and revenue minus cost equals profit margin." Uh, and so if you push down the cost, you increase your profit margin. This is basic economics, right? And so when we, uh, then apply that to AI, um, of course, um, one path is to throw OpenClaw at it. Uh, and that is a very inefficient way that increase your costs and cuts into your profit margin because OpenClaw is gonna waste a whole bunch of tokens by just going down rabbit holes that it doesn't need to and arguing with itself over stupid s-stuff that doesn't matter. Um, so path number one is, is OpenClaw. Uh, path number two is to use something like MCPs. Uh, and MCPs are better than OpenClaw, but it's still gonna waste a lot of tokens because it's gonna, um, like for, uh, for example, um, FOLIO has an MCP. Uh, you can throw an entire document in that, uh, and say, "Classify, uh, the contract type and classify all the clauses in that contract type and classify a whole bunch of other things that FOLIO has in its eighteen thousand tags." So it can do that via MCP, but it does it very inefficiently. So, so path number one is, uh, OpenClaw. Path number two is MCP, which is, uh, inefficient, but good. Um, path number three is to use Enrich, uh, where you can actually take that, and we use a lot of CPUs, not GPUs, CPUs with good old-fashioned AI. and that good old-fashioned AI says that if there's a string that says motion to dismiss, it tags it up with a CPU. It doesn't need AI to say motion to dismiss. It says motion to dismiss. And the, uh, FOLIO also knows that a motion to dismiss is also known as a "demurrer" in California. It's also known as a motion to terminate, also known as an MTD, uh, in time entries. So it has all of that coded up in symbolic language that uses not GPUs. And so option three is to spend just CPUs and be able to tag it all up, um, and then maybe throw a very, uh, cheap LLM on top of that to be able to validate that, that the tag was correct. Um, so anyway, so option one, use OpenClaw and be very wasteful with tokens. Option two, use MCPs, which are wasteful but helpful. Or option three, use symbolic AI mostly with a little sprinkling of LLMs on the top of it. Um, we're gonna f- more frequently societally go to number three, and, uh, the smartest people in legal tech are also doing number three. Completely agree with like the direction that you're going in with this one. I, I do wonder, however, like the number of people who are capable of building model three that you've described, right? The symbolic rules, the heuristics, the, the string matching, all of that. I think that is a, a very, very small number of people who do have both the, uh, domain expertise in legal and the technical expertise to be able to put that together. Um, but I think you're about to say maybe that's changing because the Claude Code is making it so easy to just vibe out these solutions Uh, that's right. Uh, there's a FOLIO resources, uh, Google Doc that I just gave to that Fortune 100 company, and I said to this paralegal, who is an amazing vibe coder but an awful regular coder, but I said, "Literally go to FOLIO resources, uh, copy all of the GitHub repos for FOLIO Enrich and FOLIO Mapper and the FOLIO MCP and the FOLIO Python library and the FOLIO API. Just copy all of those URLs and say to Claude Code, 'How can I use this within my, uh, within my Fortune 100 company? Uh, here's all the systems I have, here's all the integrations I have. Um, tell me how I can use CPUs to be able to do things that I don't need to use GP-- uh, GPUs for.'" And so I, I think that, um, yeah, uh, there are only a few people in the world that can do what I've described. Uh, I'm one of them. Uh, you're one of the others. Uh, and, uh, I would say that, um, as I make things open source, um, anybody can be able to stand on everyone else's shoulders. Remember that conversation we had a long time ago about what Gen AI is really good at, and you mentioned Gen AI is really, really good at ideation. Um, and then there's the extrapolation, interpolation, all of that stuff. But the ideation of, like, throwing your background information into Claude and saying, "Okay, here are my circumstances. Go and figure out for me how Gen AI can help," Mm-hmm. I think that should be an exercise that every person listening to this podcast should do today, if they have not already done it. Yes, 100%. Uh, yeah, go, go and swim, folks. Uh, y- th- you're, uh, nothing's gonna teach you how to swim. Let's, let's do this. Um, there was a paper, uh, from last year and, and I, I wanna get your opinion on whether you think this has changed. And the paper from last year was by a lady called Stephanie… Let me see if I can find it from my search history. Um, Stephanie, uh, Stephanie Tully. Um, and Stephanie Tully in 2025, um, published a research that says lower AI literacy predicts more frequent usage of AI, a- as in when someone knows less about AI, they tend to use AI more. When someone knows more about AI, they tend to use AI less or trust it less. Do you think that's shifting, and do you think the market is moving since this came out in 2025? I think that, uh, that's-- I, I believe that study just anecdotally. That just seems right to me. Um, and part of that is the Dunning-Kruger effect, uh, is that, uh, the more, uh, the less I know about something, the more I think I know about it. Uh, but then the more I learn about the thing, the more I realize I don't know. Uh, and this goes back to Socrates, right? The, the more I know, the, uh, the s-- uh, the dumber I am, essentially, Yes. Yes So anyway, so I, I think that, um, the, the less, uh, uh, facile, uh, someone is with AI, the more they use AI and trust AI. I, I believe that. And the more facile and the more you use AI, the less you trust it, but maybe the more you can use it as an, a really extremely, uh, effective tool, uh, that maybe those who, uh, are more, uh, neophytes. So I, I believe this concept, uh, that you've just described, uh, that is probably true. Uh, and the real question is, um, as society's floor keeps getting raised and as the dumb neophytes, uh, become no-not neophytes anymore, but actually moderate users and actually better users, I, I wonder if that goes away. And w-wond-wonder if even, uh, the neophytes are gonna be more skeptical of the AI and more, um, using it as a tool rather than a magic oracle. Have you seen that, uh, meme where you've got this kind of bell curve, uh, where on one end there's an idiot, on the other end there's a Jedi Knight, and in the middle it's like normal person? Mm-hmm. And it, it feels a little bit like that where, um, at least the first half of it, where the idiot's gonna be the one to go, "Oh my God, AI solves everything. I'm gonna use it for everything." And the normal person who's more knowledgeable is like, "No, no, guys, come on. It's a probability… It's a probability machine, even if it's seemingly intelligent." But then I think there is that extra kind of tail on the other side where someone who's so knowledgeable about AI, someone like, um, Andrej Karpathy, for example, is the Jedi Knight of I'll use AI for everything, knowing what the limitations are. And I think this is where, you know, Dunning-Kruger aside, I think this is where the, the sort of usage chart is starting to look like based on what I'm looking at market I think that's right. And any particular user, uh, who is a non-expert in a field that uses AI for that non-expert, uh, um, you can imagine that, uh, our-- the data scientists that you and I work with, uh, refer to things called precision and recall. Yes that, uh, precision is the thing accurate, and recall, did it get all the things that were needed to be able to give me the right answer? Uh, and so a, an expert-- So you can imagine a non-expert asks, uh, something about, uh, we'll say, you know, an expert field, uh, law, physics, et cetera. Uh, and if that non-expert says,"This gave an amazing answer," but the expert would say, "No, no, no, it was not precise. It was not accurate for these reasons. Here's the bad output, Yep it was missing all sorts of things." The recall was also bad. Uh, and so precision is maybe easier to be able for, have the non-expert figure out because if you throw enough bots at something, it'll realize that that case was hallucinated, uh, because they could be able to validate and say, "That wasn't quite precise." The recall part, the recall is, is harder because, uh, because if you need the corpus of data, that is you'd need, in legal, you need yesterday's case. You need Yeah and yesterday's regulation, and if that's not part of your dataset, ain't nobody in the world that is gonna be able to pull, uh, you know, make, uh, uh, s- uh, purse out of a sow's ear, uh, as my grandmother would say, right? Uh, there's, uh, again, another way to say it is garbage in, garbage out. If you don't have Yep case, you can't have the recall to be able to say that it's missing yesterday's case. So I, I would say that as non-experts use these, uh, these systems, um, they are not able to assess precision and recall in the way that experts can. Uh, it- it's, it's a little bit like, um, this, this idea of like you can't hallucinate the truth. Um, um, and if you can, it's like the broken clock thing, right? You just happen to get it right. Um, so being able to find the data, organize the data, I don't think that ever goes away. Um, in, in a conversation I was having with, uh, an innovation team a couple of days ago, I just casually mentioned, and, or, or rather I asked like, "Hey, guys, are you finding that the KM team is now subservient to the AI team?" And I got this kind of quiet nod. Um, and, and I, I feel like, I feel like that pendulum has to swing back at some point, um, where the knowledge team ultimately is the one that, you know, to use a poker analogy, holding all the cards Uh, yeah. Uh, I, I think that's, I think that's right. Yeah, there's, uh, the knowledge team has always been, in my opinion, the, the most important, uh, holder of the cards because they, um, yeah, there's, uh, what is the value of your law firm? And people said, "Well, our value is in our people." Uh, and, you know, there's, as someone leaves the firm, that value walks out the door. Uh, and really the only sustainable value for a law firm is that lawyer had in their brains that they put into a knowledge base, whether that be a document corpus or a here's how to do this type of matter in this jurisdiction, um, putting that in a knowledge base before they walk out of the door. Um, I think the smartest firms are realizing that, hey, we need to enrich this corpus, uh, otherwise we're gonna have a whole bunch of people walking out the door and taking all of our firm value with it. Uh, so we need to be able to, uh, make sure that this knowledge base, uh, is, is good. So, uh, God bless the curators from the knowledge management and the librarians of the world that are trying to, uh, have been trying for decades to build up this knowledge base, and maybe the firms are figuring out now that, hey, we should have listened to them I think that's a nice callback to the very first thing we said on this call, which is the $500 million investment by Kirkland into its own AI program or w- whatever it is that they're trying to build. Um, because this sounds like a big part of it is gonna be the retention of the expert knowledge that Kirkland has on the market Uh, and, and that makes me think, you know, we talked about, you know, is, is that part of that $500 million, uh, that might billable hours? Uh, you could also imagine, uh, you and I have talked before on the pod as, uh, as I'm a lawyer that, uh, the law firm, say, Kirkland says, "Hey, lawyer, why don't you put the stuff in your brain into this corpus?" Um, I as a lawyer could say, "Hey, I was thinking about jumping ship to this other firm. Um, what are you gonna pay me to essentially, uh, d-d-- what's the value that's gonna transact?" So I wonder if they're, uh, maybe the $500 million goes to what they're gonna pay their lawyers to take stuff from the lawyers' The royalty that we talked about once upon a time. And we go back to the Spotify economist. Everything calls back to everything else, Damien Yeah. it's true. Yeah, and I, I… And royalties, and I guess, yeah, if… But the question is like, like if there's Practical Law-y kind of insights that now I transfer from my brain into, you know, there's the way to do this kind of deal or this kind of litigation, if that's now in there, what's really unique to me, uh, versus something that every Tom, Dick, and Harry, uh, that is doing litigation, uh, knows these things? Uh, and what's the value then for this, uh, I- I as a lawyer just provide pablum something that everybody knows. Uh, are my royalties gonna be just as high as somebody who has the unique insights that does trillion-dollar deals? Oh, no. I mean, uh, okay, the, the obvious answer is no, it's not going to be. But I, I, I don't know where that value lies or how to define that or how to like, you know, put the parameters on that. Um, but this is gonna be an interesting question. Perhaps we can explore that when we get together next week in person I-- That'll be so much fun. And I, I have two minutes before my next thing. Let's talk about optimism. Um, I, I continue to be optimistic, uh, and I'd be interested in yours. I'm, uh, continuing to be optimistic about open source, and I'm con- optimistic about, um, the excitement I see in my law departments I speak with and my law firms. Um, open source plus excitement on all sides of the ecosystem make me very, uh, optimistic. Uh, what are you optimistic about? I'm optimistic about the benchmark for what is good enough. I think, I think the benchmark for what is good enough from gen AI is moving up, and also the acceptance of what is defined as good enough is actually getting more realistic. People are no longer expecting 99.99% accuracy because you never got that from humans in the first place. So I think that there's a convergence of where the good enough line sits, and then once we have that, there'll be clarity on what gen AI can actually do in this ecosystem Amen. Uh, that is an excellent way to end it. Uh, Horace, I love our conversations so much. I love them in 2D. I'll love them even more in 3D. Thank you so much, and let's get this out unedited. So, uh, you're gonna see this warts and all, folks. Boom! Damien, have a lovely weekend, and listeners, thank you for listening. We'll see you again soon. Thanks everyone.