#69 Why VCs Are Betting Billions On Legal Tech

Hello everyone and welcome to Pearls on Gloves Off. I'm Mariel Carroll, your host and chief operating officer at Goodwin. Today I'm speaking with an old friend, Jake Saper, a general partner at venture capital firm Emergence Capital. Jake saw the legal tech wave coming before almost anyone else. While most investors were still steering clear of this space, Jake was leaning in, which is why we were connected years ago. As he started learning about this space, he and his firm placed early bets on companies like Simple Legal and Ironclad that are now foundational to how legal work gets done. Jake has spent over a decade at Emergence Capital where he's helped build category defining companies including Zoom, Guru, Assembled Logic, Unifi, and many more. Most importantly though, he's one of the clearest thinkers out there on what AI is really going to mean for SaaS and for legal and for the way we all work. This conversation was really informative for me and frankly, Jake just speaks my language because he's not just a big picture futurist, but he's also a framework builder and pattern recognizer, which you will see in this episode. We're going to talk about why he believed in legal tech so early on, how his thinking has evolved, and where he sees AI reshaping our future not just for software, but but for the entire legal ecosystem. Plus, we talk about what he looks for in founders, how to navigate a crowded tech market, and what's real versus hype when it comes to AI. I hope you enjoy From AI to.

M and A change is accelerating. Stay ahead of the curve with Exhibit A charts and graphics on the latest legal and market trends that are impacting your business. Visit goodwinlaw.com exhibit A to check out our data driven insights.

Well, I've been, as you know, super excited to get you on this podcast for a while now. Let me just preface, you are one of the earliest investors that I know in legal tech and you were looking at it before people even knew what legal ops was. So you saw this wave coming and you stuck with it. You're not specifically legal focused. Tell us about that.

We are not legal focused, but we are focused on B2B software more broadly. So the firm is a little over two decades old. Started with the thesis that software would move from on premise to the cloud, and that thesis largely worked. Our very first investment was in Salesforce and we've been lucky to partner with great companies like Zoom and Box and of course Ironclad on the legal side of things. We've also done other kind of more vertically focused software companies A company called Ziva Systems, which is actually the largest vertical software company focused in the pharmaceutical industry, Lend and mortgage and a bunch of others. So that's the style of investing we do. We tend to invest pretty early as well. So we aspire to be the most important partner to our founders. We make very few investments so that partners have time to really go deep and try to help each company grow. And I have been here for almost 11 years now, which is crazy, but I love this job. I feel very grateful and lucky I get to do it.

Yeah. I have always said legal is no different than any other industry. So I appreciate your approach. Where you see this as B2B or any other vertical. I mentioned earlier you were early to this space. What attracted you? What did you see that others didn't back then?

Yeah. So back then, like way before there was even AI or at least AI as it was good enough to be deployed in B2B software.

We didn't even have legal software back then.

Honestly, a lot of my thought process was this is an extraordinarily high value function that is extraordinarily inefficient. And so if you just use that framework of like who was paid a lot and where are their inefficiencies, legal is like top right.

Yeah.

And so that was a pretty obvious place for us to spend time and try to make investments. Our very first legal investment was in a company called Simple Legal, which was a billing software company that helped lawyers manage their billing process. And that company sold a few years ago and. And then we invested in Ironclad, which is obviously a company you and I got to partner closely on.

Yeah, I remember meeting you and Santi and you were getting educated on this space. And when you first landed on this is an inefficient vertical, I just remember you being like, let me get this straight. It's totally inefficient. Every single company in the world has this function and it's a totally untapped software market. I'm like, yes, we have nothing. That was the beginning. And of course we had an era of really old clunky on prem software. And then you arrived at the time that new entrants were coming in that were really focused on simplicity, user interfaces, cloud based systems. And then there was an explosion probably around 2015, 2016, into more of these and tons of VC money going into this space. What made you kind of stick around and do you think this space has evolved in the years that you've been involved?

Yeah, there are a couple Dynamics that have happened since our first interest in the space. One is the rise of the Legal Ops Persona. That Persona was not a super common Persona over a decade ago. And Mary, you had a lot of insight leading Bach, leading the organization that was championing that Persona. And if you have a Persona like that whose job is to make these function run well, you better believe that those people are going to demand great software.

Yeah.

And part of our thesis around investing in Ironclad was people are going to demand better software than the existing CLM software, particularly if they're empowered to be thinking about this for their whole jobs. So that happens and I think that continues to happen, which is a good thing for the industry as it makes the functions more efficient. The other thing that's very Obvious now is AI. We have been excited about AI in B2B software since about the time we met Mary, about 2015, 2016. Our vision back then was something we called Coaching Networks. And the idea was how can you use AI to help coach workers in real time as they do their jobs and learn from everyone else that's doing that task to improve the coaching for each incremental person on the network. And when I first met Jason and Kai at Ironclad, even before they'd started the company, we were talking about this concept. We had this idea like would it be possible to have a contract and be able to coach someone on the right laws to use in a given situation to ensure that the contract closes 10% faster than the average contract? And back in 2015, 2016, that was a cool concept, but the technology wasn't really there yet.

Sure.

And obviously with the rise of LLMs over the past few years, that has become very possible. And you're now seeing that in Ironclad and other obviously legal software as well. In many ways I think we, we at Emergence got the concept right. In Coaching Networks, we got the branding of it wrong. Copilot is now the way that that concept is referred to generally in the industry. And it's probably better than Coaching Networks. But we like the network effect concept because it really is that if this works well, the more people you have using the product, the better the product will get for everyone else using it.

And AI has obviously played a huge role in disrupting this space. I think we're at a very different moment in time with legal tech as AI is everywhere. The legal teams are continue to be overwhelmed, but now they're also overwhelmed by the amount of tech that's coming at them. We've gone from, we have Nothing. Can somebody please build some tools for us to. Now we can't even stay on top of what is out there. And I think you and I both note software and AI is not always what it claims to be and have like fancy demos. As an investor, how is your approach to investing in legal tech shifted over time? It is like the hot thing right now when it comes to AI. Everyone talks about AI is perfect for legal. I have mixed views on that. But curious. Does that change for you?

Let's unpack. Like is it perfect for legal? And then like how we think about investing. So I think the reason why people are excited about AI in legal one is legal deals in words and LLMs are good at words. And so if LLMs are really good at predicting what the next word should be and the law has like massive corpuses of data, then it's an obvious place to train models and deploy models. So I think like technologically there's, there's logic behind why this could work in the law. The other concept here, as we mentioned before, is this is a very high paid and inefficient profession and if you are able to make it more efficient, that is attractive in many ways. The obvious question that everyone, many people are wrestling with now is what happens with the per hour billing model in a world where things get more efficient. And I think like you're going to see more and more evolution as that progresses.

That's why I came to this job, so I can solve that problem too.

It's a fascinating question. I don't know. I don't think anyone really had the full answer yet. It is possible that no organization's business model will change more going forward than the law firm.

Yeah, I agree with that totally.

Or they die.

Yeah. For someone who wants to do transformation, which is my passion, like there is no better place to be than right here, right now at a firm that is open to it and is interested in thinking about what the model and what the firm of the future should look like and trying to get there. Because AI is just technology and being a first mover and adopting it, it's so easy to adopt. The adoption is much easier than any other software we've seen in our history that if you hear that it's ready and someone else is using it, great, we'll buy it tomorrow and roll it out to everybody. But the problem is that does you no good when your entire business model is completely at odds with efficiency. And AI is really from a large part an efficiency tool. So yes, the firms we really have to think about all the incentives, the pricing, the structure, the go to market, the delivery and how do we pivot so that we're aligned with our clients and giving them more value which I think makes us more profitable. And the whole thing makes sense. It's just how do you get from A to B?

Yeah. And the shipping goes the most hard. It's easier to be a GC right now. GC has a free market incentive to try to make their team as efficient as possible.

That's right. You have invested in technology on the legal side that has been primarily focused on the in house teams. And there's good reason because in house teams were really eager to get tech in again to the point about legal ops. Like it's someone's job to do more with less because you're a cost center and, and when you're on the revenue side there's less incentive to do it faster and cheaper.

But I don't think it's unreasonable to think that you could move to some sort of more outcomes based model in legal.

Oh, I think it's inevitable. Yeah, absolutely.

And I think honestly that'll be good because it's more aligned for everyone. Like I've always felt like when I'm using legal services like so wary of, okay, we're going back and forth on this but this is a minor thing and I'm wasting money. It doesn't matter if the thing was like we're all just aligned on getting to some outcome we agree is good and I'll pay you X dollars for that. I think that the cognitive burden of the consumer of legal services would also be lowered.

Yeah. And it's a trust relationship. You're supposed to be a trusted strategic advisor. And when everything is. Are you actually calling me to just burn time? You can't help but think that when the model is completely misaligned. So like I said, I think it's inevitable. The thing I was most excited about for AI showing up in our space is not even the potential that it brings from a tech perspective, but just like forcing these conversations finally it's healthy.

Because it also like it forces you to recognize like what is the real value I'm adding and how do I charge for it. The ideal pricing model is based upon the value you bring.

Absolutely.

And I think per hour is a poor proxy for that and it gets poorer as AI gets better. And so the best model will be how am I bringing value to you? What is it worth to you? And I want to take 10, 20, 30% of the value I'm bringing to you and take it from me.

Yeah. So going back to then, from an investment perspective, how do you look at legal AI now as there's just a proliferation of companies.

Yeah. So one of the things you said was both exciting and scary as from an investor years, you're like, yeah, if we hear someone's using this software, we'll buy it, we'll roll it out to everyone. The reason why that is exciting, obviously is there's a lot of like people are comparing to each other and they are adopting, which is part of the reason why you're seeing such growth in these companies. The reason it feels a little scary is I think there's a lot of products right now that are being deployed but not heavily used.

Yeah.

And we're now coming up on roughly 12ish months since some of these things started to come out. 12, 18 months. And so I think you're starting to see in second half of this year you'll see even more the kind of reckoning, or at least the initial reckoning of, okay, we bought this, we rolled out to everyone, but like 10% of people are using it or everyone's using it, but it's not actually I don't know about its value or what's wrong. And so I don't trust it. And the flip side of the law being an expensive, important function is you just can't mess it up. My experience talking to lawyers who have used these products is there's no second chance for the AI that these lawyers are giving it. The lawyers are skeptical coming in, you give them the tool. And if the tool has a hallucination or is wrong in some capacity, I think a lot of lawyers reaction is, I passed the bar. It's my name on this, my bar number associated with this. I'm not letting this software mess up my good reputation in the law. And I think that's like a totally understandable reaction. And so I think like over the next six to 12 months, you're going to see this kind of reckoning of legal products that have been sold that either weren't deployed or deployed but not used or were used and people discovered, I don't trust it. I don't think it's going to be the whole industry, to be clear, but I think that there are going to be a lot of vendors that kind of go through that trough of disillusionment, so to speak. And to answer your question, for me as an investor, it goes back to just like first principles of investing, which is instead of just looking at the top line revenue curve of a company, how much have they sold? Your job is really to understand how much value is this product bringing. And I think we too in the venture industry have been starry eyed about this technology and have let the fast growth of a lot of these application layer companies brainwash us a bit to say, oh, this is all awesome, but really what we're investing in isn't revenue growth. What we're investing in is value creation. Yeah, revenue growth should be a lagging indicator of value creation. And what I care about more than even how fast your revenue growth is going is like, did you actually completely change the way Goodwin does their job? Like how many people at Goodwin are using the product, what's been the percent uplift and are they continuing to use it? What's the retention curve look like? What's the uplift curve look like over a 3, 612 month period? And if I can build confidence in that, and I can build confidence that the product itself, the AI, is getting better as people use it. This is something my partner Gordon calls the Emergence effect. The idea that like as more and more data gets fed into the product with more people using it, the product itself gets better and better. If I can see that positive trend over time, then I'm pretty excited to better that company. I'm also willing to bet on ourselves at Emergence to help them figure out how to monetize the value they're creating. What I can't do with a VC is figure out how to make a product add value to a target customer. That has to be the company.

Yeah.

What I can do as a VC is help a really great product, technical founder who has cracked that nut figure out how do I take this thing to market? What should my VP of sales look like? How do I think about channel partnerships? How do I price it? We're doing an AI pricing event in our office in a few weeks, which I'm super excited about. But it's a bit of a side note. But like in general, I as a VC am much more focused on what is the value your thing is creating than necessarily how fast is your revenue growing.

Yeah, it's such an interesting conversation. I want to react to a couple of things you talked about, which is the risk, right? You try it, it hallucinates, it made a mistake and we just, especially on the law firm side, our tolerance for that is zero. I think on the in house side again, their risk tolerance is much higher. What could possibly go wrong is a lot less. They're dealing with internal issues. And so the ability to turn on all sorts of LLMs and AI tools internally, they're going to adopt faster, they're going to see value faster. Because if you're giving advice on can I hire this intern? And it's fine again, what's like the worst that can happen, like really bad things could happen for a firm. And so we're going to be much more careful, you know. And I spend a lot of time these days talking to VCs and investors, just trying to help them get educated on this space. The speed at which development is happening in the AI space is also just overwhelming. So if you see a demo and I've had this experience where you see a demo three months ago and you're like, that's cute, that's not going anywhere. And then all of a sudden three months later you hear everyone talking about them and you're like, did I miss something? Let me see it again. And it's a whole different product and team and they're just moving so fast. But again, anyone who's evaluating tools in this space would be crazy to bet on one. And to say I'm going to sign up for a five year term because it you can't. You've got to hedge your bets and try a bunch of things. And if something like you said, doesn't work out great, what's the renewal rate, what's the retention versus six months from now? I just rip it out and put a brand new one in overnight because now that one's the one that's better today and it's going to be a different one that's better in six months.

This is where it comes down to that proprietary data point I mentioned before, which is if you've got a bunch of products that are all doing the same thing and one of them integrates a new foundational model faster than the other, then that demo is going to be much more impressive. But the other one may then update their model three months later and then that's just as impressive. So it's kind of this cat and mouse game. I think the way for an application layer company to get around that cat and mouse game is to grab and make use of this proprietary data. So back to the CLM example. If you're building a CLM tool and you have a bunch of data on when you use this clause in this specific context, the contract closes 10% faster. That is data that lives within your walls and you could capture and make it possible to surface to the next user who's in that situation to say, hey, you should actually use this clause. And when you do that, it becomes harder for a new entrant, even if they've updated to the latest foundational model, to meaningfully outpace you. And also there's a much bigger downside to ripping you out. So if you're good, when you've deployed this type of thing broadly, even if someone comes along with something that looks flashier because they've integrated the latest models, if this thing has been trained on a bunch of your associates, very specific traits and behaviors, you're going to be more loathed to rip that out because it feels like there's something in your. IT almost in the software.

Yeah. I missed CLOCK this year for the first time since its inception, but I think it was 10 years of clock.

It went on without you.

It did, which I'm proud of. It should. But one of the things I heard is that someone went up on a stage and said, clm is dead because AI agents will take over. And I know that's a topic just SaaS generally, that a lot of people are talking about. Is it the end of SaaS are AI agents. The future would love your perspective on that and maybe tell the audience what AI agents are to start with.

Sure. They're not perfectly well defined. Different people define them differently. I'll tell you, like, the simplest way I think about them, which is it's AI that takes action for you.

Yep.

So if the AI that most of us are using today is AI that answers your questions, AI agents are AI that take action for you, that could be sending an email on your behalf, that could be doing some sort of booking on your behalf, that could be writing some document on your behalf, et cetera, but you're basically giving the agent agency to take some action. And it feels very likely this is the direction that AI is headed, that this is the direction AI in the enterprise will head. There are lots of questions and risks around how to put the right guardrails around this stuff, because if you give it agency, it could obviously do bad things. And so one of the biggest challenges right now in the underlying technology layer is how do you actually put boundaries around these things to make sure that they don't break stuff? But the promise is crazy high. And it's really exciting. But it does beg this question that you alluded to, which is, if we get to a place where I can just pull up a natural language interface and say, build me a CLM that's specific to Coca Cola, and has all specific things to me, why would I buy that versus ironclad? Like why, like why wouldn't I just vibe code my way into my custom CLM for me?

Yeah, exactly.

We're big believers in vibe codes. Like we, we are big investors in bold new and we've seen a bunch of growth and these things. So there's definitely a use case for these. But I still don't believe, we still don't believe that this is what's going to dominate, particularly at the enterprise software level. And putting our money where our mouth is. We just raised a billion dollar fund to invest in enterprise software vendors. And so obviously I still believe that enterprise software vendors will exist even in the age of AI agents. And there's really three reasons why I think they will. The first is when you're buying software from a vendor, you aren't just buying the code. You are buying an opinionated perspective on how to solve a problem. You're buying a package from someone who has spent. People have spent their lives dedicated to solving this very specific problem. And so they have a lot of perspectives on the right way to do it that is very valuable. And that's a big reason why, you know, that people buy software vendors today and not just build themselves. But there are two other reasons why I think it's unlikely that software vendors will go away. The second is once you build it, you need something or someone to maintain it. And to your point, before the technology is evolving so quickly that even if you build an amazing CLM that's like world class in on May 9th of 2025, it's out of date on June 9th of 2025 because the underlying models have changed and the architecture has changed, et cetera. By very virtue of the speed of this technology development, it actually puts more onus on having an entity own the kind of updates of this thing. The third point I would make is people buy software from vendors because they're looking for a throat to choke. They're looking for someone to blame, someone to fix something when it breaks, to say, and this is separate from maintenance of keeping it, you know, on the latest technology. This is just, hey, I tried to do this query in the CLM and it didn't work. Fix it.

Yeah.

And telling an AI to go fix it, there's no guarantee that it's going to work. Having a human who's guaranteeing an organization, a vendor who's saying I will fix this for you is really important, particularly as you get to higher value products. I think super down market when you're selling like an SMB thing where it doesn't really matter as much if it breaks, so what? But if you're selling a CLM to Coca Cola, that is mission critical software. And so Coca Cola wants a human being on the other side of that to say, go fix this and to guarantee it. Which is I think, actually why back to this kind of shift towards outcomes based pricing. We talked about law firms moving towards outcomes based pricing. Software companies are likely to move in this direction as well because the big part of their value, if there's less value coming from the writing of the code, because that becomes more commoditized, there's more value around guaranteeing the code works. And I think that becomes an increasingly important differentiator in a world where people can build applications very easily. So the one that's actually backed by a vendor that is willing to put their name and potentially a financial guarantee behind, hey, I promise this CLM is going to work. I think that becomes an increasingly important differentiator in the age of AI.

Do you think that's different than if it is like a legal AI agent that is going to put contracts together or whatever. Related again, this whole like death to the clm. Because you could still go back to that vendor and say your AI agent messed up and please fix it. Does that change things?

I think the fact that the ironclads of the world are billing agents in will be better for both the vendor and the customer. But the whole point of it is it's wrapped around the ironclad brands and the brand guarantee. And so you can go to them and say, hey, this agent broke or teach me how to use the agent or fix this thing and there's a person who you trust who okay, I will fix it for you.

Yep.

And that is different than if you as Coca Cola spin up your own agents and try to make it work.

Yeah. I still think there's value in the whole workflow as well.

That of course, I think that's also something. It's such a good point, Mary. I think like in our delirium about the power of AI, our thinking has become too binary. There's like workflow software, which is like old and pre AI and then there's AI, right. Which like in that world, workflow doesn't matter. That's not true. Because a big part of ironclad value is the fact that it can help you coordinate your whole organization between HR and legal and sales and whatever through a very complicated workflow and permissioning thing. And like that, as long as there are all those people in the organization, as long as you want their input on an important contract, you need workflow.

Yeah, yeah. So as I think about it, we have all these tools that are coming at us now. Some are really specific AI that relates to like due diligence or patents or litigation and they're probably really good at that. Like one specific use case. Then you have this whole general legal AI that you know you can interact with in just a very human way. Then you also have traditional B2B legal tech SaaS that has AI in it, which I think is probably super powerful at least today. And then at the other end of the spectrum is like the OpenAI's, the Microsoft copilots, the Geminis of the world that are just maybe one day since they're sitting in all our stuff, we'll get good enough. I think nobody knows which direction all of this will go and where we should put all our bets. Do you have a perspective on kind of that huge spectrum and where this is?

I think the framework I would use to answer that question is the Jobs to be Done framework, which is what is the job that you're trying to do right now and what is required to do that job? Well, love that there are some jobs where AI and just AI, an AI query engine may actually be all you need. If you're doing a deep research task and you really just need to understand deeply the way that this law has been applied in a bunch of different contexts and what's the precedent there, then that may be enough to have either a legal specific search engine or even perhaps more of a general search engine like an OpenAI. If your job to be done by definition requires the coordination, collaboration of lots of different people. We talked about the CLM example before, but there are a bunch of others. Obviously in the law then probably most effective to have a workflow tool that has AI agents embedded in it. Which gets back to the earlier point. I think in our excitement and in some cases hysteria around AI, we've sort of lost the plot a little bit that there's still organizations that need to make use of this tooling. And I think that will continue for a long period of time, particularly in a high value field like law, where clients are willing to pay for it to be 100% right. And as long as there's still a lot of people involved in the lot, I think there will be because of clients desire to make sure everything is right. My bet is that this combination of kind of workflow and AI agents will probably be the predominant factor, form factor for kind of the legal situation. But I think within legal you have to think about what's the job to be done to your point. And I think like even within like specific tasks, if you think about something like ip, which you mentioned, there's a bunch of cool stuff happening in IP right now. Some of it wasn't possible before AI, some of it is replacing legacy workflow tools.

Right.

There's a whole concept around patent discovery where you can mine your company's black channels and zoom channels and email, et cetera to identify using AI stuff that is patentable. That's pretty cool. That wasn't possible before. Like the idea that you would have an always on thing that's looking for patentable ideas and then you can draft the patents using AI obviously and then it can go into a patent management system and there's a bunch of legacy old patent management systems that look more like workflow tools where you have approvals and document storage and everything else that have existed for a long time. I think one of the questions for the next gen startups that are starting here in patent discovery is will they be able to move over and take over the workflow tool? But it's a different form factor for each. So again, using this jobs be done framework I think is the right lens.

I love that lens. And also you're surfacing jobs that we didn't even know we had or like we didn't even know that was a problem to be able to proactively find ideas for patents. And that's a great example. The other thing I like about that framework is the AI is so general right now. You get these tools that it can do anything, it can do everything. We should roll it out to all the lawyers. That's overwhelming. It's too much because it's like what should you use it for? What is the job to be done? Everything too wide. That's why I think you're right. So much of what I look at across this firm or even when I was running legal departments or at Ironclad, it's the workflow orchestration is like 90% of it. And then yes. Wouldn't it be cool if we could tack on some AI to do the rest? So I think that's the right way to think about it.

We're starry eyed with AI for like a reason because it is, it's absolutely insanely cool. And also think about it in the context of how it helps get your job done.

Yeah, yeah.

In some Cases all you need is chat if it's just a research interface, but in a lot of cases you need more complex workflow for whatever the thing you're trying to do is. And you need like good old fashioned great SaaS tools and people and architecture to do that.

Well, that's right. Okay, so the other thing I wanted to ask you about, and you touched on this with why SaaS is important and why you want to find that team that you can have the one throat to choke. How when you're looking at tech, whether it's as a investor or even as a buyer, how are you evaluating that? How much of the decision, at least in the early stages, the founder, like what signals matter? And I follow on questions which maybe I'll just put into one big question. Right. So there's so many companies now that we look at that are two to five people companies. And 10 years ago you'd be like, no way, you guys aren't ready. You'll never pass the security, you'll never be able to develop fast enough for a company like us. But of course now every AI company has only been around for a few months or it's a small team and you hear about billion dollar teams that are five people. Big question, how are you evaluating? What are you looking for? Team matter.

Yeah, and I think this is true. Like this is true for a vc, but it's also true for a potential buyer within a long firm or GC or whatever.

Yeah.

I would offer two heuristics when you're thinking about this type of thing. Like when you're trying to evaluate founders at the early stage, particularly in this kind of AI world. The first is a little less AI specific and the second is more AI specific. The first one is founder problem authenticity. How authentic is the founder to this specific problem? And that could be because they themselves faced this problem when they were in a previous life. Some of my most successful investments thus far have been investing in someone who was at Stripe and they had this support issue they had to fix since they looked around to find tools they could buy to fix it, couldn't find anything, so they built it themselves. And then they were like, you know what? This is the other problem. Companies have the same problem. I'm going to spin this out and actually sell this to other companies. This company called Assembled that had that story and there's a bunch of others. There's just an authenticity that can come from having lived. Like yes, there's almost nothing better than if you were your target user and you Know that there's nothing on the market that can solve your problem because you would have much rather, yes, bought the thing versus having to quit and start held it yourself.

Exactly.

I love when a founder has that profile. They can also have that same thing, not necessarily by being the buyer themselves, but being involved in the industry. Obviously Jason Ironclad is a lawyer, was a lawyer and so had more perspective on CLM than the average person. In his case, he teamed up with Kai, who was obviously a technical genius. And that kind of concept of you can find teams that have this pairing of founder kind of problem authenticity and incredible technical prowess. That's a really green signal. And then the second category that I was referring to that's more AI specific is the pace of the team. And this kind of gets back to the fact that the technology is evolving so quickly. It was always important that teams act quickly, move quickly when they're, particularly when they're younger startups. Now it's almost an order of magnitude more importance because it's not just your code base that you're trying to update. If you know a new version of Claude gets pushed and you haven't updated your stuff to incorporate that latest model, you're out of day in an hour.

Yeah.

Whereas before, like, because foundational models are themselves evolving so quickly and applications are in large part derivative of the foundational model advances, you are now required to be super on top of what's happening across that landscape, open source, closed source, et cetera, and what's best for your specific job to be done. So you just need an insane level of iteration, speed, et cetera. So practically, what does that mean? If you're a buyer evaluating potential early stage startup, you can look at the team dynamic that I described before. You can also look at their product release history. You can ask them for it, show me what are all the features you built, what has changed and when did they all ship. And if you see something that's oh wow, these folks are releasing incredible new features every month, every two weeks, like this is crazy, then that tells you that they're probably on the cutting edge of things. If it's slower, obviously that sends a different message. The caveat, of course, to that example is particularly in law, you need to make sure that the thing that you have works well. And if it doesn't work well, it's not like move fast and break things type environment because it's a 100% accuracy type situation. So you need to make sure that the team is marrying this kind of incredible pace with Ensuring that whenever they're making an update that the thing is bulletproof.

Yeah. Such an interesting observation. So I remember back, I have known tons of legal tech founders over the years. Back in the sort of earlier days, even the Jason and Kai days, almost all the founders you met had this kind of story. Right. I was in a dark discovery room, like, why am I here all weekend? I could write something that could make this better. That is how so many of the earlier legal tech companies came to be. I hear less of that now. I actually hear, not infrequently, we were just going to do general legal or general AI. And then we found out there was such great TAM in this market and there was such a need and it makes such a great use case. And so I'm in there, I find it. I get flooded with calls from these types of founders who would love to learn from us. And so they want our investment of time to tell them how MA works and how diligence works and how patents work. And I think there's, to your point, a lot of learning that needs to happen. And so they need a lot of design partners.

One other thing I would say in that, Mary, it's not just the design partner need, it's also. Startups suck. You get beat up in the best startup. Yeah. The first investment that I led Diligence on when I joined the firm was Zoom Zoom beat the quarter ahead of the quarters until they went public. And even after that was an incredible company. Still is. And they got punched in the face a million times. Even on the way up in startup land, you get punched in the face constantly. And I think it's. I have noticed that it's much easier for the founder to bounce back from being punched in the face if they have real authenticity about the problem they're trying to solve.

Yeah.

If you came about solving this problem intellectually because you're like, oh, this could be a cool use case, then, like, if you get beat up a bunch, you don't have like the gut need to be like, no, I know this is wrong in the world. I'm going to come back and I'm going to fix it again.

I believe in it.

I believe in it.

I understand that we need this. And you just, you'll see, right?

You're just much more likely to be like, oh, man, this sucks. Let me pivot to another use. Case law didn't work, so I'm going to go to construction.

Interesting. Interesting. I also think, and I'm. I bet you have to play a big Role in this as an investor in these companies. The founder life. Wow. It's gotten really hard. You know when it was just coding and you were up all night, but you knew that you couldn't release something until two weeks and you're all up and then meanwhile you're slowly hiring and thinking about the next step and like bring on new customers now. Fast and furious. Your company hasn't been around and you are raising and you are building and you're trying to keep up with the third party systems and you're trying to learn the customers and bring on an onboard and it's. I feel, I feel stressed out every time I even talk to them. The life sounds so. It's so much pressure and you can't even sleep because like you said, like your hours, you're talking in hours now, not just like weeks.

Yeah, it is. I like, I have more concern for founder mental health than I ever have.

Yeah.

Because of these dynamics. Which is why, like you really, really want to feel like a deep need in your bones to change the world around whatever problem you're facing. Because it needs to be more than, certainly more than ego, more than desire to start a company or be a CEO. Certainly more than money.

Yeah.

Like you need to like really believe that the world needs to look different and that you have a unique insight on how to do it because otherwise it's just not worth the misery.

Totally. It's painful. It's really hard. You sacrifice a lot along the way.

My parents are co founders and so I've been like along the journey with them across a number of companies and I've so I've seen this obviously very personally and I so deeply admire them on so many levels. My brother actually is working with them on this most recent company and it's really cool to see. And they really believe they're building an iPad based wine menu. Their passion about wine. And they think that the way that humans buy wine, people buy wine is stupid because they don't really know in restaurants like what they should be choosing because it's just a static menu. They have this thing called venue that allows people to know what they're eating and what the restaurant has and do the smart pairing, whatever. I'm not passionate. It makes a lot of sense. I personally don't have a lot of passion for wine, but they do. And so they like really believe that the world needs this and so are willing to put in the hours to make this thing a reality. And I deeply admire them for it.

Yeah. Oh, passion matters. People Used to always ask me like, why do you spend so much of your time doing all this stuff for the community and for the ecosystem? And I'm like, I feel a duty, like, I feel like I have to. We can't stop now. Totally. So just generally, as an Investor Looking at B2B, are there trends, do you think things are going to go look very different and in what time frame?

Yeah, so we talked about this a little bit before, but I think the next big wave in B2B will be the incorporation of these agents into the software. So it going again from AI being a tool that helps you search or gives you advice or recommendations to a tool that takes action. Again, that leap is a scary one. There's a lot of things that can and will go wrong when that happens, but it's clearly hugely valuable. Again, think about with the job to be done lens. If there's a portion of the authoring of some document that you can do by outsourcing to an agent and have the agent do its own research, pull from a bunch of different resources and then come back with a synthesized answer that would be super powerful in a bunch of different legal contexts, I believe you have to trust it that it's pulling from the right research and that it's not hallucinating and that all that kind of stuff. And that I think is the direction things will head across a bunch of different B2B software applications. We're going to see a lot more browser control. Like you're going to see the application will take control of your computer and actually do the thing for you, which I think is going to feel really weird. So almost think about like autopilot for your computer.

Yeah. Self driving.

So you're sitting at your computer. Yeah. Self driving. So you're like, I want to, I want to do this expense report. And you say, go do it. And then you all of a sudden you see your mouth start to move and it's moving between the various browser pages of Concur and Uber and where else you pull from. And it's just like you're watching it do the work you used to do.

Yeah, it's like early macros of the 90s.

It is, it's like a little bit like Visual Basic if people remember that. Yeah, exactly. It's obviously going to be much, much more powerful because this thing is connected to the web and so it could do anything. So that's the direction things are headed against how they'll be deployed, how they'll be priced. All that stuff is like the stuff on the horizon that we'll have to figure out. But again, what I want to anchor everyone, prospective founders, existing founders, buyers of the software around is really focus on the job that you're hiring this piece of software to do. Do not focus on how cool the tech is.

Yeah, yeah.

And like the agent stuff is going to be gonna make people starry eyed when they see the autopilot take over your computer. And it's gonna be like, whoa, that's insane. And that might be the right thing for a specific job to be done, but it may not be and the risk reward might not be there, et cetera. And so your job as a investor, whatever is to evaluate your job to be done and think about what are the, what's my risk tolerance? What are the best form factor to try to solve. And then back to your earlier question when you're evaluating, does this thing work? How do you sort through all the various outcomes or offerings on the market today? Talk to other people to understand did it work for them? I think like one other thing that might be an issue right now is if you're at a law firm and you see some other law firm signs a big contract with AI software provider X, your reaction is probably to go, oh God, we need to go talk to that software provider X and buy it. That's not the right reaction. The right reaction is to call your friends at that company.

Yeah, exactly.

And say what happened? Did you deploy it? What deployment look like? Have you started using it? Where is it useful? Where is it not useful? And then come back with that data and say, okay, what are my needs? What are my jobs to be done? How do they differ from law firm X? I'm law firm Y, I'm a different firm, I have different needs, whatever. And this may or may not be a good fit for me. But like the kind of lemmings purchasing behavior is not healthy.

Yeah. How fast do you think this agent stuff is going to happen? Like how soon will that be?

Here I was at like an OpenAI private product roadmap thing last two weeks ago and I think they're forecasting like by the end of the year that this will start to be more of a reality. But I think like things take generally this technology is taking longer, it takes less time to actually develop the technology and it's taking longer to roll out and solve specific jobs to be done than people would expect. And realistically like I think you'll see early adoption of this stuff in 26 and I think that you'll Start to see scaling adoption of it in 27, which happens to be when feels like there's some emerging consensus around AGI, which is a whole nother thing. That could be very crazy. If I were a betting man, and I guess technically I am, since I've been educated better in my job, I would say this starts to scale in 27, but we're going to be making investments in this now because the startups are going to get kicked off and start contesting.

Such an interesting space to be in. So a final selfish question for myself. So my job here is about transformation. I oversee all the jobs to be done, whether it's legal work, I have hr, I have finance, I have marketing, I have it, like you name it. And there's many jobs to be done across the board. And as we think about upgrading our tech stack, there's the question of do you go with trusted incumbents? Do you go with kind of forward thinking, new, smaller upstarts? Do you kind of wait? Because we want to leapfrog and wait for agents or people who are starting to think about what the agents will look like for someone like me, what advice would you give on, you know, we want to be ahead, we want to be the firm of the future. But like how do we do that mitigating our risk too there the easy.

Answer to that, and then I'll give a better answer is test stuff. But you ring fence it. So you try stuff with a specific team and if it works, and if it works, then you expand it more broadly like that. That I think is a good place to start with any of these pieces of technology. And I think the more nuanced answer is it does depend on each function and what the job to be done is for that function. There are going to be some situations where the incumbent piece of software is doing a nice job integrating the latest AI and so you don't necessarily need to change. There are going to be some where there's just a whole new architecture that is taking place in that function and the existing products just aren't going to, they can't do the architectural change necessary. So one example in kind of the more sales and marketing side of things is historically marketing tooling, which is like a lot of like data around leads and who's downloaded your white paper and who's fitted your website and all that kind of stuff sits over here. And then there's a bunch of like sales tooling that sits over here that actually does like outbound emails based upon what you know about the customer the prospect, et cetera. And they're two separate sales stacks. You've got like sales loft and outreach on the sales line of things and then marketing. You've got zoom info and a whole bunch of others that do this. It is becoming increasingly clear that like the tech stack of the future will unify both of those things. Like they shouldn't be divided. It makes a lot more sense to have them connected so that the sales outbounding tool is actually informed dynamically by all the things that are happening. So if someone's on your website and they click on this thing, then the email you send can be like, hey, I saw that you had interest in X, Y, Z. Do you want to talk to Mary about this product we sell or the service or whatever? Yeah, we invest in a company that is called Unifi intentionally because it connects us to, that is focused on providing that service to corporations. So I use that example not just to tout unify but also to give an example of where like the legacy architecture may require a new entrance because like it wouldn't be enough for each of these players to add AI because it's. You need to kind of the thesis is you need to connect across both, but that may look to different domains. In finance, there's a bunch of interesting stuff that's popping up right now to help CFOs do better forecasting and all that kind of thing. So you may see some next generation anaplan adaptive insights that come out that are pretty interesting and obviously AI native. So back to my lazy answer is I think it does make sense to test a variety of these things, but I would do so using all the frameworks we talked about today. So understand if it's very real estate company in their founder problem, authenticity, how quickly are they moving? Is there someone who understands the domain plus a technology expert? Try to make sure that you've got comfort there. Talk to your friends to see like did this work over here? So like you're building that kind of thing. And then I guess step three would be like test it in a specific area and see if it works and then expand it or not based upon.

Awesome. So helpful. Jake, this was so fun. I feel like I could talk to you for another hour but unfortunately I'm sure you have to run somewhere. So thanks so much for your time. It was great to catch up. This way.

You're so welcome, Mary. This was so fun. Thank you.

All right everyone, thanks so much for listening. We'll see you next time.