Turning Legal Services to APIs | Jay Madheswaran, CEO of Eve

Jay Madheswaran is the cofounder and CEO of Eve, a legal AI platform for plaintiff law firms. They recently raised their $47M Series A from Andreessen Horowitz, Lightspeed, and Menlo Ventures. He was previously a partner at Lightspeed and the first engineer at Rubrik.

Jay's favorite book: The Truth Detector (Author: Jack Schafer)

(00:01) Introduction
(00:44) Overview of Legal AI and Industry Impact
(03:53) Daily Operations in Plaintiff Law Firms
(05:49) Identifying and Launching Eve's MVP
(08:58) Framework for Building an Effective MVP
(12:02) Acquiring Early Customers (Zero to Ten)
(14:20) Scaling Beyond Early Customers: Growth Strategies
(16:08) Encouraging Word-of-Mouth and Inbound Growth
(18:21) Product Development and Customer Feedback Loops
(20:27) Eve's Technology Stack and Internal AI Usage
(22:16) Team Structure and Leadership Development
(24:20) Role and Impact of Designers in Early Startups
(27:15) Future Trends in Legal AI: Consolidation vs. Specialization
(30:29) Exciting AI Advancements Relevant to Eve
(31:58) Rapid Fire Questions

--------
Where to find Jay Madheswaran: 

LinkedIn: https://www.linkedin.com/in/jayanth1/

--------
Where to find Prateek Joshi: 

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 
X: https://x.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.494)
Welcome back to the podcast.

Jay (00:05.262)
Prateek, thanks so much for having me on again. It's good to be a repeat contender here.

Prateek Joshi (00:11.334)
Yes, one of the very, very few ones. Now north of 160 episodes. So yeah, I'm really glad we were able to do this. It's been a couple of years and congratulations on all the progress and momentum that you've built with Eve. So I want to start with the basics. So I know legal AI is pretty broad and there's so many things, so many different pieces to the puzzle. So what are...

all the different types of legal work that AI can do.

Jay (00:44.78)
Yeah, I mean, this is one of the exciting things about where we are in the legal industry in particular, Pratik, because people don't realize just how in depth the legal industry is going to be impacted by this technology. And the reason for that is like, fundamentally, if you really summarize, I know a lot of your audience is kind of engineering business focused. So if you really summarize like the legal, what it's about is it's an unstructured data transformation problem.

So you have large amounts of text and you have to apply human intelligence and techniques to transform that text into different forms of other text, documents, court filings, contracts, agreements, et cetera. As a result, different industries have formed around this. And actually, don't know people might know about it, but even the promotion path of...

Lawyers is kind of indirectly mapped to like how much tech they deal with, right? Earlier on in your life, you're dealing with a lot more kind of grunt work and you have to put in the hours to kind of move up the ladder. And that's just how the entire space works. Right? So what does, what does this mean? It means that the legal AI space or the legal space in general is largely tied to human labor. And as a result, firms have come up with different forms of handling that, you you have.

defense firms, and these typically represent companies and handle complexities companies have, ranging from simple things you could imagine, like reviewing agreements, contracts, like sales contracts, to different types of litigation, like IP litigation or business contract litigation, et cetera. And then you have kind of the plaintiff legal landscape, which is what EVE services.

And plaintiffs represent people like you and I, right? We don't have the money to pay a couple hundred dollars an hour for a lawyer. So plaintiff law firms are merged to try to align their incentives with yours. And they work on a contingency model and they deal with representing and advocating for you to the best of ability from end to end. Right. And then lastly, you kind of have legal research like these task specific components that have kind of emerged. So you have players that do.

Jay (03:10.05)
very specifically things like contract review, transactional document analysis, right, and help. And you have legal research providers that help with the legal research component of it. And that's just kind of how the industry is shaped up. But what we do for the case, we help the world's best plaintiff firms really represent and advocate for the clients using technology and in a whole different way than before.

Prateek Joshi (03:38.73)
That's a great summary. Now, when you talk about a plaintiff law firm, even without Eve in the picture, just their average day, like what are the things they have to do on an average day?

Jay (03:53.582)
So plaintiffs, let's take labor and employment. So if you're in a labor and employment firm, typically you're dealing with things like wrongful termination, sexual harassment cases, or things along those lines. And these things are called claims. So a lot of these claims are actually, there are some things you're allowed to fire people for and there are some things you're not allowed to fire people for. claims largely represent litigatable

reasons of wrongdoing. And as a result, some of the challenges you have as a plaintiff firm is you're potentially taking some of our firms deal with 300 unique calls with clients before they are able to take on even one case. So why is that? It's because taking on a case is very expensive for the firm. So they have to make at least $5,000. Some might have to make

even $50,000 per case based on their particular business model. As a result, a lot of work, sometimes 10 to 40 hours of work, goes into really qualifying a client to understand whether or not is this a real lawsuit, right? Is this a valuable lawsuit? Like doing that work requires a significant portion of time before you're even able to represent the client. So that's like the very beginning stages of what Eve helps with. But what's really cool and unique about us is

Once you do that work on the platform, learns about that case and that client and how your firm operates. And then the same type of learning is easily applied through the entire litigation process, going towards complaints, which are when lawsuits are filed, to discovery, which is when you're learning more things about the case from the other party or you're giving information back to settlement when you're actually resolving the case in some way.

Prateek Joshi (05:49.686)
All right, now I want to talk about the launch of I want to take you back to right before the launch. And I know the company has been around for a couple of years, but this use case was not identified. So before the launch, how did you spot the need, or how did you identify the need? And also, how did you decide what goes into the MVP of Eve?

Jay (06:17.166)
Yeah, I mean, this is like the million dollar question, right? Maybe a billion dollar question. But this was hard Prateek because when we were starting off, what was this? This was like 2023 or something along those lines, maybe early to mid 2023. It was whenever GPT 3.5 and then 4 came out. But that really shook up law firms across the board, because any owner was able to just go try.

a chat GPT, right? And then they would ask it invariably some sort of general legal question, and then is able to spit out the answer in seconds. That's like a mind blowing moment for a lot of a lot of firms. As a result, we had no trouble kind of getting in front of really any law firm we wanted to, because everyone was looking for help understanding what this meant, and solutions to help make their own life better in terms of how they thought they could use it. Right. And this is likely what you know, Harvey, I'm sure

ran into and why they kind of specialized in defense side. but we were kind of hearing that from a lot of the defense firms. We probably talked to every one of the Amla 100 and we were hearing, similar things, but very different things on the plaintiff side. Right. And this was the first decision for us to make is like, what do we actually build? Cause the MVP for defense looked quite a bit different than MVP for plaintiffs where plaintiffs

They cared about representing more clients and getting through a larger volume. And they were not hurt by their labor going down, right? Versus defense, their charge billable hours model. And they care about, you know, making their work more valuable and making work that's non-billable more efficient. As a result, what was really exciting to us was in the plaintiff space, it touched every facet of how they operate way more than on the defense side. And that to us,

felt like a meaningfully different problem where we're actually changing how law firms are run, which is much harder to do on the defense side, given where the business models look like and where we were at back then.

Prateek Joshi (08:28.692)
Now, taking that a step further, and if we were to provide guidance to a younger founder on what framework can they use to decide what features should go into MVP? Because you can spend months, years just building and building, and then you launch it and nobody cares. So what can they use to decide, okay, this is the minimum number of things I need to include in this product and just ship?

Jay (08:58.678)
Yeah, I think this is where a good product discovery goes into play, right? So you have to actually talk to your customers and understand. And there's a lot of tactics for this. Some of the common tactics for new founders is to actually not try to sell your product, to instead just try to solve the problem for the customer. So what that means is you're actually encouraging them to use competitor solutions. You're like, well, why do you actually need us? You should go use X and Y and see if that solves your problem.

And really use that to drive to the understanding of what is the heart of what you're solving. What is the core pain point that is unmet by current solutions and current things? And is that a valuable enough problem? Are they going to pay for it? Because eventually, they have to pay the bills if you're an enterprise solution at the very minimum. And you're trying to do all of that upfront.

And then the second thing is I actually kind of, think one thing that's changing now is the rate of development, right? Like back in, back when I was, you know, we were starting Rubrik, it actually took a lot of engineering resources to build the MVP, right? Because it needed to be reliable. needed, like, you can only back up the data once. And if you don't do your job, you're, you're screwed. So you have to, you know, have to do.

implement the entire infrastructure to be able to do that before you can even sell and use it. I think that's changing a lot in especially vertical AI solutions, right? I think the time to develop a product is way slower than it used to be. And I think you should utilize that to your benefit by basically making up for common challenges in product, you know, product management like things where you're talking to lot of customers trying to make the best out of the decisions you can.

by creating a higher confidence behind those decisions by actually giving them a mini product they can play with. It's just kind of like how design evolved. Back in the day, you used to do product discovery with slides. And then it became product discovery slides and Figma images. And now I think you can go a step further and show a little bit of interaction patterns and actually show and get a bit of value back.

Jay (11:20.566)
And that drives you sooner to, they going to pay for it? Right. so I think that's probably like what I would change, you know, is to get even more analytical about solving their problem and leaving yourself out of it. Cause as founders and builders, you want to just build and you, your brain automatically ignores what else they could be doing, which is not great. So a way to get around this by basically giving them a competitor to use and like almost forcing them to use a competitor, to

to drive clarity to you in terms of what you need to build. then lastly, build it early. don't take this whole, it takes a long time to build an excuse, build something easy quickly and iterate ASAP.

Prateek Joshi (12:02.646)
Yeah, that's great framework. okay, so now the journey from zero to 10 customers is always looking back. I now you're bigger, so it look nice and easier, but at the time, always zero to 10 is hard. Nobody cares about you in that phase. So you have to make a case for you and your startup. So the journey from zero to 10 customers, what was the biggest challenge?

in acquiring and retaining those customers, right? So what happened in that phase?

Jay (12:39.296)
Yeah, I mean, we had a, again, like, this is what gave us confidence to double down in the plaintiff space. So we probably had like the Goldilocks golden scenario happen, which is one of our earliest customers, which was a large one. they ended up, you know, we, we gave them three months to adopt the product and the attorneys did not listen to us at all. And they just started spreading our login to every single member of the firm and they were fully on boarded in like three weeks. So we were, you know, back in the day, we were the first.

productionized cloud application. and we were running into like QPS limits on their side, cause they were still rolling it out slowly. and you know, we were, we kind of had to like be on fire continuously since then. Right. Because the product demand and how much it was able to do is so lopsided and how much we charge for versus the benefit they get that it's like a no brainer to adopt the moment they see it, you know, in, customer calls, like I.

I see our customers, new customers, you know, when we're demoing the product and showing them different challenges they have trouble with today. And you can see halfway through the thing, they start going, go Eve, go Eve, go Eve. And like, they start chanting when Eve starts generating a long form document. And so like the value is so obvious, right? That for us, the hard part has been how do we continue this rapid growth and still make good decisions in terms of, you engineering.

not taking on tech debt. Like now if you take on tech debt, it bites us like instantly. Right. So like doing things in a very fast growth way is really, really hard. And that's oftentimes what we deal with here.

Prateek Joshi (14:20.028)
And now that let's go to the phase where you have your early handful of customers, clearly something is starting to work. Now after that, you want to speed things up. You want to get more customers, expand faster. So in terms of product or sales and marketing, what are the things you tried? And what worked and what didn't as you went from the early handful of customers to the next phase of growth?

Jay (14:47.754)
Yeah, I think what we were initially trying to do is the typical stuff, Like lot of outreach to kind of get the visibility of the product out there. And what we underestimated was how much word of mouth growth actually happens when you have a good product. So when we had a good product and we had some early customers, they were actually spreading the product without us even knowing about it. You know, we would get forward to these random threads of internal discussions people have.

And that's how we got meetings on the, on the calendar. Right. So shifting from kind of this pure outbound into largely inbound motion was kind of really large for us. And as a result, like we just ended up putting more resources towards, you know, product development and making an amazing product that encourages more of this word of mouth growth and aligning our sales efforts towards like generating more inbound as well.

Prateek Joshi (15:45.974)
And if you look at the tools and efforts, now that inbound is working and you wanna do more of that. So what has worked in terms of making more people wanna spread the word for you? So is it content, blogging, encouraging and sending? What have you done to push that even further?

Jay (16:08.44)
So it's a few things, right? One is, think, just making, having a larger and larger value prop is pretty big. So not taking it for granted, but also staying up with the changes in technology. Today, AI changes every three months, which I'm sure you're very well aware of. We're hearing about diffusion LLMs out of nowhere, for God's sake. so things are changing so rapidly in this space, and not resting on our laurels is becoming more important than.

Prateek Joshi (16:27.082)
Yeah. Yeah. Yeah.

Jay (16:38.072)
You know, you have to reinvent yourself every three months and also tighten the life cycle between new tech that is mature to getting it in the hands of customers ASAP. That's becoming incredibly crucial. So we put a lot of effort in terms of how we architected the product, how we develop our own internal development life cycle, who we hire to make a lot of this happen where now, you know, when Sonnet comes out, we're able to hit it against eval and launch.

three, seven in relevant areas within two days of launch. You know, that's kind of what the level at which we have to operate to kind of stay on top of how good the product is. Right. And that is eventually what matters, right? Like really good products is what people end up talking about. and so I think that's, that's like the start. Second is like law firms are a little unique in the sense that, it's a, they're very used to referrals.

Right. So law firms themselves refer clients to others as a result. Like if you show that you're actually causing some real change to their business, right. Positive change their business. And you have multiple customers that are really well known in the space that have used you and seen success. It is a huge unlock in the space.

Prateek Joshi (17:57.502)
Now, going into the craft of product building. Now, in the last couple of years, clearly something is working and you shipped a great product, people love it. What goes into building the product, especially what have you put in place at EVE to make sure you continually ship great products?

Jay (18:21.718)
It's a lot, right? So it's basically tightening the feedback loop cycle from customers, new prospects to product development life cycle. So it's kind of keeping that process as tight and iterative as possible is something that's really benefited us. So a few things we've done is our team, we've really put a lot of effort into hiring the best engineering talent we can, right? Like they come from places like, know, OpenAI, Anthropic and...

other really well-known founders in the space we've kind of hired as engineers. Our talent is tense to be more senior and each person is able to handle large segments of the architecture themselves and push it forward. That's one component. Second component is legal is a unique space, right? Most lawyers don't know ins and outs of legal. So being able to tie that together with

people that we've hired that are experts in legal internally and our own customers and getting that feedback, organizing that feedback. And we do use LLMs for this, right? Like we use transcripts and actually analyze what are the common pain points? What are the things people are complaining about? And that gets fed into how we develop products. And it prioritizes different roadmap items based on what comes up. And if you think about it,

in normal development life cycle that would get delayed by three months. Right. So what would happen is you would go through a quarter of sales, then sales would produce report. Like, here's why we won deals. Here's why we lost deals. And then you would then take that feed it into some sort of product roadmap, but there's no reason it needs to be delayed by three months.

Prateek Joshi (20:06.986)
Now going into the technology stack that we've used to build the product. So as much as you can, can you walk us through the technology stack behind Eve? And also, part B maybe is within the company, where do you use AI internally for your own work?

Jay (20:27.414)
Yeah. I mean, in terms of the technology stack, it's kind of like the, like an advanced rag stack, right? It's kind of the way to think about it. But the thing that's unique on top of rag is we really, really care about hallucination rates being super duper low. We care about the UX UI patterns to be able to make it so that end to end time to produce a high quality, trustable document is, is low. Right. And what that means is there are some times we're getting to the 80 % and giving them UI feedback.

to get to the last 100 % is a good idea. And then sometimes we're putting a lot of ML resources behind getting it to 95 % accurate is a good idea. So we put a lot of time and effort into that. So that's kind of like, that's kind of it. And on the internal side, we use AI a lot, right? So our sales training enablement.

uses AI significantly, our sales outreach uses AI significantly. Our product development, like PRD development, uses AI a lot. Our engineering team is huge fans of, know, cursor. And in fact, I use cursor to help make small little apps to help our sales team out a lot. So we are huge developers and users of AI. And maybe one of things that's unique about what we're doing is, you know, I have a bit of an engineering background, but I also kind of run our go-to-market teams here.

And as a result, I'm able to use and actually develop sometimes applications and inject efficiency gains via technology in the sales process, which is actually not, it's surprisingly hard to do. And that's, think, something we're benefiting from for sure.

Prateek Joshi (22:16.862)
One of the things that keeps coming up, especially as you go from early to this growth phase, is structuring the teams. And there are so many other infinite ways to do this, but obviously some ways work better than others. So how have you structured the team? And also how many direct reports and what are standalone divisions versus what have emerged? So basically, what's the structure of EVE today?

Jay (22:44.396)
Yeah, I mean, this is one where I wish I had a good answer for you, right? Like given our fast rate of growth, we still have a very flat structure. think most people report to me directly right now. And most report to my co-founder on the engineering side. So very flat, but this is an area where I think it has to change as we continue to hire more. The other thing is we have natural kind of leaders emerge from the team that are able to

Prateek Joshi (22:55.254)
Okay.

Jay (23:14.44)
own various different parts of the team management or ownership over important components of the company. And I think this is one area we've really benefited from because almost every single individual in the company is turning into a leader in their own category. And that's helped us kind of keep up with the growth that we do see today. to answer your question, there's nothing too crazy we're doing, I think, that's different from

Anyone else, we still have the same organizational structures in our sales team, go-to-market team, product teams. The maybe one unique thing we are doing on the engineering side is we actually pair designers into smaller pods with engineers. And designers actually do a lot of product level thinking, you know, similar to our thing around how iteration cycles are not faster, right? Designers are able to do quite a bit of the work that you would expect.

from a product manager and maybe even a little bit of an engineer. And that really accelerates product development lifecycle.

Prateek Joshi (24:20.278)
It's amazing. think some of the edit really shows when a really well designed product, people want to look at beautiful products, they want to use them versus clunky. I think having good design aesthetic, both in terms of functionality and also just the plain look of it is very important. So maybe a quick sidebar on this question. The designer, when you hire a designer, obviously many different definitions. It's not just about picks like,

picking the right shade of blue on your website, it's not that. So when you interview designers to bring them onto your team in the early days, so what do you look for? What are all the things a designer should be able to do in a fast growing startup in the early days?

Jay (25:05.816)
So early days as an initial designer, you have a massive impact on the entire direction of the company in a way, right? Because early on, small things like usability and even visuals, make a huge difference. Because the more, it sounds silly, but the more beautiful it looks, your company looks a lot bigger than it seems, which can help with trust and just...

getting someone kind of excited to take the call, right? Even things like that, it affects across the board. And usability, to your point, goes beyond just how it looks. It's actually like how much sense does it make to actually use your product, right? Can someone figure it out without getting on a support call to use it? And that also affects metrics across the board, right? Because sometimes you start a pilot,

Prateek Joshi (25:37.728)
Right.

Jay (26:02.24)
And if your own salesperson doesn't know how to use a product and they're teaching your customer how to use a product, you kind of have a big problem on your hands. So that's like the immediate impact I think a designer has. The other impact oftentimes, especially in companies like us, is to be able to become the voice of the customer internally. As a designer, you oftentimes are having a lot more calls and relationships directly with the customer.

Oftentimes it's your most avid customers. So being able to translate and understand and really put yourself in their shoes and be able to relate that feedback down to engineering execution is massive, right? Because like, as I'm sure, like most people know, the reason engineering takes so long is you have to deal with a hundred different tiny little, you know, what happens if this type of questions, which if you do enough of them incorrectly, you end up creating an inferior product, right? And

the better you understand the customer, you're able to make a lot of decisions a lot more effectively. And sometimes engineers aren't in the best place to do that, but engineers combined designers, especially if you're early on, makes a massive difference there.

Prateek Joshi (27:15.668)
Right. Now, looking forward, legal AI, obviously, it's picking up big time and there's so many different areas. You mentioned like plaintiff, there's defense, there's transactional law. Now, if you look at how it's going to evolve in the next two, three, five years, do you see more consolidation or do you see more specialist legal AI companies emerge to service different parts of the legal world?

Jay (27:46.55)
I anticipate there to be probably a bit more consolidation. Because what we're seeing is we're actually handling the case's end to end. And given that and given the quick traction we're seeing, we're able to hire talent that are able to even pump out highly accurate individual steps into the overall product very quickly. So then if you.

If you think about where that goes, it's effectively that it is possible that the more data you kind of can pull together on a case for a firm and are able to handle and organize that structure, structure that data for the purpose of rag and other things, really, really effectively, you are actually able to increase accuracy on every single individual thing. Right. So from that perspective, like I expect there to be quite heavy consolidation on the plaintiff side. but legal is complex.

And I think what's going to happen with the entire legal industry is much hard to say, because there's all sorts of things that law firms are doing. And I think some areas in which you could go, which is kind of exciting to me, is there are some firms that are using Eve to really kind of reimagine how law firms are built. So instead of relying on the best you could do to improve your goods and services as a law firm is to hire a better attorney.

Right? Like that improves your capability and litigation. And maybe if you're the owner, you build up some of the skillset over the many years you spent litigating. But now with things like you can teach Eve to become the best litigator in your firm. Right? And you're up-leveling the entire firm across the board just by the mere presence of using it. What that means really, if you think about it, is now for the first time you have a way to consistently make

the good, is your legal services continuously better, and your cost margins actually change towards taking up 70%, 80 % of your revenues on labor. Now it's looking at 30%. So if you play all this out, one of the ways to go is some law firms could start specializing to almost turn some forms of legal services into almost like APIs, or like a consistent product that gets better and better, which is kind of what happened with the tech industry.

Jay (30:11.8)
you if you think about things that eventually became software, that's what they managed to do. And Eve is going to become kind of their building blocks, just like how AWS powers a lot of building blocks and legal tech. Eve is going to power a lot of how legal law firms are actually built and run.

Prateek Joshi (30:29.888)
Amazing. I have one final question before we go to the rapid fire round. Now, so many advancements happening in AI. What AI advancements are the most exciting to you as it pertains to Eve?

Jay (30:45.344)
Yeah, this is a good question, right? So I think there's a few things that are really, really exciting. I think this whole reasoning models is going to be game changing over the long term for Eve and legal industry in specific. Because there's a lot of how you reason about and take actions on data that actually matter a lot in attaining good outcomes for your clients.

And the better we can get control over steering that reasoning, we can actually make a significantly better product that starts improving the efficiency and capabilities of attorneys across the board. So I'm really excited about that. I think voice to voice capabilities or voice agents, when they become more mature and a bit cheaper, are going to be game changing. Because as I mentioned, people take 300 calls sometimes to qualify a client, right?

If you're able to inject this legal intelligence that we're building into those calls, you can change up workflows quite significantly, you if you think about it from the first principles. So I think that when they get more mature, I'm quite excited about those two.

Prateek Joshi (31:58.646)
With that, we are at the rapid fire round. I will ask series of questions and would love to hear your answers in 15 seconds or less. You ready? Alright, question number one. What's your favorite book?

Jay (32:07.041)
I'm ready.

Jay (32:12.134)
I have many, I already gave you one before, so I'll pick a different one. I, I've enjoyed the truth detector by Jack Schaeffer. if you guys have read it, it's about like an FBI agent that, tells you how to get the truth out of things and people. And as you can imagine, it's generally useful in life, but really useful. found in sales.

Prateek Joshi (32:35.034)
That's fantastic. Yes, I think that's actually a very, very good point. mean, sales, most people are just inclined to be nice and anything but tell you the real truth, which is like, I'm not going to buy you, not this quarter. Like basically, I think it's just, very, very interesting. All right. Next question. Next question. Which historical figure do you admire the most?

Jay (32:49.304)
That's exactly right.

Jay (32:56.364)
Ooh, that's a good question. You know, I don't know. think I've always liked Newton as a kid because just the amount of progress he's been able to make on humankind is just insane. And I think that's just incredible. eventually, if you think about it, probably gave birth to AI in some weird way. So I don't think too many people can really get that credit.

Prateek Joshi (33:26.44)
has been an important but overlooked AI trend in the last 12 months.

Jay (33:31.714)
I think people are underestimating that the models that we've already have have been getting continuously better. and this is just the beginning, right? Like people don't really realize that as the capabilities that get model even get 10 % better. There's a lot of different use cases that previously were just outside the reach of models and AI that suddenly become possible. so. You know, I think, I think last year was a little early for agents, AI agents.

I expect over the next like 18 months, we're likely going to start seeing like the very first signs of agents kind of doing things you would expect, you know, since we saw Star Trek, you know, things like that start becoming more and more possible.

Prateek Joshi (34:17.398)
What's the one thing about legal AI that most people don't get?

Jay (34:22.018)
people still don't really understand just how impactful Eve is going to be on the legal industry, right? Like the amount of physical human labor required in things that should just be done with technology is just ludicrous in the legal space. And the amount of time saving that Eve does, people are drastically underestimating, I think, the impacts it will have on US, right?

lot of people actually need to leave the representation. They can't get it because the case won't even get taken unless it nets at least $5,000 in fees to a lawyer. As a result, you can imagine how things have been getting more more inequitable over time.

Prateek Joshi (35:05.364)
What separates great AI products from the merely good ones?

Jay (35:09.55)
You got to drive results. You have to be able to see it, see actual change on your business by using said AI product. If you're not doing that, then you're kind of a gimmick, which I think a lot of products are right now.

Prateek Joshi (35:25.526)
What have you changed your mind on recently?

Jay (35:30.892)
I think I mentioned this a bit before about agents. I was really big pessimist on agents because from kind of a mathematical background and also doing AI for a long time before LLMs, probability, when you add a lot of steps together, you need really, really high levels of probability to do anything meaningful. So I was incredibly skeptical. But now I think

We're running into a lot of cases where I call it a probably distribution is good enough. There is actually a good enough outcome in using an AI agent, which I underestimated. So I think that's going to start changing. And I think I've changed my mind about that.

Prateek Joshi (36:11.893)
Yeah.

Prateek Joshi (36:16.438)
Yeah, that's a very fun part. If you've been doing AI for like a long time before, it's hard to unsee the cascading probabilities. that's my first question. How can you, I mean, if you multiply a bunch of like less than one, one point or numbers, yeah, it's just like a small number and either like 17 % accuracy, it's not good enough for me. I need like, definitely higher than that. So that's fantastic. All right. What's your wildest?

Jay (36:23.872)
Yeah.

Jay (36:29.9)
Yeah, even 0.9 by multiple times gets bad quickly. Yeah.

Prateek Joshi (36:45.032)
AI prediction for the next 12 months.

Jay (36:51.128)
That's a good one. man, this is tough. Wildest.

Jay (37:02.412)
I think we're going to taping out in terms of large LLM improvements. And I don't know if this is common consensus or not, but that's my prediction for I think the AI hype in terms of the next model being awesome is we're probably going to start seeing that die down this year is kind of my guess.

Prateek Joshi (37:24.884)
Right, right. It's gonna be priced in. like you can't, like, can't, yeah, like, people are gonna expect a certain amount, and if you just deliver that, it's gonna be, yeah, that's great. All right.

Jay (37:35.916)
OK, I will make one more prediction because that one might be too obvious. I think someone this year is going to figure out a use case where applying compute, thinking for a very, long time and spending potentially $10,000 on one run, is actually ROI positive. So I think we haven't seen that yet, but I think we're going to start seeing that this year.

Prateek Joshi (37:39.797)
today.

Prateek Joshi (38:00.768)
Final question, what's your number one advice to founders who are starting today?

Jay (38:07.866)
do it. I think that's, that's it. Like I've, talked to so many people that are, that are thinking and considering and, and, it's never. Timing has never been better. You know, if you, if you think about it, there's so much opportunity. Like you're talking about entire, like trillions of dollars of services industry being able to be touched by AI. And I think a lot of the boring spaces are actually really, really good to go after. Right. Like.

Logistics, transportation, medical is a really, really large field that many things don't actually have to do with even drugs, right? There's a lot of boring problems in medical that are really interesting. Finance is a large one and you know, your local plumbing supply store probably has a lot of interesting problems. so I think, I think like the real thing is just, just go do it, right? Like don't overthink it.

Prateek Joshi (39:04.054)
He's like, that's a fantastic way to end the episode. It's like, just do it. And that activity is going to produce so much information that you don't need to go ask people because that is so much data. Just do it and then use the data to do more things.

Jay (39:18.954)
Yeah. mean, think about it, Pratik. Like when we started, you know, even, you know, 10, 15 years ago, I tried doing startups in the past, right? And it was so different because you do it and then you were actually blocked. Like you didn't have a good enough network to go ask people for questions. You didn't know like what the conversion rates for, I don't know, B2C should look like in terms of how much you spend on marketing to how much you convert. And now you can just go ask these things to, AI. And it's not great, but...

way better than you in most things. And I think if you just start doing it, getting that data like you said, you're gonna make way more progress than you think.

Prateek Joshi (39:50.11)
Yeah

Prateek Joshi (39:57.428)
Yeah, no, 100%. Jay, this has been a fantastic episode. I loved the discussion and it was always fun to see how builders build. And I always try to share that knowledge with more people. And in this case, it was really nice to see the journey and obviously he was growing fast. So thank you so much for coming on to the show again.

Jay (40:17.912)
Thank you for having me Prateek.