Jason Boehmig, CEO of Ironclad on Balancing Risk, Innovation, and AI Opportunity in the Legal Field

Jason, thanks for sitting down with me. We met a couple of years ago. Now we both have shared investors in Excel. And I think what I've been so impressed by is your founding story and the success that you've had with Ironclad. Last year you passed 100 million in ARR. You guys are growing at a really strong clip and you guys are fusing AI into really every part of your product. And what I would say is going AI native. And I'm really excited to just sit down and talk to you about what that story and what that journey's been like over the last. What will be, you said on Thursday, 10 years?

Yeah, a couple of weeks.

Couple weeks. 10 years. Yeah. Crazy to start. I'd love to just hear from you about the founding story, how you started Ironclad, your journey, what the vision was.

So I was an attorney. I was actually an attorney for companies like Ironclad.

And I would just meet these founders and help them with financings, things like that. And I just started automating some of my own workflow. So the simplest thing was like incorporation docs, their certificate of incorporation. There's like founders stock purchase agreements, and you got to like fill in all these numbers and make sure that they populate amongst different documents. So I was actually in charge of the automation of incorporation for the firm. And then we started doing some other things like open sourcing financing documents. You may have heard of series seed documents, which are kind of like open source seed financing documents if you're doing preferred stock. And kind of this like theme of how can we standardize the practice of law, how can we automate? Some parts of it led me to start tinkering on nights and weekends. I hired a software engineer to tutor me for like 50 bucks an hour, and he would kind of like help me get unstuck on automation stuff that I was doing. I actually didn't start with the idea that I would be a startup founder. When I was doing this, I was like, I just want to be a more efficient attorney who uses technology like the state of the art in my practice of law. I actually really liked being an attorney, but what I realized is, like, no one was building for lawyers. It was so widely considered to be a horrible market that there was like no software companies that would even want to talk to you as a potential customer. So that's kind of why I had to.

Start doing it my own. And the companies that were doing it were like 20 plus years old on like really archaic stacks.

Yeah, I was going to ask like, did you look for software solutions initially?

Yeah. So that was my initial one was I just want to like use the best off the shelf stuff. When I found the off the shelf stuff wasn't good or even current, I started automating it. And eventually I realized I'm kind of the only person doing this specific thing which is trying to automate the practice of law and particularly around like my work, which was corporate law, contracts work, and quit my job with a bunch of student loan debt and started working on basic.

The continuation of the Basic automation was fortunate enough to randomly attend a lecture at Stanford on like a Tuesday afternoon. There was maybe a dozen people there and it was small enough where everyone went around the room and said what they were working on. And I was like, I'm a lawyer, just quit my job to focus on automating the practice of law and contracts and I can code a little bit. And then room went around and on the other side of the room was my co founder Kai, who said, I'm a software engineer from Palantir who just quit my job to focus on automating the automatable parts of practice of law. And I'm learning a lot about the legal profession. So we kind of looked at each other and did the Spider man meme. Lawyers have one tool, which is Microsoft Word, and software engineers have Unit Test, they have GitHub, they have built all of these great tools for working with text. And so could we take some of the principles behind those tools and apply them to the lawyer work? I actually one time counted the number of clicks that it would take me to take one Word file, compare it with another Word file and output a PDF that was named the way I wanted to. It was like 17 clicks and sometimes you're doing 25, 30 red lines and it's just like a huge waste of time. So I made a script that would basically take one folder, redline it against this folder, output into a new folder with a naming scheme post starting the company before we even had any actual automation. What we had is an email alias. So it was like adminironcloud AI and you would cc adminironironcloud AI on transaction you want to do. And then it would basically put that transaction on a dashboard and walk you through it. So if I want to do an NDA with you, I would say, hey Dylan, let's do an NDA. I've cc'd ironclad here. And then Ironclad, like the AI would reach out to you and say, hey, what's Your company address, what's the email we should send it to? And it would collect all the information and it would send you the docuSign at the end of the day. And for me it would show a dashboard like collecting information From Dylan, preparing DocuSign signature requests like storing file in your Dropbox folder of legal documents.

Really cool.

And we just walk you through it. I mean the email was of course me.

So you were the AI?

Yeah.

Are you still the AI behind Not.

The AI, Although our AI is named Kai after my co founder Contracts AI.

Oh, that's really funny. And how has that changed in particular over the last, let's say two years as the more modern AI models and tech has really started to work its way more into production.

Yeah, so the idea around Ironclad from day one and still continuing to this day is those two parts of the contract making the contract and then managing the contract after it's been made. What we realized was workflows were a better way of making contracts that than just an AI that's on an email alias. Because even if that AI would get something wrong 5% of the time, you really can't have that in a business contract. Like whoops, we put the wrong information on page 37 and you didn't catch it. Sorry about that. It doesn't really work. So we had to be like 100% accurate. And the way we could get to 100% accurate on the contract creation was through really great workflows. And of course once you have a workflow.

At the end of that workflow you have a bunch of really great structured data. Incidentally, adding workflows to this old school software category called contract lifecycle management helped us truly in the business school sense of the word, disrupt the category. Around 2018, 2019, Kai started, my co founder started saying, hey, I read this paper. When was the transformer paper?

2017.

2017. Okay. So I think like yeah, 2018 we started saying, okay, there's a potential breakthrough here. It's not necessarily.

Usable yet, but we need to start seriously thinking about the ironclad AI piece here because we're making contracts great in our workflows, but if we can get the ability to extract from 10,000 PDFs all of the data and put them in our database as well, that's going to generate a ton of value for our customers. And so we started retooling with the idea of layering in AI specifically at that point into just the data extraction.

The thing that stands out to me is that you had A clear use case you were trying to apply AI technology to. And I think that's very different from where a lot of companies are at today in that they're looking at AI technology and they're thinking, where can we apply this in our company, in our product? But I think with you guys, it seems like you always had a clear area within the product that you're trying to apply it to, and that really narrows the scope and, and that helps you, I imagine, get something to market faster. Has that been the case?

Yeah, yeah, I think we were pretty quick to get something to market and I don't know, like, we do find that attorneys are pretty amenable to AI, which I think, like, it kind of runs contrary to the popular narrative. I think we've definitely done a lot over the past 10 years to win attorney trust, which I think, you know, maybe has some analogies to other markets, but like, winning that trust, you have to be really precious with it. But if you have it, you can do some interesting things and you can like, have more of a dialogue with, with users that do trust you. And I think that that, that accounts for kind of like some of the delta between what we see in our user base and what like the general industry.

There's something really interesting in there that stands out to me, which is like your vision when you started the company, it sounds like, was to be this AI native software solution for lawyers. But back in 2014, the AI just wasn't ready. You were creating classical ML models. Now there are a ton of AI native software solutions coming to market, focusing on lawyers, focusing on the legal market. You guys are sort of an incumbent where you're north of 100 million in ARR.

But you're still a startup, you're still a tech company. You have a lot of tech capability, but you've earned that trust over the last decade with your users and with your customers. How are you thinking about as a founder, as a CEO, just balancing all this so that you're not too late, but you're also not too early as you're thinking about.

The more exotic AI solutions you're trying to bring into the product and how you think about really transforming the product to be an AI native solution, an AI native product when it's super competitive, but it's not clear if the tech's really ready yet. How are you just like balancing all this as a CEO?

I think one thing, that one tool I use is just like thinking on a product by product basis and what the end goal of a specific product is. So with respect to clm, I think it's about making sure that we are using AI in every part of the CLM product that we can possibly generate user value in. One of the interesting things about starting with an existing scaled product that we've embedded AI into all the parts we can is it's let us get a bunch of data back from users around like what they like, like where they see the value. And it's also led to the ideation of new products for us.

How risk tolerant are you as a company with testing these features out across your user base, across your existing flagship product to learn?

I think I would distinguish between.

Risk tolerance that we have as a company and risk tolerance that we have to put an experience in front of our customers. So I'm comfortable taking a lot of risk and I think that's one of the benefits of a founder led company. But I don't want my customers to be taking any risk in using the product.

There's a couple ways that shows up. One, we're rigorous in testing. So there's been a lot of noise around AI products and standalone AI products in legal space and we have really focused on our flagship product in terms of the public narrative. We are in rigorous testing and have been for some time around a standalone product.

Like a standalone AI native.

Yeah, exactly. But like you could say we've been slow to launch that. I would say we have a really high bar for what we're gonna put our name behind.

In the industry.

Can I ask you a question on that?

Yeah.

How much tension do you feel at this stage of the company from investors, from your team on responding to the noise in the market, on getting something out there versus this thoughtfulness that you're taking in? Really want to be thorough. We don't want our users to have any risk. Yeah, I'm sure they're asking for it, your users, but you're seeing what the evals are coming back. How much tension is there that you're facing as a CEO right now?

I think a lot of that tension has been alleviated by the fact that we do have clm. Is this natural place to put a lot of AI where whether it's that first draft AI review or the data extraction, you can look at the ironclad product and just you will naturally experience a lot of AI and that's had business benefits for us. Like we're growing at a good clip. Like you know, the unit economics of the business are improving quarter over quarter. All the stuff like the investors might otherwise be pressuring on. I think the AI goodness from layering that into our existing product has accounted for that. But for me it's about making sure we don't get complacent with just that because I really do think we have to disrupt ourselves. And you could say launching new AI products is actually a really risky thing for us to do because who knows, it could cannibalize our existing product line if it's really successful. And it's hard to predict how that's going to play out in the market. But I think that's the kind of stuff we have to keep pushing the boundaries on and making sure that we are taking enough risk.

How do you as a CEO think about you are an AI company. The company's vision was to be an AI company. You are an AI company. But how much of the technology.

Is strategic for you to develop yourself versus the application of it is the strategic angle.

I'd say our current view, which seems to roughly match with the large model providers, is that the verticalized application stuff is things that we're going to need to be uniquely good at and we're going to develop our own proprietary stuff around. Whereas the foundational capabilities.

We should stay on the latest state of the art and invest in anything that allows us to stay on the latest state of the art. I'd say specifically with respect to the foundational model companies, our approach is to constantly be evaluating that and.

We rely on OpenAI. We do some stuff with Google Anthropic's in the mix, so we're not all in on any one of them and we want to.

Have a sophisticated viewpoint on what each of the models do best and then apply that in our application because we are really covering a wide variety of use cases. It's everything from like a commercial contract negotiation to summarization of like 270 pages of M and A docs, to extracting data from 10,000 contracts. And they're the skills that the models need to have in order to do those different tasks can vary.

How does that compare to what you're seeing from competitors? I think some companies are saying, hey, we're going to go take Llama three, we're going to fine tune it. It's going to be ours. That's our competitive advantage. You're taking a more balanced approach, it sounds like, which is some stuff you're going to develop yourself, but you're going to rely on the capabilities from other companies developing these foundation models to just get better and better and be quick to leverage Those and evaluate those. Have you seen over the last 18 months, 24 months.

You guys actually be able to move faster than companies that have said we're just going to develop everything ourselves?

Yeah, well, I'd say this is where the technological landscape is changing so rapidly. So I'll give you a good example which is.

Case law, right? Like you don't want to make up and hallucinate a case statute.

And it's really important that there's almost no level of hallucination on that. You want to be pulling from the actual cases and citing them correctly. It's something that like isn't that relevant to a contract lifecycle management platform, but to a general AI product, pretty relevant, I'd say like nine months ago.

The best way to do that was to go to OpenAI, pay OpenAI a lot of money, strip out the Reddit trading data, put in Delaware case law data and you didn't hallucinate anymore. The problem with that is that model stuck on GPT3 and it doesn't have GPT4, it doesn't have GPT4O. You lose all of the benefits of the main branch. And this is where like having a great technical co founder like Kai is a huge advantage. Because while that might have seemed like the right approach, what 9 months ago Kai was like, I don't think that's going to be the right approach in a year. And really there's going to be benefits from staying on the main branch. And what we found is one, the.

Hallucination level is so much less on GPT4L than it was on GPT3. But two, there are techniques that have evolved like.

Application of REG that help you supplement and make sure that you're not hallucinating the case law. And if you have that Delaware case law.

Database being on GPT4O plus some customization plus rag on your good data set. And the data sets have been really improving in the legal field over the past year. Harvard just put out an incredible.

Clean database of case law in the past couple months. Those convergence of data, getting better techniques and the foundational models have produced a much better result than the approach from nine months ago of kind of forking the model.

That's crazy that. I mean nine months is a very quick timeframe to shift technology strategy. Have you seen competitors take different approaches though and say we're just going to go, we're going to develop everything ourselves. There's this AI wrapper.

Framing that I think was wrong last year that emerged like, oh, this company is just an AI wrapper over X.

And so I think that pushed some companies to say okay well we're going to then go develop everything ourselves because we want to end to end develop the technology stack and all the AI models. I'm sure you've seen some companies take that approach in your space. How has your technology strategy which is like just stay on the main branch, be creative, focus on the applications and the user value.

Fared out, would you say, compared to the competitors?

I'd say from a strictly technological.

Viewpoint and results driven viewpoint, extraordinarily satisfied. Like 11 out of 10 on that one. And again huge credit to my co founder because I would say as the non technical co founder I was like that sounds pretty good to be on your own, fully custom trained, OpenAI provided model. And Kaya had a very strong opinion on that which turned out to be totally right. I think what's interesting is that still hasn't fully played out is the marketing angle around that. I mean the market's discovering it, but when you look at results of one company versus the next. But that's taking some time to come out and I do think there's a marketing message which is we have an LLM that's completely custom trained for you that does land it sounds better and. Sounds better.

Yeah.

So I think the market is still playing catch up to the reality of the results. But. And it's going to be interesting how that plays out.

I think what's tough for earlier stage founders too is that that story also sometimes can sound more compelling to investors like hey, we're going to go create a specific model for X. But to your point, that might have an advantage for only three to nine.

Months and then you're worse off and.

Then you're worse because the main branch, to your point, you've now forked.

Yeah.

You're not getting updates from the main branch. The main branch is getting better and better.

Migrating customers off that difficult. Yeah.

So something we were talking about is enterprise adoption of AI has actually gone down year over year. When you look at applications that have made it into production and so I think figuring out okay, are you replacing lawyers? Are you fusing AI into workflows? Are you doing something in between? Is a question that a lot of companies that are incumbents and I hope you don't take offense by becoming oh no, I like how I'm taking both.

We're an incumbent and a startup. Yeah.

So incumbents are trying to grapple with how forward do you go? Do you say, okay, ironclad is now your AI lawyer versus ironclad, which you know and love now is AI native and can make you 10x more productive than what you're used to. Because we're pulling the latest AI technology into all parts of it, which again, from a positioning perspective is not as exciting as this company over here that's saying we have an AI lawyer come use it. It's going to completely automate away your need for a paralegal or a general counsel, which like, I don't think the technology is ready for that yet. And so how do you just think about this and manage this, these dynamics?

Yeah, I'd say a couple things. So one is this is where I do think it's really important to have a multi product strategy. If you're an incumbent and you can't lose the direction of your main product, there's so much momentum behind that. If you're an incumbent, it's an existing category. You should have AI everywhere. But don't lose sight of the big.

Picture on your main product, like pull AI into it.

Yeah, and I think this kind of goes to some of the like Geoffrey Moore crossing the chasm stuff where I think there's like four zones that the company operates in and you can have different zones operating in the same company at the same time. Like you have your performing zone where like things are working and you're really about like getting into the unit economics and the performance of the business and the product. And then you can have like an incubation zone which is where you're trying out new ideas and those can be judged on totally different things. And I think it's very important to not lose sight of that main product, but also be making those incubation bets, some of which are gonna pay off, some of which are just gonna get you more information.

And I'm curious, just when you guys talk about your technology strategy, how much do you care about.

How you're achieving that delta in performance versus the fact that you just are gonna have a delta in performance and that's what's important.

Yeah, I don't think we care at all actually. So I think having that delta in performance in our very specific vertical of understanding and recommending contract language.

That'S where we feel like the 10 year sustainable advantage is. And we've got unique viewpoints and data that we can use to sustain that for some time. But as to how we get there doesn't matter. But I do think it's critical for our success as a company to make sure that it is There we did.

Talk about how enterprise adoption of AI, at least according to this report that Bain put out. Especially in the legal profession, when you look at adoption in production is still like single digits from your perspective. You're probably like one of the.

Most knowledgeable experts when it comes to AI adoption of the legal profession, especially in the enterprise. What does it take to get to 100%? What still just fundamentally does not work well enough? Features, capabilities that are just still not possible today and that are limiting that adoption?

I think it's less about features and it's more about.

Sophisticated sophistication of buyers and.

Getting them comfortable with.

New ways of thinking about data. And so you know, obviously like we have zero training contracts with all of our providers such that our customers data is not being used to train any foundational model and isn't even getting stored there.

And I think that's part of the reason because we do have a very sophisticated contract based story around.

The data and how we are protecting our customers from.

Their data winding up somewhere they don't want it to wind up. That to me is where the gap is at least in the enterprise. Is it starting to grapple with where and how data is used in generative AI applications and having a lot of discomfort with that.

So you think the technology is good enough today? It's more about customer education, getting comfortable with data security.

Yeah. And I think that explains the gap between what we see, which is a lot of AI adoption and what the industry reports are, is there's like we're one of the only, if not the only scaled company in legal technology that is in corporations and.

I think particularly in enterprise there's just no chance of a series A startup getting the legal team comfortable. They just haven't been around long enough. There's like not enough funding, there's not enough sophistication of the data report and legal is very sensitive to that.

Interesting.

And it's only because we already have a product in there and we have our AI stuff is built on the same architecture and it's more of an addendum to the existing work they've done than a net new process that maybe enables us to show up in a different way. But I think it's going to happen. More companies will mature.

It will get more comfortable, there is going to be more of an industry standard.

I'm curious as a final question, the last 18 months have just been crazy for every company. What's been the biggest thing you've been surprised by over the last 18 months?

I mean, it sounds obvious, but pace of change, like, it's just amazing. Thinking about to nine months ago feels like almost as long ago as starting the company, which was 10 years ago, like, totally different state of the world.

And, you know, it's also giving me a ton of energy as a founder. And as I talk to other founders who are kind of like 10 years in, it's a grind, you know, to be a founder. But having this like, boost of energy around, absorbing new information every day, trading notes with other founders, getting hands on with a new product, that's been really fun. So that's the real benefit of all this, to me as a person, is boost of energy.

Really cool.

Yeah.

Well, thanks again for sitting down. This has been awesome. I really appreciate you doing this.

Thanks for having me.