Steve Loughlin (00:13):

Welcome to Spotlight On, I'm Steve Loughlin, Partner at Accel, and we're so fortunate to have Cai GoGwilt, the CTO and Co-founder of Ironclad here with us today. So thank you, Cai, for being here.

‍

Cai GoGwilt (00:23):

Yeah, thanks for having me. Excited.

‍

Steve Loughlin (00:25):

And I think, I know you go by Cai, but recently you've been going by C-AI because you've had a name change, which is a full commitment to your company. 

‍

Cai GoGwilt (00:34):

The running joke at the company for a while has been, you can't spell Cai without AI. Yes. Yeah.

‍

Steve Loughlin (00:40):

Well, thanks so much for being here. I know that there's a lot of exciting things happening for Ironclad happening in the software ecosystem right now, but why don't we start with just a level set, kind of what Ironclad is, this company that you guys started, how long ago, what year did you guys found the company?

‍

Cai GoGwilt (00:55):

2014.

‍

Steve Loughlin (00:56):

2014. 

‍

Cai GoGwilt (00:57):

Nine years now.

‍

Ironclad’s mission, inception and Cai’s interest in legal

‍

Steve Loughlin (00:58):

Okay. And what do you guys do? What's the mission of the company? What is Ironclad all about?

‍

Cai GoGwilt (01:02):

So Ironclad is a digital contracting platform, and our mission is to power the world's contracts. So the oldest piece of business technology is the business contract. And what we're doing and what we have been doing for the past nine years is taking this piece of technology, transforming it, and bringing it into the digital age. And so we also believe that the business contract is kind of the atomic unit of business. And so by innovating on this, we're actually setting an incredible new foundation for the entirety of business and commerce.

‍

Steve Loughlin (01:32):

It's amazing. So can you talk a little bit about, let's go back to 2014. Where were you in your life and how did you meet Jason and come up with the idea for Ironclad?

‍

Cai GoGwilt (01:42):

I was a software engineer at a company called Palantir, where I worked on military and intelligence applications of big data and AI. Before that, I did my undergraduate and master's at MIT and computer science and physics. So in 2014, I left Palantir. I was really interested in applying ideas that help make software incredible to other industries. I believe that the way that engineers kind use technology to innovate themselves, AI being a good example here in generative AI, was a good precedent for how other industries would innovate collaboration, and the future of work. And legal just spoke to me. 

‍

Steve Loughlin (02:26):

Can you talk a little bit, how did legal speak to you? Legal  doesn't speak to a lot of people. Did you have some experience with it or was it just intuitively kind of observing their craft? You were fascinated by it?

‍

Cai GoGwilt (02:37):

Both. Okay. So I had friends who were lawyers at Palantir. I made friends with the legal team, great people there, and I got to know them early in my time at Palantir. And so naturally after leaving Palantir wanted to talk to them, talked to my friends who were in legal and just got absorbed by how the way they thought about their work was so similar to the way as a software engineer, I thought about my work.

‍

Steve Loughlin (03:02):

And your original idea. So the way the story was told to me was Jason told me that there was some engineer from Palantir who was one of the 10 people in 2014 in the barrier interested in legal tech, and it was some guy named Cai and he wanted to build the GitHub for lawyers. Is that true?

‍

Cai GoGwilt (03:21):

That is true. Okay. And the 10 people part is also accurate. Back in 2014, I was like, I need to learn about all the people who are interested in this, thinking it would be a several months process. And the next week I was like, okay, I've met everyone.

‍

Steve Loughlin (03:37):

Oh my gosh. You guys met at a Stanford lecture, is that right?

‍

Cai GoGwilt (03:40):

We met at a Stanford lecture, yeah.

‍

Steve Loughlin (03:42):

Okay. And then you started working out of a motorcycle garage or something like that. Is that what happened or how did it go from meeting each other to working together? 

‍

Cai GoGwilt (03:49):

I like the motorcycle garage story that Jason tells. I think the part that he forgets about or ignores is we worked out of there for an hour because it was so clearly a bad idea. So yeah, Jason was parking his motorcycle and the motorcycle garage was thinking, Hey, the Silicon Valley thing, a startup thing, let's make a co-working space. And so they turned the upstairs area into a coworking space and it was just way too loud to get any work done at all. So within an hour we were like, yeah, we can't work here.

‍

Steve Loughlin (04:22):

Go to a coffee shop from there?

‍

Cai GoGwilt (04:23):

We actually ended up at this weird art studio, this maker coworking art studio. 

‍

Steve Loughlin (04:31):

More complimentary to what you guys were trying to do.

‍

Cai GoGwilt (04:34):

Definitely more quiet.

‍

Ironclad’s AI thesis, and the road to Series A

‍

Steve Loughlin (04:36):

So you guys get together and you spend the first couple years, I mean, before we met you, you guys had been working for three years. Can you talk about before you guys raised your series A, what you guys were working on and some of the twists and turns that you guys experienced?

‍

Cai GoGwilt (04:51):

The first domain name was ironclad.ai. We had this thesis that AI was going to go through a moment in the next three to five years or something, and that by building the foundation for legal, for digital contracting, we would be perfectly positioned in three to five years-ish to take advantage and innovate the practice of law with this new revolution in AI. And then three to five years later, we met you and we pitched to you on - in three to five years, AI's going to go through a moment.

‍

Steve Loughlin (05:25):

Yeah, it was really interesting. I remember meeting you for the first time in our San Francisco office with, I think it was you, Jason, Vas and myself and Vas and I thought we were walking into a legal tech pitch, but it wasn't a legal tech pitch. It wasn't even an AI pitch. There was AI kind at the end, but that was when enterprise collaboration was having its moment. And what was very obvious to us was you were picking the contract as this core object of something that everyone inside the company touches, and you were building the tooling for everyone inside the company to collaborate more efficiently starting with the general counsel, but salespeople, HR people, procurement people to be able to collaborate with the general counsel. So they weren't working out of their inbox, they were actually having this workflow happen. And I think that's what got us super excited.

(06:21):

And then there was some footnotes, some asterisk that was like, oh, by the way, we're going to have all this legal data of how people work together and the content of the contracts, but we didn't really talk about that for the first few years after. It was all about collaboration. And so you guys have built an incredible business with great customers, great logos. What was the moment where I think every board meeting we talked about should we do something with AI? We even saw emerging companies who were just focused on that but didn't have the data or the workflow. Do you remember when you decided, Hey, we need to actually start investing in this?

‍

Cai GoGwilt (07:00):

Collaboration was having a moment when we first met and it continues to be fundamental to Ironclad. The contract is like this collaborative moment that people come together in across different functions. It's not just the legal team, right? But we were also starting to lay the groundwork for AI. I mean, you gave us our annual plan framework, right? Thank you for that.

‍

Steve Loughlin (07:22):

Yes, you're welcome.

‍

Cai GoGwilt (07:23):

Yes. But I don't know if you remember company, company intent number four or five was always build the legal AI system or something like that.

‍

Steve Loughlin (07:34):

Yeah, no, totally.

‍

Cai GoGwilt (07:36):

So in the background, we were always working on it and I think we had several debates on the boardroom saying, okay, we need to up the investment here. We need to lower the investment here. But I think what that means is we've been investing, as Ironclad, we've been investing in AI since at least 2017. 

‍

Steve Loughlin (07:55):

That's totally fair. But it was more preparing for the moment, and then there was a moment.

‍

Cai GoGwilt (08:02):

We kept thinking there was a moment.

‍

Steve Loughlin (08:03):

You guys were a company that did have years of work data customers, and everyone says, oh, we got to allocate resources to this new generative AI set of user experiences to solve problems for our customers. How did you guys practically do that inside the company? Did everyone start working at it right away? Did you start with a smaller team and then do proof of concepts and then start adding? What advice would you give to the person who's sitting there with a customer base that's demanding it? They have the data to be able to do it, they have the infrastructure to be able to do it, but how do you staff it and how do you kind of create milestones around it?

‍

Ironclad Contract AI and what it does

‍

Cai GoGwilt (08:44):

We had this organizational inertia and we had to overcome that inertia. And to be clear, I am part of the organization, and so a lot of that inertia actually came from me, despite me being one of the people working on it. So we launched this thing called Ironclad Contract AI a few weeks ago.

‍

Steve Loughlin (09:04):

What does it do?

‍

Cai GoGwilt (09:05):

It answers all the questions you have about contracts in your organization.

‍

Steve Loughlin (09:10):

So I'm a lawyer, I'm sitting at my desk today. I open IroncladAI, what was my workflow before Ironclad AI and what do I do now?

‍

Cai GoGwilt (09:18):

Have I ever promised a customer ever that I would keep their data in the EU? Your solution right now is either maybe craft some keyword searches and hope you hit something or just assign an intern or a team of interns to read every contract or what are the variations of different data residency requirements we have in all of our customer facing contracts with Contract AI, you can just type in the question and it'll use the concepts around AI agents and LLMs to actually answer that question for you and show you the chain of thought and reasoning. 

‍

Steve Loughlin (09:51):

This is live with customers now.

‍

Cai GoGwilt (09:52):

This is live with customers now. Wow. Yeah, it's pretty insane.

‍

Steve Loughlin (09:55):

So you said something recently that stuff that you thought would take years is now taking a week or two weeks. And I know if you talk to a lot of folks in the space, there's so much experimentation happening that people are kind of loathed to forecast where this is going other than it's going to be transformative, hard to recognize, we got to run these experiments. But if you look at your domain and you've launched contracts AI and you think I'll kind of give you two bogies. One is a year from now, one is five years from now in the legal context, what are some possibilities that you're excited about of experiences that you guys are going to enable for folks to interact around contracts?

How Cai wants to evolve Contract AI over the next 1-5 years

‍

Cai GoGwilt (10:44):

Right now, Ironclad contract is really capable of answering these questions that the legal team has around variations in language, what they've agreed to across a certain set of industries or something like that. It even does weird things help you justify arrays. It'll do that. It'll come up with a creative way of doing that. But as we add tools and capabilities to contract AI, it's going to be able to do more and more and more things. So looking out a year from now, my hope is that ironclad contract AI will be able to intelligently answer questions around supply chain management or things like that, be able to look in your contractual obligations, try to anticipate a natural disaster or extreme weather phenomenon happening that was unexpected, and help you answer your questions as a business, as an organization about what your exposure there is based on the contract, the data locked inside of your contracts.

‍

Steve Loughlin (11:47):

How do you think what you're doing, this is highly relevant to other companies in this context, affects the way that Ironclad’s platform, you guys have tons of integrations with folks like Salesforce, DocuSign, other players in the ecosystem. How will these platforms interact with each other given this context?,

‍

Cai GoGwilt (12:12):

It's going to be really interesting. I think first we're going to have to learn as humans how to interact with these new set of features, and then we're going to have to learn how those AI agents interact with each other. 

‍

Steve Loughlin (12:24):

Is there going to be a whole new set of APIs and interfaces between these apps?

‍

How AI agents fit into the mix

‍

Cai GoGwilt (12:29):

Almost certainly. Yeah. It's going to be really interesting. So a topic that's starting to come back into the forefront is AI agents, and this is an area that we're really pushing on.

‍

Steve Loughlin (12:43):

Yeah. Can you talk a little bit about Rivet and kind of the work you guys are doing this to? 

‍

Cai GoGwilt (12:48):

We started working on Ironclad contract AI a few months ago. The idea was basically like, let's just take a big swing. We know one of the biggest problems confronting our users is they want to be able to understand and answer questions about their contractual obligations across their entire set of contracts. And we were like, okay, retrieval augmented generation RAG is this thing that's been around, if you've seen chat with your PDF, that's the concept that's being used. And so we were like, at a bare minimum, we're going to have RAG, but what is the ceiling here? And so we just started working on the ceiling. We do your inspiration from Lang Chain Auto, GPT, and decided to try to build this agent that could use Ironclad to answer any question. And it didn't go very well. Within two weeks I was like, okay, nevermind. We're just doing retrieval, augmented generation.

‍

Steve Loughlin (13:44):

I think I talked to you at that point and yeah, you were low. You were kind of like, that's not working. And I think you gave me a demo that froze and you kind of hung up the phone on me or at some point hung up the Zoom.,

‍

Cai GoGwilt (13:57):

One of our engineers, Andy didn't give up. And over the weekend, maybe the weekend after we talked, he came back after the weekend, he's like, I built this thing, Rivet, and I refactored our entire agent into it and I think we can make it work now. I was like, this is a crazy idea. Visual programming languages are like, I don't know, I'm seeing it work in some domains, but I've always been suspect of them. And then I actually tried using it and it was a total game changer. All of a sudden we went from putting break points and logging what the AI agent was thinking about and just going through reams and reams of logs and code and just getting utterly confused about why things were breaking to suddenly it was like, oh, that's why it's breaking. It's right there. I'm just going to change it.

(14:46):I'm going to experiment with it. I'm going to iterate quickly. And we were able to suddenly add skills to our AI agent. We were able to stabilize the agent loop, and then we didn't hit any more walls after that. We just kept going. And at some point we were like, okay, this started as a experiment to see how far we could go. We haven't reached a limit. We're just going to stop now. We're going to productize something because we really need to productize something. And that's how Rivet was born. That's how ironclad contract AI was released with the feature set it has now. And that's why I'm so bullish on over the next year, we're just going to see the capabilities to that feature massively split because we actually stopped ourselves from exploring the bounds of this technology. We were like, we know the bounds are well above what's necessary to be game changing in our industry.

‍

Steve Loughlin (15:39):

And what's kind of the future of Rivet.

‍

Cai GoGwilt (15:43):

We are excited for Rivet to be part of the LLM tooling infrastructure. We think it's really powerful that it's open source. Already within the first week, we had a thousand stars on GitHub. Within the first 24 hours we had our first community genuine community pull request. I think there are five plugins for it that have already released from LLM evaluation tooling to assembly. AI is adding a plugin for their large foundation models around audio. And we're in a phase right now where everything is so new that usually when you think of a new technique, you hold it close to the chest and you say, all right, no one's discovered this. And it's non obvious. But right now, the discovers people are making are mostly obvious and other people are making them in parallel. So by being part of the open source community, we can share those ideas and we can actually advance the concepts around LLMs and AI agents much more quickly than anyone working in closed quarters.

Decisive moments, and knowing when to swing for the fences

‍

Steve Loughlin (16:44):

I think one of the things that everyone is dealing with across our portfolio and across the boards I'm on, the question is, when everyone's having their kind of generative AI moment, can you give us a little more detail on how that happened at Ironclad and maybe extrapolate a little bit to how other people can learn from the way you guys, it seemed like from the outside in went from day-to-day business of growing your company to repositioning the company to take advantage of this innovation wave.

‍

Cai GoGwilt (17:18):

Funny enough, so we launched Ironclad contract AI a few weeks ago, and at the launch event, I think I said something like when I discovered what AI was now capable of, my first thought was it's time. But yeah, no, my first thought was not its time. My first thought was like, oh fuck. And part of this was, well, maybe to back up, I was in the office, one of our teammates, James comes over to me and says, Hey Cai, have you looked at GPT3 recently? And I was like, yeah, I looked at it maybe not recently, but I looked at it, how much could it have changed? And he was like, no, man, you have to look at it again. The legal engineering team is like has been looking at it all morning. They're obsessed with it. And that made my ears perk up because as you know, our legal engineering team is kind of our secret weapon. These great people, former lawyers, former consultants.

‍

Steve Loughlin (18:19):

Yeah, they're the canary in the coal mine of this thing could be working.

‍

Cai GoGwilt (18:22):

Exactly. Yeah. They're not going to get hoodwinked by technology being applied to legal in particular. And so I went over, talked to a couple of legal engineers, they showed me what they were doing, and that was when my first thought was not it's time. It was like, fuck, because I was about to, I'd written up this huge document and was getting ready to present how our AI strategy was going to work. I thought there were these five points and you're going to do them. It's going to take the next year. And then we were going to have leadership in AI, and I was staring at this clear proof, that strategy that I just labored over was just not going to work. And so that's kind of what got us started was just seeing this technology and we immediately started prototyping something. And I think that is maybe one of the key things for people working on this. You've probably thought about what the AI moment means for your customer base. So focus on the customer problem and then just take a massive swing for the fences, right?

‍

Steve Loughlin (19:31):

What do you mean by massive swing for the fences.

‍

Cai GoGwilt (19:32):

Right now with generative AI, I feel like a lot of people see it as a hammer and a hammer looking for a nail. You see? Okay, cool. Retrieval, augmented generation, RAG, we can make chat with your PDF. What can we do with this? That's classic hammer looking for a nail and you'll find a nail. But instead, think about the thesis of your company and think about how you've been thinking three to five years from now. AI will have a moment and we will be the best positioned to do this thing for us. One of those big things was AI will be negotiating contracts.

‍

The necessity of looking ahead to identify customer problems

‍

Steve Loughlin (20:09):

Got it. Yeah. So you're saying don't just bolt it on. When people took web and we're like, oh, we can do it on mobile, and they just copy and paste it and said, now you can see it on mobile versus rethinking the entire experience because you now have this new device in front of you. So you're saying, look at the capability of this. Think about your customers and don't just copy paste.

‍

Cai GoGwilt (20:32):

Start with the customer problem. Look at especially the customer problems where you're like, three to five years from now, we're going to do this massive thing and ask yourself, actually, can we do this today? 

‍

Steve Loughlin (20:43):

Think the other reason you have to do that is the reality of the software markets is that the way people are buying is different. We went through this whole, everyone grew when companies are growing on seat based pricing, but now you really have to price on value consumption usage and the ROI that you're delivering, and the good news for that is generative AI enables those experiences. The whole premise is that you're taking one person and giving them an Ironman suit of whatever that function is and making 'em 10 to 20 x more productive. And so I think a lot of it is not just aspirational, it's also existential. If you don't do it, absolutely no one's going to buy your product.

‍

AI efficiency gains are already materializing

‍

Cai GoGwilt (21:27):

People need to be embracing generative A* because we've started saying, or for a while, we've said lawyers, AI is not going to replace lawyers, but lawyers who use AI are absolutely going to replace lawyers who don't. And I think those massive efficiency gains are actually happening. We talked about AI for negotiation, which was the first thing that we launched with generative AI earlier this year. We've seen companies negotiating, I don't know, 20 times faster with the technology. Maybe it'll be more concrete. One of our customers, Orange Theory, Orange Theory Fitness, do you know Orange Theory Fitness?

‍

Steve Loughlin (22:10):

I've heard of Orange Theory Fitness. I've actually been to Orange Theory Fitness. Quite painful.

‍

Cai GoGwilt (22:15):

Yes.

‍

Steve Loughlin (22:16):

They're good at what they do.

‍

Cai GoGwilt (22:17):

They get you into the orange. Yeah. So their business went bananas a few years ago, and their membership agreements just proliferated and became like every state, every county had a different membership agreement. Got it. Charlin on their team was tasked with consolidating their membership agreements down to one thing, which should have taken months because there were so many variations that she had to consider using Ironclad AI and the generative AI capabilities. She was able to do this in less than a week. And then another company did a similar thing in terms of adapting their template language and saw actually, I think a 50% reduction in turnaround time or processing time even as their number of contracts exploded.

‍

Steve Loughlin (23:04):

And are they feeding this back to you? You're observing the ROI. Are they also having that moment themselves where it's like, this would've taken me.

‍

Cai GoGwilt (23:12):

I mean, this is happening. This is happening across I think a lot of different functions within a business, right? In legal, we're seeing it a lot across our entire customer base of people saying, I can now do 10 x more than I could do before, internally ourselves, where our brand design team has fully embraced midjourney, and I've been talking to them and they're saying, our Ironclads brand expression is quite expensive to produce, but we've been able to, I don't know, a hundred x their output in this very expensive to produce brand style. We're producing custom images for every single piece of content we put out there because they can do it. Do it. Yeah. Podcasting, right? We're bringing podcasting podcast production in-house because of systems like D-Script that just make it so much cheaper. So I think a lot of different areas are seeing this 10 x improvement, legal being absolutely one of the first to see it with Ironclad. And so it's not just a matter of these products are going to get 1 x better Ironclad with generative AI is 10x better than any other product on the market, but it's also teams are getting 10x, 10x better. And so if you're at a company that has a generative AI policy that says you can't use generative ai, I feel like you probably only need to get out of that company. You are going to get left behind.

‍

Unintended consequences, and what keeps Cai & Steve up at night

‍

Steve Loughlin (24:37):

Every company needs to become a generative company. Every function needs to have their platform that gives them the 10 to 20x return. What are the risks to all this? What keeps you up at night?

‍

Cai GoGwilt (24:48):

Turn that question back on. What keeps you up at night?

‍

Steve Loughlin (24:51):

Any experimentation phase of technology, there's going to be things that really work. And then the things that don't work, especially in this context, I think is going to have to do with privacy. The unintended consequences of where the data's going, what the source data is for the model, creating experiences for people that we're just unintended. It's always unintended consequences. I mean, I think there's obviously a whole part of the world that's going to use the technology nefariously, and that's something that's in one bucket, but I think the experimentation is going to yield good outcomes and unintended consequences of bad ones and how we respond to that. But I think that's also just how innovation works. And so society's always dealt with this, how do we manage it? I think it's overall net positive, but it doesn't mean that it's going to be smooth sailing all the way. Yeah, no, definitely not smooth sailing. I hope. I mean, I think the fear is that I, the unintended consequences lead people to slow down on the innovation, which I don't think is the right thing. I think the right thing is to lean into it and figure out how to manage it, but I'm overall optimistic and excited about it.

‍

Cai GoGwilt (26:09):

Yeah. No, I like your take on it. 

‍

Steve Loughlin (26:12):

Feel free to use it on your next podcast.

‍

Cai GoGwilt (26:14):

Will do. Yeah. No, I think you're right. I think there's a lot of fear around it can now do my job function, am I going to lose my job? But I think human society, if we look at technological revolutions over the past centuries, it changes the face of human society. And I think generative AI will change the face of human society, but what we're left with on the other side is human society and get used to it. So yeah, I think the benefits that generative AI can bring far outweigh the negatives, but it is going to be a bumpy ride.

‍

Steve Loughlin (26:52):

I think it's a requirement to lead from the front. Were you and Jason aligned on the generative moment and what needed to be done, or was there some arguing that happened in that context with the caveat that actually my observation of you two as founders is one of the things I love about user, you guys have a deep amount of respect for each other. You have different skill sets and capabilities that are very complimentary, but you guys are also not afraid to debate and try and do what's best for the company.

‍

Cai GoGwilt (27:20):

Jason and I were very, very aligned on we need to take this disruptive technology, this disruptive innovation, and lean into it. There's no authoritative textbooks on how to use generative AI. Unfortunately, one of the best places to get information right now is Twitter, which is horrifying to me. X, X, yeah. But yeah, it's the best X/Twitter, whatever we call it, is one of the best places to get information. But there's also a ton of misinformation there. And so I think a lot of the top level alignment that Jason and I had and the entire team had on pushing and taking this moment and making it our own was maybe obscured by also the details or maybe guided the details of people are like, should we be training our own large foundation model? Should we be taking open AI at its own game? What are our moats now? What are not our moats? There's been so many trends and a new trend comes up every week. So I think the devil's been in the details and certainly working through that. 

‍

The necessity of staying hands-on and close to the problem 

‍

Steve Loughlin (28:32):

I think that's normal for a platform shift with the experimentation that's happening. I think the biggest mistake people can make is not engaging in it and just watching versus getting your hands dirty and kind of playing with the raw materials. How rivet emerged, right? It was because you guys made the decision to jump in 100% and start building, and then you ran into roadblocks and you had to solve those problems, and those become the ingredients to remote actually over time.

‍

Cai GoGwilt (28:59):

Yeah, no, I heard someone talk about how right now people are building tools for using large foundation models and compare the tools that are being built to everything that came before React in web development. React was built by Facebook because they'd experienced how difficult it was to build a complex application on the front end. Similarly, all of these tools that are coming out around large foundation models suffer from the problem of no one's built. Very few people have built very complex or sophisticated agents with large foundation models. So it's really hard to project into the future and say, okay, these are the problems that are going to hit. So that's part of what makes me so excited about Rivet makes us so excited about Rivet is it's actually been built because we've out of of necessity, out of necessity.

‍

Steve Loughlin (29:50):

Had to.

‍

Cai GoGwilt (29:51):

And what makes it such a no-brainer for us to open source is we've benefited from kind of the open dialogue that we've had with other teams confronting these issues. And we think that Rivet is a great place for people to come together and share those best practices.

‍

Steve Loughlin (30:08):

Cai, thank you so much for joining us today on Spotlight On. We look forward to watching your success and hopefully have you on the show again.

‍

Cai GoGwilt (30:16):

Yeah. Thank you so much for having me.