Transcript
Amit
Nick, can you tell us a little bit about your story and your background?
Nick
Absolutely. Right now I'm building GitLaw. But before that, I founded and sold a company called Whisk to Samsung. I founded it in 2012, sold it to Samsung in 2019, and it was basically a food AI platform that was able to reason about recipes, meal plans, and shopping lists. I built that before LLMs, before even things like TensorFlow and the libraries and services that made AI easier. We were literally building it manually on a computer in our office in the UK. I've also built before that an HR tech platform called Air HR. I'm an investor in about 20 to 30 different startups, mostly deep tech, LP in a couple of funds. But most recently, the last year, I've been focused on building a platform that basically makes legal document creation and review free, mainly because I was fed up of paying so much money for legal fees. I think I spent way over a million dollars on legal documents and processes, and I just thought there must be a simpler way to do this.
Amit
That's a problem that all of us as founders and investors have, so that's a great one. I want to start with your background. You were doing AI before AI was hot and LLMs came out. What is the biggest difference in building an AI company back then and now, in terms of the use of technology? I'm guessing back then you were using machine learning, and now you're using LLMs?
Nick
There's a massive difference in capability. Things that were really hard in 2014 to 2015 are now extremely easy and basically free. We were building natural language processing models, not deep learning based initially, because that was the best way you could do it. Then of course the deep learning models made it easier, and platforms like TensorFlow made it easier to use deep learning for NLP. But practically what we were doing was taking a recipe and figuring out which part is the ingredient, the amount, the instruction, which makes a big difference. Rice cooked versus rice uncooked is very different. Beef lean versus beef not lean, big difference. So we extracted all this data, created nutritional information, perishability information, and that allowed us to do things like meal planning and recommending the right recipes.
What we did back then, you could replicate now in minutes using one of the cheaper LLM models that's essentially almost free. But the biggest difference is that back then AI was one of the most difficult parts of what you were building, and it was a real differentiator, a real moat. We had to hire PhD-level university lecturers and professors to take the latest knowledge from academia and apply it. That very quickly changed. Today the AI is no longer the most difficult part. Other startups entering the food space didn't have as good AI as we did, and we ended up winning because of that. Within our financial constraints, we were able to build great AI. Other startups tried the same strategy and were not successful.
Today, AI is not really the real differentiator anymore. Everyone's using the same frontier models from Anthropic, OpenAI, and Google, and patching it together. The difference is a little bit in how you orchestrate it, but again that's not really that difficult. Most people use platforms like LangSmith or one of the various Lang versions. So really what the differentiator is today is much more about workflow and other things. It's harder to build a moat with AI, but it's also liberating, because as a small company you're able to use exactly the same AI as a well-funded incumbent. A big company like Harvey or Legora has raised hundreds of millions of dollars, but it's a leveler. It also means everyone's got access to great AI and the experiences you're able to build are incredible. It's never been as fast-moving as it is today, which is super fun, but it's also stress and anxiety-inducing, because you wake up and there's a new model, or a new competitor, or Claude just released Cowork and the whole industry blows up and Thompson Reuters has a 20% drop in stock value. The space is changing constantly. It used to happen every three or six months and now it happens every week. But ultimately what we're all able to build feels slightly like magic. It's amazing.
Amit
I think you covered a lot of very important points. I'll pick one. There used to be systems where maybe half the value-add came from the underlying platform, like a cloud provider providing 30% of what you're building and 70% built by you. But now it seems like the foundational models are growing exponentially and keep adding functionality. They want to be product companies now, Claude for Excel, Claude for legal. It seems like 70 to 80% of the value is now being provided by the foundational model, and you're doing a very thin layer on top. We are a vertical AI venture builder focused on financial services, just like you're building for the legal space. We still see some space left, things related to complex workflows, regulated industries requiring human oversight and compliance. But have you started feeling like Claude or Claude Code might be able to do 99% of what we're all doing, and therefore there's an existential threat to vertical AI?
Nick
I think it's definitely on everyone's mind, but that isn't the conclusion I come to. Your cloud example is great, because building a website before cloud was hard. You had to buy hardware, plug it in, run your own server, which meant the quality of the service you could build on top of that was only part of your focus. Then that became really easy, a couple of clicks. Databases became a couple of clicks. Other things became a couple of clicks. Slowly you could spend more and more time building much better experiences, because you could focus on that part and didn't have to focus on the rest. Was that bad for software? Absolutely not. Software grew like crazy. I think the same is true for AI. Suddenly you don't have to worry that much about that specific part, and you can instead focus on other parts.
Of course there is an existential threat when some of the models become so good at doing anything. The question is also a lot about customer acquisition and who owns the customer. If the models own the customer, and the customer is always in the chat interface or in the Claude interface or in the Cowork interface, and that's where they expect the service to be, then it's harder to own the customer. You can still get into the customer experience via an MCP integration or some other connection, but it's harder.
If I look at the legal space specifically, there are some important things I think Claude or ChatGPT won't do. If two people are negotiating, you want a good record of the revision history between the two parties. You want workflow that's fixed, like the signing process. One party's AI has certain context, the other party's AI has different context, and you're not blending your negotiation tactics with the other side's. You also want to provide templates or standard practices for your business. There are a whole bunch of workflows and bits of context that become really really valuable. A really easy way to think about it: if you're going to create an employment contract, would you trust a lawyer who is great at using AI to create that contract more than you trust yourself using Claude directly? The answer is probably yes. So what's that lawyer doing? They have lots of context and lots of expertise. In software you can basically replicate a lot of what that lawyer brings to that experience. I do think for quite a long time there's going to be space for software providers to build that kind of lawyer-led experience with all the right workflow tools built in.
Now, the interesting thing I've seen since Cowork and Claude is that every single founder or company with their own agenda basically defends their own thesis on LinkedIn. It's fascinating. Everyone is saying Cowork doesn't apply to me because of ABCD, very self-serving. What we should all be doing is try not to be too self-serving. Try to look at it from a market perspective. Try not to look at it in 2026, but project forward to 2030. That's the horizon I'm trying to work to. AI is going to come a long way by 2030, and it's three years out, so it doesn't feel too abstract.
On that horizon, I don't think the existence of these horizontal platforms like Claude and ChatGPT will be a threat to our business. I think we have lots of value to add, and I see that in our product every day. Within the next three years a whole bunch of other stuff will also become apparent in terms of how we can add more value. But is there risk? Of course. This was literally my thesis from a year ago in the investment deck. I don't care about Harvey, Legora, or any of those players. I literally don't care about the legal tech industry. Of course I look at them for inspiration, but I'm not worried about them at all. I am worried about the horizontal frontier models. That's the risk. They will connect brilliantly to my Google Drive. They will collect loads of great context and memory about me. That's what I'm worried about. And that's been true right from the beginning, so it's not a shock. But does it mean I'm not thinking about it and slightly anxious? Of course I am. I'd be negligent if I wasn't.
Amit
Since you mentioned looking 3 to 4 years out to 2030, what do you expect will happen by then in general, and specifically to your business?
Nick
I think the models will continue to get much much better. We're seeing that with every model release, and you don't have to be a genius to predict that. I think they'll continue to become cheaper too. Some people say they're going to start marking up the price, but I think that's absolutely ridiculous, because there are so many open source models and so much competition. If someone were to ramp up their price, there are so many alternatives now which are maybe not as good as the latest leading model, but not far behind. All the frontier models will become better and cheaper. The platforms will also connect more seamlessly with other software, so you can easily access the context in other places, the documents in your Google Drive, your SharePoint, wherever you happen to use. That's the main trajectory.
I don't think they're going to go super deep into verticals. I think they will create use cases and examples to show people how to build on top of them. The Claude legal plugin was one of maybe ten or fifteen example plugins published at the same time. People focused on the legal one because so much money has gone into Harvey and Legora, but it was just an example plugin, a lightweight way to show people how to use their models. I don't think they're going to go super deep.
There is some counter to that. Google has gone deeper than others. They've built e-sign into Google Docs. They're probably the ones I'm most thinking about in terms of going deep into legal, because you can see their product strategy has led them a little bit that way. ChatGPT going into health is worth noting too. And you could expect Apple to go quite deep into health because they've got Apple Health and connections into different medical record systems. But Apple haven't really done anything in AI, so there's still a lot to come.
I think we are basically super early. There will be some big vertical plays. I don't worry that it'll be specifically what we're trying to do. We're not building something for lawyers, we're building something for businesses, for businesses to create or review a contract. And it's free, or you can pay for more. I don't worry that we're going to have a whole bunch of competition for that in the very near future, or that the frontier models are going to try and enter that space. I think they're going to keep building their horizontal tools and keep offering more examples of how to build plugins on top of them.
Some of the big enterprises spending a million dollars or more on AI platforms will go and build their own solutions on top of frontier models because they want to own the data and keep it internally, and it becomes expensive enough to warrant building internally. But for us, we're free or $20 a month. Most companies do not want to invest the time and effort to try and build something that's subpar and maybe 30% of our product to save $20 a month. Who cares, right? You've got more important things to do when everything's moving so fast. If I was charging $500 or $1,000 a month for something on top of the frontier models, I might be more worried. But because of our price point and who we're targeting, I'm actually not too worried.
Amit
A lot of things we were taught growing up in the product and tech industry seem to be falling apart one by one. For example, we were taught you have to develop technology moats, but it's very hard to have a technology moat anymore. It seems like sales, marketing, and customer acquisition are now the most important things. And similarly, you used to have to build a large team really fast, but now you can have a very small team and do a lot. Some people are saying you can build a one-person billion dollar company. One of the things we think about is that if your TAM for your product or segment is small, if you're doing anything which is not the hottest or largest segment, you might actually be better off, because the foundational model companies and big tech companies have no interest in small business lending, for example. But they do have interest in health, in finance as it relates to stock markets, in what Bloomberg or Google Finance does. These are massive industries with massive TAMs. But if you go to a midsize industry with a smaller TAM and something very specific in it, it seems like it won't be natural for big tech or foundational model companies to attack. A lot of things are very counterintuitive now. What do you think?
Nick
That's a really interesting point, and there is a counter to it. If you've got a smaller TAM, the amount of money you can justify investing into building something is smaller. And then how do you compete against all the vibe coding army? In every single possible field, you've got founders vibe coding their way into everything. If you've got enough capital, you can build a really amazing experience as long as you don't charge too much. But there is always going to be the fact that anybody can vibe code their own version of your experience. So there's always a little bit of a balance of how much is that competition worth worrying about.
You've got on one side the horizontal frontier models potentially taking the big TAM topics. On the other side you've got all the vibe coding people, which is not just one or two people in a bedroom. Usually it is, but using Replit with Claude Code on top of it, they're actually building really really good apps and moving super super quickly. So in the middle of that, if you're trying to build a venture-funded startup, you've got to find that sweet spot of the right market to go after and how to differentiate sufficiently from both sides.
But if you do it, it's an amazing time to build AI companies. Look at the scale and speed of Replit, Bolt, Vercel, and Base44. I can't remember the exact number but Base44 went from nothing to $80 million in like six months, some crazy timeline like that. You would have never heard that before. Lovable talks about themselves as the fastest growing startup in the world. All AI companies that have been successful have got there super quickly. So this is what also creates that anxiety for founders, because you can be really successful really quickly if you get it right, but you're competing against so many people and the frontier models who've got billions and have massive talent density. It's a fun time to be building.
Amit
Something very related to this is domain expertise. I know if I look at your background, you did an HR software company, then a food AI company, and now something in the legal space. If I was to guess your answer it would probably be that domain expertise doesn't matter, that if you're a learning machine you can learn pretty much any field. But more and more in vertical AI people are talking about domain expertise as important, because there are very specific things in verticals. In financial services for example, you really need to know what the regulator will allow and what it won't, the compliance requirements, where you'll find the data and where you won't, which players are easier to integrate with. There's a lot of tribal knowledge. How do you think about domain expertise as a moat?
Nick
I don't think it's a moat, because you can hire domain expertise and it's not very expensive to hire. If I want the best financial regulatory knowledge person on my team, I imagine they're probably available on market for $200 to $300k, maybe a third of that or slightly more. So if that's the gap, I can fill it. There's obviously value in domain knowledge, but I also think there's a lot of value in knowing how to build a product, knowing how to align it with what users want, knowing how to bring it to market, knowing how to put together a team. There's lots of value in all of those things, but I don't think any of them are necessarily moats because you can usually hire the gaps you have.
As an investor, I definitely look for strong founders who have grit and can stick it through the ups and downs, and there are probably more ups and downs daily in AI than in previous periods. Do they have to have knowledge of the exact industry? I don't think so. I would personally prefer to invest in a founder who had built other tech companies before and had a track record of building software products. They don't have to be the founder, they could be a lead designer, product manager, or lead engineer who built their own personal projects on the side. That's probably harder to learn than domain expertise, which you can hire.
Amit
If you look at a company like Harvey, one of the co-founders and CEO was actually a lawyer, and he describes that it really helped him because he knew a lot of this stuff, and their go-to-market was very interesting. Instead of going to smaller companies first and working their way up, they actually went to the top 100 law firms in the US and started there, which he ascribes partly to having connections from the legal industry. So in cases like that, it could help a lot, right?
Nick
I think so. Obviously if you have both domain expertise and the ability to build a successful company, which the Harvey founder has proven, having raised successfully and built a team and a product that's actually used by real people, that combination is fantastic. If you can have all of it, perfect. But if you're a founder and you don't have it, I think you can make up the difference.
And one of the things I have learned over the last 10 to 15 years in tech and venture is that there are actually no patterns. I was told that when Anthropic was raising their Series A, 21 top tier firms rejected them, and then the co-founder of Skype actually led their round with a bunch of other investors. It's fascinating how, with all the things everybody has learned, some of these top VC firms still passed. But I imagine the founders of Anthropic are pretty impressive when you meet them. Before they founded Anthropic, if you met them, I imagine as an investor you'd think they're special, they've got something special about them. They have serious technical expertise as well. So on paper or in person when you review that proposal, I imagine they're very very impressive, which is why the ex-founder of Skype invested. But that said, the path to success has many many ways through it. It does not require a Sequoia seed round and a YC founding.
Amit
Okay, let's talk about GitLaw now. Can you tell us more specifically what problem you're solving and who are the customers?
Nick
Customers are small businesses today, and that can extend to medium-sized businesses in the future, and also individuals and enterprises over time. Small businesses who want to create or review legal documents. If you want a legal document, it's like a vibe coding tool. It's like Replit or Bolt or Vercel. You just type in what you want and it will generate it for you, not from scratch from a frontier model but based on templates that are tried, tested, and reviewed by lawyers. We have specialised sub-agents that are specialised in different areas of law like employment or privacy, and those call their own legal knowledge database to review the document properly. The same process works for reviewing a document you've already received. And there's free e-sign, so if you want to take it all the way through to signature, you can do that for free too. Every user gets $5 free a month in AI credits. If you pay, you get $20 of AI credits a month. Those free credits allow you to process a good number of documents through the process.
What we're really trying to do is make legal as close to free as possible. The AI is almost free, it's so cheap. So how can you justify $500 or $1,000 a month on top of the frontier models? I don't think you can. There is lots of value in orchestrating the different models in a certain way and providing legal knowledge to the frontier models. And we also do things like remembering the dates when your contracts expire, organising your legal documents for you, knowing what your insurance liability cap preference is and applying it to your contracts. We do all of that, but we charge $20 a month or offer a free tier. Because as we've spoken about, customer acquisition is one of the most expensive and hardest things to do. My logic was, if we just offer it for free to everybody as a product-led growth strategy, rather than paying $5 to Google or LinkedIn to get a customer in, we just give it to the user directly, which feels like a fairer exchange.
Amit
In terms of markets, are you serving the US, UK, Europe?
Nick
I've got a British accent, but we're a US company and we serve the US first and the UK second. We have lots of interest from all over the world. Our third largest market is actually India. We also have a lot of interest from Australia and Europe. But number one is US, number two is UK, and then other countries after that. We get tens or hundreds of new companies signing up every single day, and we have thousands of companies using it. The general feedback is super positive. People have tried going directly in Claude or ChatGPT and they come back saying this is what I was looking for, this is better, I trust it more because it's using these templates and the result was better.
Amit
Companies like Harvey, Legora, Casetext which was bought by Thompson Reuters. Can you differentiate them in terms of customer segment and what problems they're solving versus what you're doing?
Nick
Most of them, everyone you just mentioned, are focused on lawyers. They're providing tools to lawyers and to law firms. Harvey is targeting elite law firms. We don't target law firms. We target businesses directly. The reason for that was somewhat personal. I don't think lawyers will ever take legal as far as it can go in terms of value to the customer. I believe AI should drastically reduce the cost of many types of legal documents. But that's terrible for lawyers' business model. Their whole business model is selling hours. If they sell fewer hours, revenue drops drastically. And for partners who are in control of law firms, they've worked their way up so that they can become partner and get the benefits. It doesn't make sense for them to use AI to reduce bills. So I believe the mission is to drastically reduce costs and drastically increase speed, so it doesn't take weeks, it takes hours or minutes.
We will and do already connect with law firms as well. A user can create a document on our platform, share it with their counsel, and their counsel can provide feedback. There are lots of questions some lawyers have around that, like what about a document created on someone else's template. But there are definitely lawyers who are AI-first and are happy with that. The industry is changing. But we are focused on small businesses. Most other players, probably 99% of the ones you've heard of, are focusing on lawyers. That's the simple differentiation.
Amit
We've seen a flavour of what you're building. YCombinator created the SAFE which has been used by thousands of startups, and Open VC created a bunch of documents for VC firms. There are like 100 different types of legal documents, so why not have a platform for all of them. And it looks like you're also creating it as a system of record, keeping track of when agreements are expiring and flagging other issues. Do you think one day you'd also extend this to actual litigation support, like when a company gets a scary notice from another company, gets frightened, reaches out to a lawyer who charges $400 to $500 an hour, and it's hard to know if the lawyer is trying to solve the problem or billing time?
Nick
Yes, 100%. We don't publicise it yet, but we will soon. We have an email assistant at assistant@git.law. You can already forward whatever notice you receive to that address and the AI assistant will review the email or document and get back to you. And yes, we are building with lawyers to build the human-in-the-loop part. The lawyer also reviews the AI's work, and the AI can help judge the right price for the level of review needed. You'll be able to choose the level of lawyer review you want, like self-serve, basic, or premium. The future is something like: do I want to pay $200 an hour and have someone review it for $50 what the AI did, or do I want to pay $1,000 an hour and get the best lawyer in the country for $500 for that document? Depending on how important the deal is, you'll choose accordingly. And often it will probably be the $50 option. Sometimes it'll be the $500 option because there is real value in paying someone a thousand bucks an hour when they've got so much expertise on that specific type of contract.
We also have Zapier and a public API going live soon. So you'll be able to connect it into your other apps. When your CRM notices something happens, create a document automatically. There are 8,000 apps on Zapier.
I'm really looking forward to the agent-to-agent communication era. I think we're already there in many cases but it's not yet common or used by everybody. I believe we'll all be doing agent-to-agent communication where our agents on our behalf will be communicating with other agents. In that world, our agent will need the law firm that our agent uses essentially. We're not a law firm, but we'll be the platform through which you connect to a law firm, and which also automates a whole bunch of stuff along the way, so that when the human at the law firm actually sees it, it will be condensed into a simple easy to understand brief with a document to review and all the track changes already there. That's the future I see. For businesses and individuals and enterprises. GitHub is used by everyone from individuals up to the largest companies in the world. I see the same kind of future for GitLaw, but it'll take us a while to get there.
Amit
Do you also see a world where all the apps become like an app store? So GitLaw has an app on ChatGPT, and if I'm reasoning with ChatGPT around a new customer we've onboarded, it creates an agreement using GitLaw in that flow?
Nick
Absolutely. Wherever people want to use it, 100%. An app on ChatGPT, an app on Claude, an MCP that other AI agents can discover in MCP marketplaces and automatically create an account in and connect and do everything. We'll just go where the space is going. What I've just described is already happening, so it's very likely to go in that direction at least in the next few months. On ChatGPT specifically, we'll probably end up having to pay for ranking in their store. Interesting deja vu moment coming there.
Amit
Realistically, do you think the monetisation model or how we charge customers will change for AI companies? Today your subscription looks very much like a SaaS subscription, but will it change into a performance-based or outcome-based pricing model?
Nick
It kind of is usage based already, because we charge you for credits upfront which is basically the same tokens we are paying for. So it's usage based but with tiers of free or $20. Do I see other pricing models? Yes. I think especially when it comes to looping humans in, paying per hour but a fraction of current rates will be a model that appears. But I'm not actively thinking about any pricing changes in the near future. I can foresee it though, because it's quite natural. We already pay for legal services on that basis. People already prefer fixed price over the "I don't know how much it will cost" model. When it's at our level of cost, whether it costs 50 cents or 60 cents to process that document, who cares. But when you start adding a human in the loop and the difference is between $60 and $1,000, then people care. So yes there's lots of scope for that kind of model, but right now I'm not actively working on any pricing changes.
Amit
Yesterday I was at a fireside chat at UC Berkeley and the lecturer was talking about Ben Horowitz's Hard Things About Hard Things. There's a chapter on the struggle, basically how all the fires are burning, everything is stacked against you, and you still keep going. You've done multiple startups. What keeps you going, and what have been some of those near-death moments?
Nick
With Whisk, on the face of it we were successful. We got an exit to Samsung, we raised some money. The truth is we raised three rounds of financing in 2012 to 2013. Then we basically ran out of money. We almost shut the business down. We went from 20 people to two. We were extremely close to death and we pivoted the product three times. So I'm very experienced in failure.
What keeps me going? I find personally that exercising regularly helps with your state of mind. It's hard sometimes when you're stressed and worried you don't have enough time in the day, everyone's doing 996, working all the hours of the day. But I do think it's important.
Much more important for me though is my wife. I have a wife and two kids. Spending time with them, her support, her belief, and her understanding that everything might fail has been mentally important to me throughout my journey. Everyone has a different thing, but for me it's my wife and my family.
And I have to say, I'm an exited founder, I've made sufficient capital that I don't have to worry about money, and yet this is some of the most anxiety I've felt. I did not expect it. The space is crazy, and I think not enough people are talking about how hard it is to be a founder right now. I've got a kid, I've got a family, and people talk about 996 hours in San Francisco. It's hard. Finding your way of surviving is important. I think luckily I've always had a natural willingness to stick with it. And my family is my secret source.
Amit
Well said. Thank you so much Nick, I found it super useful and very interesting. A lot of great stuff shared today.
Nick
Really enjoyed the conversation. Thank you for the time.
End of episode · Ep #14