How to Build an AI Moat for Your Business (Without Consultants)

Sean Weisbrot: AI isn't just the future, it's your moat. In this episode, I'm joined by Alberto Rizzoli, the co-founder and CEO of V seven Labs to talk about how smart companies are building defensible advantages with ai. We cover why relying on consultants isn't enough, how to start experimenting internally, and the real world impacts on AI agents, on workflows, cost, and job security. If you're serious about future proofing your business, you don't want to miss this interview. How can business owners look to build AI moats around their business in 2025 and beyond?

Alberto Rizzoli: It's a question everyone's asking themselves, and at the moment, AI is changing the landscape across the legal world, the accounting world, the insurance world, any, any company that has a back office and that processes a lot of documents, has this, uh, uh, looming threat of AI potentially replacing the work of many, many people. And the best way to resist that and to create a mode is to start building with it. There's a certain, uh, clarity that you get when you start developing your first AI agents within your business, and you actually see what they're capable of, what they're still not capable of. And that actually opens up your idea of what an AI strategy would look like if you're a business owner in a law firm, um, in an insurance company, in a financial services company, or a PE firm. And, uh, rather than spending millions of dollars in consulting to be told what AI will bring to your business, I always recommend people to just build a proof of concept to something. And it's going to suddenly bring Clari clarity to the C-suite that AI can understand all of your documents, that it can complete the work of analysts, but only to a certain degree. And understanding that degree also can bring some excitement as to. Hey, we can actually serve five times more customers than we do today if we properly adopt this technology. And then you start doing sort of an actual organic digital transformation of the business rather than something where you pay someone from the outside to come and tell you, uh, how you should adapt to the times, which, uh, is usually a recipe for disaster. If, uh, the leadership isn't really gung-ho about the adoption of a technology from the start.

Sean Weisbrot: I think one of the reasons why companies look to people outside is because they lack the expertise to be able to know what it can do for them and what their first step should be. So if your message to people is, don't hire someone from outside, test it yourself. For someone who may not have the expertise in house or as a CEO in particular who may not have a technical background, how can they take that first step if they don't know what that first step is?

Alberto Rizzoli: First off, there's a lot of products to get started with that don't require any technical knowledge at all. Just, uh, using, uh, opening ID research, which is available on chat, GPT, um, is a good first step. Developing your own custom GPT is a good first step. And this, uh, this can be done by a 12-year-old. That can also be done by the CEO of a, of a P firm. Um, on, on the second hand, there's always someone that loves to tinker in almost any business. Okay. They might not be someone that's ready to start adopting an AI product like V seven or like many of the other products that are out there. Uh, but these companies can usually help you with the first implementation of AI agents. And they, they usually have a small professional services component to it that helps out. And then on a third note, whatever recommend, recommend is not to hire an external firm that will effectively, um, replace inefficiency with inefficiency. But rather to, uh, adopt software and then to start considering hiring people that are AI enabled already. Um, I think if you are hiring for paralegals today, you would be much better off hiring a paralegal that already knows how to use AI tools for their own job than someone that is inept to it. That first person will also be this driving force for change. And they will start to immediately think, Hey, this is a task I can do with ai and this is a task that I need to do manually whenever something is delegated to them. And everything that gets done with AI has 10 times less cost and a hundred times more speed of execution. And so that brings some advantages. Um, I do think though, that I, I don't, I don't think every business has to become a tech business or every, especially your services business, but they do have to become tech enabled if they want to compete in the future. Or at least AI enabled, especially if it's any form of knowledge work.

Sean Weisbrot: So I, I love how you suggested looking for people that are already tinkering with it. I've been playing with AI for at least eight months now on a daily basis using different tools, constantly looking for new opportunities to automate what I'm doing. I'm also using AI to build tools for myself to make my own work, you know, more automatable. I have a friend who. Uh, she had a job as a quality assurance person, so a QA tester man, manual tester for a software. And she lost her job with the company. And then she ended up finding another job with the company after a few months of, of looking, which is quite normal. But in that time, I was pushing her really hard to figure out how to use ai. 'cause I said, look, your job is not safe. I go, I, I don't know why you lost your job, but any job you're gonna get in the future. If you're not using AI right now to help you do your job, your job will be not safe. Please take this time while you're looking for a job and figure out ai. She didn't, she got this job and finally something happened that caused her to wanna just try out what would happen to, to automate parts of her job. And she, she did a test and she was amazed with how quickly she was able to figure out how the AI could help her to automate her job. And now she said, I'm never gonna go back. I did something in 10 minutes that would've taken me a week to do. She said, I'm gonna be the most efficient person in my entire company because no one else is using AI yet.

Alberto Rizzoli: That's the right mindset, I think to have something that we've been all afraid of is that this new AI revolution is going to be worse than the industrial re revolution, as in it's going to leave us without jobs rather than just make everything more efficient. Um, but there's also this interesting trend where even after the, the, the latest industrial revolution demand just, uh, continued to. Stay abreast of supply. We now have materials that we treat with throwaway mentality. We created a world of disposable material for better or for worse, of course. Um, we have food that we leave on the table and we're okay if it, uh, molds up in a refrigerator, uh, at least in most parts of the developed world. And that's something that will be unheard of before, you know, the, the, the. The industrial revolution and the, the invention of modern machinery. And in a similar way, we're now creating a world in which these intelligence resources are available on tap. And what this intuitively tells us is that, hey, all these people is, are gonna be out of a job. All of these lawyers are gonna be out of a job. Um, all of these developers are gonna be out of a job. What ended up happening is people are creating 10 times more websites, more products on vibe, coding apps like lovable. Uh, or V zero. Um, everyone is able to enhance their non-digital business with a digital interface that has AI enabled to it. So we're really pushing also this, the supply side and, and seeing that there's still a lot more demand than we expected for all of these things that we're currently doing manually and the right mindset to have. Uh, because a lot of the shifts are gonna happen in the short term. There will be some, some painful shifts happening at companies that, that don't adapt is to be like your friend. And ask yourself, I'm a QA tester. I'm doing this work manually where I'm clicking on a UI to, to find out if there's any issues whenever it gets deployed. But I can be a, be an aa, an AI enabled QA tester, and I can have five agents that I configure and maintain over time that are continuously testing this software and that person on the job market. Is extremely valuable and extremely in demand. So there's also an opportunity of making out of a, a threat or a, a difficult situation, an enabling one by saying, okay, I am the person that is going to bring this new technology into the firm. I know the rules of it because I've done it manually, but I also know how AI can use it. Or at least I know other businesses that I can work with to make, uh, AI actually do this job at a much bigger scale. And, uh, we're starting to see these, these champions of AI also within businesses. They're the, um, you know, scrappy analysts that is saying, Hey, I'm looking at all of these investment memos every day. It's boring. All this financial information and yeah, I get paid well, but you know, is this is eventually gonna be done by ai. Now I want to be the person that develops and develops is a big word here. You don't actually have to write any code for these things. Uh, but that is effectively configuring these AI to do my job. And now my job is that I, I'm effectively a chief of staff of Theis and people come to me whenever they need something delegated, but I don't have to spend my Saturday night doing it. The agent is gonna do it while I'm at the pub. Um, and, and so there is really a way out of this that, uh, requires embracing the technology. Um, and luckily it's never been easier to learn it. Um, in a way it's a compounding story where if you need to learn anything about how to enable ai, you can actually just ask AI about it. Uh, and even chat, GPT or Claude will tell you how to get started on something.

Sean Weisbrot: That's a really good, uh, point for me to start from is that I currently use Gemini for a lot of things. Like it's my chief marketing officer, my chief product officer, my chief revenue officer, my chief customer success officer. It, it's constantly giving me advice on different situations that I find myself in and how to handle it where. I would need to have at least five to 10 people on a team who I'm constantly going to advise, which then wastes their time, prevents them from doing their job. But if I'm going to the ai, the AI has nothing better to do, but wait for me to ask it something so I can say, Hey, I'm in this situation, what do you do? And it tells me and. One of the things that I'm trying to automate is the asset generation for my own podcast because, you know, there's, uh, audio files, there's transcripts, there's uh, images, thumbnails, all sorts of things that you have to create and it takes a lot of time and I don't have someone to help me do it. So I. Instead of hiring someone I am having, I'm building an AI software to automate it for me. But I didn't know what exactly I needed to do and I didn't know exactly how to build it. And so I said to Gemini, Hey, interview me and I'm gonna tell you what I need, and then you can give me all of the feature specifications, the user stories, the technical documentation, even down to. All of the database tables and the relationships between these tables, and I'm gonna input this into Cursor so it can start developing this software for me. And it did it, it took about half an hour to have this conversation with Gemini and then I put it into Cursor and I said, analyze this document and turn it into a phased approach to develop this software. And I said, okay. And now we're building it Cursor and I together. And you know, years ago when I was building my last software, it would've taken three to six months to come up with all of that stuff. Yeah. And between 3, 4, 5 people all disagreeing about what to develop, when to develop it, how to develop it, how they relate to each other, should we even develop it. And it's just amazing. So, uh, I, I haven't gotten into agents myself because I'm not sure exactly how I know about N eight NI know about make, but I haven't. Gotten into it because there's so many people that say, oh, well you need to code this, or, oh, you don't need to code this. Or, oh, you can copy this. There's just so many different people saying so many different things and trying to teach you their own way and make money doing that. And so I'm confused and so I'm just stuck at the beginning and I just say, forget it. I'm gonna keep doing what I'm doing. What I'm doing now is working for me is do you find this to be the mindset of some of the people you talk to of I'm just stuck, I don't know what to do.

Alberto Rizzoli: I think it's very common. Uh, there's. There, there's a lot of hesitation towards agents because we see them as this very open-ended, uh, creature that may or may not return with results on the other side. Mm. Uh, yesterday we saw OpenAI release, um, an agent, uh, feature that allows g PT four effectively to browse the web and make purchases for you and or book a holiday for you. And that's a really tall order unless it knows exactly your preferences of location, hotel appearance, vibes, et cetera. It's really hard and uh, we're, we're not quite there yet, in which you can just tell, um, an agent, Hey, go buy me some shampoo, because I've run out and it knows exactly what you've bought it. It's, it's more a little bit more complex than we think. It can't just easily look into your, your purchase history. I mean, it could, but it, it's still a, a pretty technically complex, uh, system here. So the best, there's effectively two areas of agents that are becoming very popular. There's, uh, consumer or solo, uh, professional agents, uh, which are developed, uh, in a very, uh, similar to, to Zapier or N 10 as you mentioned before. So, uh, with small bits of intelligence in the middle, but it's largely an API to API connection where you say, Hey, I'm gonna get the recording of the podcast into, um, a video editor. The editor is gonna detect some clips and segment it, and then another node is going to upload them to YouTube or to TikTok. And that's something where you're defining a fairly systemic process. You're not just telling Gemini, go and make me famous, and it figures it out for you. That's still not quite there yet.

Sean Weisbrot: Mm.

Alberto Rizzoli: And in our world, within V seven, we develop agents for highly professional. Um, workflows, uh, the, uh, the, the process of evaluating an an m and a deal is really complex. It involves hundreds of documents and reading through tons of transactional data and doing due diligence on people that have information on the internet. And so these agents are still following a number of steps, but the, it justifies the time involved in actually developing this, this workflow. Which is not five minutes like you, you do with the, with, you know, writing a prompt for a Gemini and maybe tweaking a couple of times. Uh, but it's something that requires a few hours of building a few additional hours of tweaking and then some maintenance over time whenever the rule sets change. Um, so the recommendation I have today for agents is if your revenue depends on it, then consider building an agent because it's going to help you grow much faster. If you have someone full-time that is doing a manual process in front of a computer, definitely get an agent for them because, um, it's gonna help them as a person to be enabled in technology. Um, or you could be this person as, as the listener, um, and then everyone else that works with you is also going to be grateful because it's a shared resource that can also leverage. Um, but if it's for a one-time use, if you just wanna book that one holiday in Corfu, um, it's not always that convenient to develop an agent that knows all of your preferences because you're gonna have to write all these down just for that one interaction. Um, obviously if you are a travel agency, that's a different story. So agents a little bit more of a lift today to build, but absolutely buildable also by non-technical users to automate something that you do every day.

Sean Weisbrot: So I do want to share a quick example, uh, related to travel and AI as you were talking that made me think of this. I'm going to Japan in October with my brother For 17 days, my brother spent a month preparing everything he could, speaking to people who've been to Japan to put together an ideal itinerary for 17 days. He told me nothing of his plans the day before, he wanted to have a call about it. He said, are you ready? And I said, sure. I had done zero. I told him I had done zero. I've been busy planning my wedding and everything. So the next morning before the call, I spent four hours talking to Gemini, and Gemini came up with an amazing plan for me for this trip. I knew nothing about any of the places, but then I did some research, uh, you know, manually and talked to a few people and, and made sure it lined up with something that would fit my personality. And then I had the call with him and he said. Did you do any preparation? I said, yeah, I, I just did. He's like, okay, great. Why don't you tell me your plan and we can go from there. I'm curious to see what you came up with, seeing as how you only spent a few hours on it. Okay, fine. The AI came up with the same exact plan. Hmm. That my brother did. It took me four hours. It took him a month. So it was funny to me because we agreed upon what we wanted to do. This is our ideal plan of Japan, which is very different than most people do, especially for the time that we have. And so it was unique and, and exactly what I wanted, but it didn't know anything about me. I just said I, I did a few prompts. I said, this is my budget. This is how long I'm going. This is where I'm starting. This is where I'm ending. This is what I want. Give me the ideal plan for that. And it gave me what I think is a fantastic itinerary. I'm super excited

Alberto Rizzoli: when you unpack that the, there's not that big of a secret behind it at the end of the day. Um, if, if you think of what your brother probably did, he went on Google, he inputted parameters that are somewhat equivalent that a model could think of. Uh, maybe restaurants within a particular budget in the city that you're going to, that have a dish that you love. And those are things that, uh, we've been able to index fairly well already in search. And AI is simply able to replace your brother's manual tasks there, uh, because it's already read through all this information and give you a, a, a condensed answer in a much faster way. Um, and luckily we have fairly common preferences. You know, if a, if a restaurant is great, if a scenic attraction is great for you, it's probably also fairly good for me. And, and so in, in some cases, you don't have to put a ton of effort into these, uh, these executions. Where it does differ is if you either are working on something that is pretty unique to your business, maybe you, um. You are, you know, you're in, in a particular medical profession or you are working on a, a construction planning set of documents and there's really not that much information on the internet about something equivalent, then that's where AI gets a little bit lost and developing an agent allows you to give it the necessary information so it can actually do the job. And I think we're, we're, we're going to see. More and more coming from the core models like Gemini, for everything that is in common amongst all of us as humans. What's the best restaurant in London for eating a rotisserie chicken? I'm really hungry right now. At lunch, it's gonna give me five that I'm probably gonna love because other people have also polluted the internet with information about this, and the models have learned about it.

Sean Weisbrot: Sure.

Alberto Rizzoli: But if you're looking for something a bit more niche, you're. You're looking for suppliers, uh, for a particular herb that may be found that's gonna be a little bit harder for the models, but eventually they're going to get there. But an intentional behavior is better because it needs to reason through your task it might ask you some follow up questions to learn and then go somewhere outside of the first couple of pages of index Google searches to go and find that information sometime.

Sean Weisbrot: So the thing about that. Specific example for me was I had said, look, I want an off the beaten path.

Alberto Rizzoli: Hmm.

Sean Weisbrot: I don't want to go where everybody goes. And it took me through a part of the country that people don't normally get to. So we're starting in Tokyo, ending in Osaka. Typically people go from Tokyo to the on the train to Kyoto and from Kyoto to Osaka. Sure. We're starting in Tokyo. We're gonna do Kyoto and Osaka. That's fine. But. The most of, most of the time we're going from Tokyo all the way to the far West and then South Tokyo, Kyoto. So there we're reaching an entire part of the country that most people don't touch. Mm. And that's what's cool for me is that it gave me the thing that it knows everybody loves and then gave me things that it knows I would love, because there's a few days of hiking. And, uh, these other kinds of really beautiful experiences with these bamboo forests and things that, like you just don't find anywhere else. And so that's what I said was, I want something that's really off the beaten path. And I want something with nature because I know that Japan is supposed to be this gorgeous country with tremendous amount of nature and, and these beautiful landscapes, and that's what I wanna experience. I wanna maximize my time outside of cities.

Alberto Rizzoli: You probably wouldn't have known easily what to Google search to get those answers? No. You know, places in Japan off the beaten path isn't exactly a result that will give you consistent, uh, uh, returns. And the good thing about AI is that it can look into its own brain. It's into its own embeddings and, um, that gives the data effectively of, oh, these are. These are less searched for locations. These are less indexed locations that have less information about them online. Let's prioritize these instead. Um, and it comes natural to it because that's effectively already inside its head. Uh, a bit like how a human would think of immediately. Okay? I'm not gonna send you to a chain restaurant, but I'm gonna send you to the local, uh, place that, uh, is run by a family and it looks like someone's apartment inside. Um, it's, it's more natural for them to immediately start thinking about these options.

Sean Weisbrot: And I often will follow up with the AI and go, Hey, that's a really great suggestion. I really appreciate, so I'm, I'm telling the ai you've done a good job. You've given me what I want. And that, that helps it with its training so that it gets better.

Alberto Rizzoli: To some degree it does. Yes. Um, it's very tricky to use human feedback from users to improve its results, but, but it is, it is a useful way, and that's why we had these thumbs up and thumbs down. On every consumer app. I'm not exactly sure how, uh, this gets standardized over time. It seems like a gargan actual data task because, uh, obviously preferences might differ a lot and some queries are very opinion and some are more universally acceptable as true. Um, but yeah, there's a lot of effort in making these models helpful to the average user. Um, and it's actually really hard to, to make them helpful for, for people with like a particular niche opinion. They, they need to be prompted in a certain way.

Sean Weisbrot: You said something that was interesting. I was under the assumption that all of these opportunities to thumbs up or thumbs down or to think them would, would help with its training, but you're making it seem like it's not such a big deal. Why go through the exercise or go through the effort of. Wanting the user to provide feedback. If it's not helpful

Alberto Rizzoli: data coming from users, uh, can it only re represents what your already happy users, uh, want? And it also represents the bulk of, um, of fairly simple queries, uh, most of the time. So if you look at, uh, if you think of Instagram as a, as a proxy. What it shows you in the explore page is what the majority of people that have looked at similar imagery profiles, hobbies, as you have also looked at, um, and that can be good for developing engagement, but for developing intelligence, what actually needs to be done is to go and find the gaps in the knowledge of the actual model. And it's harder for non-expert to evaluate that. We don't know what we don't know, and most users don't know what they don't know. And yourself not being a Japanese native, um, you actually don't know if those places are really the best. I know your brother put some effort into researching it, but ideally, we're now at a point in which these models should learn from an expert Japanese travel guide, from an expert, um, cuisine, uh, uh, you know, reviewer. And, uh, this is a, an ongoing effort where open ai, philanthropic, uh, Google's teams are paying a lot of money for experts to give the AI their own, uh, personal knowledge. And this can go all the way to math PhDs that are teaching AI how to perform really complex mathematical problems. In part because everyone wants to be at the top of math leaderboards. 'cause it shows that, uh, your research team is the best and it increases the value of the company itself and makes people trust the model more. Um, but at the end of the day, we're in this really eerie time in which Gemini, um, oh three and oh four or Cloud four are smarter in most topics than you and I. And so there's little that they could learn from us other than I liked this answer, which is subjective. It's likely going to, um, it's likely going to not teach the model anything new other than I made Alberto happy for this one. Um, but is it a four out, a five answer? And if it is a four outta five to an expert, what would it take for it to be a five? And that's, that's a harder task to come up with and that's what these AI labs are thinking about now because they all want to have the best model. If you're not in the top five, you don't get really as much business.

Sean Weisbrot: What have you learned in being involved with AI and running this business? What's the most important thing?

Alberto Rizzoli: There is a significant amount of resistance, uh, based on geo, so we see. There, there's a significant difference. We see the US being very happy with the AI adoption. Its professional services. We see us private equity firms saying, I'm really tired of looking at all these deals, and I want the business to grow and I want this to be competitive. So let's try and use v sevens AI agents for this. Let's just give this a shot. We waste so much money on pointless things, and if this thing really is revolutionary, we should invest a little bit of our budget into AI experiments. In Europe, we see about half the adoption rate. So for the same amount of revenue generated European businesses are, uh, half as likely to, to make a purchase and to adopt. This cultural, uh, difference is a little bit worrying because of what I said earlier, which is that the actual AI pros comes from using it, not reading about it, not hearing it about it at the dinner table, when you start using ai, and when you start using really any technology. But, but AI in particular, it opens up your mind to the possibilities. You wouldn't have just outta the blue, decided to go and ask, um, Gemini to plan your Japan trip if you hadn't used it a couple of times before. And then when you were planning your Japan trip, you thought, I'm gonna ask Gemini for this. And I knew exactly how to prompt it and how to use it, and it saved you a month of work, or it would've saved your brother a month of work. That is a mentality that needs to make its way into the professional world. Um, so the people actually start to immediately grok what AI should be doing and what humans should be doing within the business. And there's a lot that AI should be doing within the business today.