Andrew Wilkinson: AI Agents run my business and life

0:00Today's episode is an inside look to how Andrew Wilkinson is using AI agents to run his personal and business life. This is a guy who has a great setup. He shows us exactly how he's using things like OpenClaw and AI agents to be more productive and to make more money. How he uses it with his family office. How he's using it to start a startup. How he's using it to think about where the next generation business is and where he should be putting his money. This is an inside look, and I don't think he's shared this anywhere with how he's using AI agents to run his life and his businesses. By the end of this episode, you're going to have your creative juices flowing.

0:41I certainly did, and I can't wait to see you in there.

0:52Andrew Wilkinson, I think third time on the pod. Welcome back. Good to have you. Andrew, by the end of this episode, what are people going to learn? Oh man, they're going to learn that OpenClaw is hard, but amazing when it works. They're going to learn whether they have ADHD or OCD. They're going to learn how to build a crazy vector database about their complex company. And they're going to learn a few prompt tricks and whatever random other stuff we end up stumbling into. Okay, so for those of you who don't know, Andrew's a very curious guy and he gets deep into rabbit holes. So it sounds like what we're going to learn is a lot about how you're using AI in your day-to-day.

1:42The good, the bad, the ugly, and everything in between. Yeah, let's do it. All right, let's get into it. I would say I had this crazy moment like so many other people in December of last year. where I played with Cloud Code and all the AI tools like Replit and stuff. And they could kind of do stuff. It was interesting, but it wasn't there yet. And in December of 2025, I just literally, I feel like I started shooting heroin. I started waking up at three or four in the morning, rolling out of bed with a big smile on my face and literally just sitting in terminal and clawed code with 10 tabs open.

2:27And I've been doing that ever since. And my goal, you know, I've had a bunch of different things I've been trying to do on the productivity side. Like I was basically trying to build an OpenClaw agent before OpenClaw came out and doing a very poor job of it. So I was excited when I was able to do that. But I mostly have been excited as a business person. And for those of you, the listeners that don't know, I basically buy and start businesses. I mostly start businesses as a hobby for fun, but I buy a lot of businesses and I'm always astounded by the amount of administrative burden in a SaaS company.

3:09If you think about how it scales, support, accounting, there's all these very boring rote types of tasks that are now able to be automated. And so as a test, before I started doing this in our operating businesses, I basically built a business, a SaaS business, and then I've been running it entirely autonomously using OpenClaw. So that's probably a cool place to start. And I just want to say like with OpenClaw, I basically have been, we referenced heroin a moment ago. Um, you know, I have never done heroin just in case anyone's concerned about me, but, um, apparently when you do heroin, the first time is the best time and they call it chasing the dragon.

3:57You're trying to get back to that special moment. And I had this special moment. I was traveling. I went to Arizona on a trip and I forgot to bring my laptop. Um, but I'd set up an open claw agent on my computer and I basically was able to run my entire business using open claw in the back of Ubers. Like I was going to this conference, it was super, super busy. And I was astounded by how competent it was able to be. Um, and that nobody picked up on the fact that, you know, every single email that I wrote was being written by open claw. And so I've been chasing that moment ever since. Uh, and the problem is that I feel like I'm spending 50 % of my time debugging it and 50 % of my time, maybe 30 % of my time making it better.

4:44And then 20 % of my time being productive. So it's like the classic productivity treadmill. Um, but we have been able to build some pretty interesting stuff. Um, so let me, let me screen share with you and I will show you this app that I built. Um, so this all started because I, uh, I wanted to build a custom GPT or chat GPT project for me and my girlfriend's relationship. The idea would be that we would be able to query it. Hey, we're having a fight about X, Y, Z, which we do. And I asked it, I said, hey, if I was to do a bunch of psychological tests, which ones would you want to see? And it gave me this list of like 15 different screens that it wanted me to do.

5:29And so I went to Cloud Code and I said, hey, can you just put all these into a multiple choice test? set up scoring and then put them in a json file and so both of us did that it took about 40 minutes and we put the json files into chat gpt and we said without knowing anything about this this couple um tell us about their relationship and we read it line by line and like our jaws just like dropped like it nailed every single fight that we have every single issue at home and so I saw the power of these multiple choice personality tests and so I saw the product opportunity and I was like I'm going to build a business out of this and so I started this thing called Deep Personality.

6:12Let me just share my screen here. Quick break in the pod to let you know about a free workshop I'm doing that answers some of your biggest questions. The first is in the AI age, what are some categories that are ripe for startup ideas? I'm going to outline a bunch of those different categories, the ones that I think the most opportunity is. Second is there's a million AI tools out there. Which are the ones that matter when I'm trying to build a business? I'll walk you through that. And lastly, how do you actually build a business with some of these AI tools? So I'm actually going to show you live how I do this with my ideabrowser.com team.

6:51And I can't wait to see you there. It's this Thursday at 12 p.m. Eastern. and if you go to the show notes, if you go to the description, you can click a link, RSVP, and learn a few things. I hope it'll get your creative juices flowing and I'll see you back at the pod. And you could have, you own an agency called MetaLab, you could have called out MetaLab to go and build this for you, but you decided, I'm going to do this all myself. I'm going to actually productize this myself, I'm going to design it myself, I'm going to build it myself. Why did you make that decision? Well, I think I love people like I'm very extroverted, but the worst part about business is people.

7:32You know, if you think about using a let's use a screenwriting analogy. So in Hollywood, there's all these brilliant screenwriters and they write these incredible scripts. And so they have this idea and they know where they want it to go. Right. So let's say they start here and they want it to go here. They want to make an incredible feature film in between all in between. the script and the feature film are a hundred people that they have to convince 50 million dollars. They have to raise a million middlemen and all that. And I think that creativity is just compromised based on how the number of people between your vision and execution.

8:11And so I think writing for me has been a very pure pursuit where I can think about it and I can write it and I can work with AI to make it better if I need to. Um, but it feels, it feels like so clear, such a clear, um, execution of your vision. And I feel like vibe coding is that, um, like finally I can do absolutely every part of it and I can do it at my own pace. You know, if I was using employees, um, they're unreliable. They don't see my vision. They don't understand every aspect of, um, design. So you can have a great designer who sucks at front end or you can have or doesn't know how to design something that'll work in a front end.

8:55You can have a great designer who doesn't know how to write copy. So there's just all these pieces that come together with vibe coding. And basically, like in a few manic days, I was able to build this pretty crazy app. So let me let me show you how it works. So basically, you do a personality test and it takes about 40 minutes. You're just clicking multiple choice answers and stuff. And then you get this report. And so here's my archetypes. It'll be like, you're the blazing architect. Here's your superpowers. Here's your kryptonite. This is just like the basic. And then as you scroll, you can see the AI has taken all of your answers and it's written like a hundred page report.

9:38So this report literally goes on and on and on and on. And it's written like Robert Green. So it's like super impactful and deep and hits you. And it basically walks you through what you're great at and what you're terrible at, what job you should have. Is your relationship working for you? Do you have ADHD, OCD? Are you on the spectrum? Like all these different things. And so you can see, let me get, oh my God, it's so big all the way to the bottom. Um, so you can see like, these are my kind of personality traits, my attachment style, um, my internal family systems, like how much I people please all this stuff.

10:20So I built this thing and usually what happens for me is I start something and then I don't want to finish it because I don't want to hire the people and deal with it. And so what I ended up doing is in this case, just building agents to do everything. And so I can show you where we were using open cloud, we're currently using this thing called Harbor that a friend of mine built. And it's basically a harness for agents. So you can see all the different agents that are running here. So we have a dev, we have marketing agents and then we have a support agent. And basically the way it works is somebody emails support and either support just tackles the ticket or it will actually fix the ticket.

11:08So it'll send it to the dev agent. If it's a P zero, so like a super scary security breach, it'll just immediately fix it and merge the PR. Otherwise it'll, we just wake up and have a bunch of PRs. and then it'll email the person back and be like, hey, I fixed your issue or whatever it is. That works basically perfectly. Like it's shocking. Like support to me is not a job very, very soon. The one that's really exciting is the marketing agent. So we basically have it, it's hooked into post hogs. So it has all the data for the app and it actually manages a meta and Reddit ads account. And so it'll do multivariate testing.

11:51It'll create ad creative. It will set budgets and we basically just message it and we'll say, you know, Hey, can you increase the budget by a thousand dollars or we approve like a big SEO project or something like that. So where we're at with it is like, we've done maybe like$20 ,000 of revenue. Like it's still small. And where I'm really excited to go next is like, what happens when we give it like a hundred thousand dollar a month ad budget and we find ads and creative that actually work. So that's been, that's been insane. And I think that we're probably three to six months out from being able to just hand businesses off to AI to run at least basic businesses like this.

12:39harbor i've never heard of and i see a lot of stuff let me um let me just ask so it's my friend gavin made it yeah yeah to the harbor website um yeah it's harbor it reminds me a little bit of paperclip have you seen paperclip yeah i have so this is it it's github slash geek for brains slash harbor. And he's a friend of mine that I've been working on this with. And it's working really well. We just found that OpenClaw wasn't quite deterministic enough. And the other issue with OpenClaw is because it's a text-based interface, it was a little bit hard to keep track of all the agents and what's actually happening.

13:26So it's worked really well for us. Yeah. I mean, I think what's cool about it is like if you're starting a company and you're thinking about org charts, roles, responsibilities, you don't think about it within text, back and forth, right? You think about it as like, okay, I'm going to spin up this agent or employee to do XYZ and then how can I monitor what that XYZ employee is doing? Yeah, so basically it's just a GUI basically for something like OpenClaw and it's still running on Clawed. But you have all these documents with a knowledge base. You've got databases. You have all your environmental variables, et cetera.

14:09It's very similar to Paperclip. I think the problem, it reminds me a little bit of crypto, where crypto is built by nerds, so they make it nerdy. And when I saw Paperclip, I got so excited because I was like, oh my God, this makes sense to me visually. You have this org chart, but it's clearly built by nerds because it's very confusing. And so my, I mean, my kind of belief is like pretty soon Anthropic and, uh, opening AI are going to launch like basically like CEOs. So you just be like, I'm just going to hire a CEO for my company. Give me your data. I'll figure it out. I'll run it. I'll run the whole business.

14:48And right now the problem I feel, I feel like, um, you know, two years ago it was like uh we're you get you get five minutes with a super genius who has alzheimer's like the context window is so small that you just have these brief moments now i feel like with the the one million context window um they can like maybe remember a day but it's like memento right um and i feel like once the context window is like five million ten million then we'll be able to run entire companies with this yeah there's some people now selling autonomous companies i don't know if you've seen this pulsia is one of those companies i think pulsia work like i i saw that new york times story about that guy who like started the autonomous uh you know what is it like billion dollar business or whatever but and obviously that's real new york Times is writing about it, but I can't parse what's real versus bullshit.

15:47I think, you know, we're recording this April 29, 2026. You cannot run a fully autonomous company right now that, yeah, that is going to get better, that you can just like, you know, go hang out on the beach 24 seven. That being said, I think there's parts of the business that can be autonomous. Support is a good area. But you're not going to outsource product at this point, in my opinion, autonomously to something like Pulsia today. Maybe that changes in six months, 12 months, whatever. But my take is the people that are selling autonomous companies today are just selling a dream. Yeah, I kind of agree.

16:37I think open claw agents are still like zap. They're basically like zap your zaps, but that can make basic decisions and have intelligence. But you still have to tell them step by step, this is how I want you to think. This is how I want you to operate. Like, let's say you hired an intern and you gave them an email address. You would never have to say, you should check your email every 15 minutes and you should reply. And when you do reply, you should think about the following things. It's like you're dealing with a baby, a genius baby, but you have to teach them how to do every single thing. Where I'm excited is when I can just have a model that's just like a CEO model and just delegate to them and they'll just figure it out.

17:25Which I think is kind of what everyone acted like OpenClaw was, but it's totally not. Yeah. I want to go back to the deep personality test. Can I give you unprompted feedback? For sure. So when you were going through this, the content looked exceptional. My issue in general with vibe-coded projects is that because anyone could create apps and products like this, it's almost like you need to have some credibility layer in order to listen to the content. You know what I mean?

18:06So if I was the PM or CEO on this, I would be like, how can I partner with someone who's credible in the space such that when I get this output, this report, I'm going to take this seriously? I think part of the problem is it's anyone who is a psychologist or someone who would be like legit on this would never do it because it's a bit cheeky. Like I'm using psychological screens that are supposed to be interpreted by a psychologist, but I'm not a psychologist. I'm using AI to do it. And I think it's incredibly accurate. Like I've, I've hammered it on a million different models and I've looked at example tests and stuff, but that is definitely an issue.

18:54If there's someone listening, who's connected to like a Jay Shetty type person, definitely let me know because I think it's perfect for that. It really is one of those like one plus one equals five kind of things. But yeah, right now it's what I realized is like I can, when I post this in my newsletter, it goes crazy viral for a bit, but I don't have enough pull in that space where my audience is a bunch of like tech bros. They're not going to, you know, they're not going to share this with all their friends and stuff. It does definitely need something like that. I also wonder, I think about what are influencers, celebrities, micro-celebrities that people look up to?

19:35A couple, even just a couple that has gone through this publicly maybe and have come out of the other side and they're just happier. When people think of a dream couple, they're involved in this in some capacity. that's smart so you and your wife you want to do it yeah yeah my wife is like very like

20:02private you know so I'm more of the public one but I'll see what I can do same as my girlfriend she hates publicity and stuff let's see what else can I show you I mean another big piece is just data So I, you know, I feel really frustrated because at different times I've kind of standardized on different LLMs. Like I'll get into the cloud ecosystem and I'll use the desktop app. I'll use chat GPT and I'll build out a bunch of projects. I've basically just accepted that I'm going to live in cloud code until somebody builds her, which I'm sure is coming in the next couple of years. And so my strategy has basically been to take all the different data pipelines that I have and put them into a centralized place.

20:52And so I have Fireflies record all my meetings. Every night a cron goes in to the API, builds an markdown file, puts it into Gbrain. I've built out vector databases for my businesses. This is a really cool thing. So, you know, a vector database can store, you know, millions and millions and millions of tokens. And you can basically have like a service like Pinecone go and crunch through a large amount of data. And it basically makes it highly searchable for an LLM. And so I can show you an example of that. Let me just switch the screen sharing. So, okay, so here's an example. So like I did, I trained a vector database on my family office, Folly Partners, which is basically just my like personal holding company.

21:47And here I just asked it, what did I say? Breakdown how many minority venture investments I've made and how many are in the money, bankrupt or declined. And basically it went off and did this query. And then boom, I can see that I've made 132 direct investments. I invested 16 million. It's now worth 36 million. And then it can break down all the different write-offs, everything. And so this has been, um, really, really powerful. So like, if you think about, I did it for tiny as well. And in tiny, I have, uh, man, 24 businesses or something like that, tons of historical data. And the challenge when you run a conglomerate is like, I don't know So like, are any of the companies spending too much money?

22:37Are any of the CEOs full of shit? Are any of the businesses trending up, down, expenses are going up? It's just too much data for one person to understand. And so I can't show it because it's all public company data. But basically what I can do in tiny now is I can say, hey, review the last quarter and I want you to give me any icebergs or issues. Or I can ask a simple question like, how many accounting staff do we have in head office or whatever? And I've found this so useful. It's not perfect. It'll often say the wrong person is CEO of a business or the numbers are a bit off. But in terms of being able to be the eye of Sauron inside your company, I think it's incredibly powerful.

23:22And curious, within Tiny, I see how you're using it to interface with Tiny and to be a better chairperson and give advice. But within Tiny, I imagine that there's some efforts to make the company's AI first and stuff like that, right? Yeah, and that's one of those challenges that I think everyone can relate to. What is this saying? Never expect someone to understand something that their paycheck depends on them not understanding. So you have a lot of people that are living in the old world and need to be upgraded in their understanding. And often they're very attached to the old way of doing things.

24:05And so to be honest, like the challenge in tiny is that we are generally pretty hands off and where we're really leaning in is in a lot of the software businesses, because those are the ones that are at the most risk. I think we've talked about this before, but it's not like software is dead. If anything, software is going to thrive way more. But it used to be like the newspaper industry where it's like in order to compete, you have to assemble this incredibly expensive group of people that's very rare to find. Right. You need like great programmers, great designers, et cetera. And that that was like a very finite resource.

24:44So it's kind of like the newspaper industry where you had to be rich. You had to buy like$2 million printing press. And then once you have it, you actually have a competitive advantage. Now I just think software is free. Anyone can make it. Anyone can rip it off. Like my personality app, it's cool. Does it have a moat? Absolutely not. Like maybe if to your point, I had John Legend and Chrissy Teigen backing it or something, maybe there'd be something there. But I think pricing pressure is just going to go crazy because it's like the restaurant industry, restaurants are not a bad business. They're bad because there's too much competition.

25:22So if you're listening to this and someone's listening to this, they're watching this actually and they're saying, okay, I like what Andrew's doing here with this fully advisory thing. I like what he's doing with personality tests. I feel like he has a good sense to where the world's going. you know where what would you know what would you be building now if you're you know if you're you want to be building something it's right where's the opportunities i think we talked about this a year ago and if anything i've become more concerned um when i look around i'm don't get me wrong i'm not concerned i just say it's getting harder and harder and harder like i think we've all been astounded seeing how good Claude has gotten and how much it can do and how quickly it can do it.

26:15And from my perspective, I think that if I was somebody who's obsessed with vibe coding and is 20, and I would say the goal should be like, make one to$2 million by building a product like deep personality or something like that, which is pretty straightforward. Try and make a million or 2 million bucks. And this is such a sad answer, but I'd be like invested in data center stocks or invested in TSMC because to be honest, that's what I'm doing. Like I am, I'm either buying like something with a moat, which is very, very hard to find certainly in technology right now, uh, home services, businesses, like very, very brick and mortar kind of stuff.

26:55Um, but even that I think is at risk with robotics over the next 10, 20 years. Um, but I honestly, I, all I can really recommend someone does is buy Iron and TSMC. Yeah, I think with respect to the GUI, the graphical user interface, it just feels like I think about it as the hour of progress. The hour of progress or deep progress is that the UI is going away. and I don't know about you but I'm spending more and more of my time in codex. I'm spending more and more of my time in cloud code. Those have become my operating systems. So I'm spending less and less time in apps and it's easier than ever to create apps.

27:42Therefore, yes, pricing pressure, competition, harder to get mind share. Someone asked me, is Adobe fucked? And that's an interesting one because I'm a pretty avid user of Lightroom and I love, I've got a Leica, I love editing photos and stuff. And while I do put my photos into Gemini sometimes for like recoloring or retouching or something, I still really enjoy the process of like moving the sliders and making it exactly the way I want it. And so I think, you know, it's a question of like, does this being the standard matter or will someone just vibe code or an even better Lightroom and everyone will switch to that because they're, you know, a big, slow incumbent and someone else will do it faster.

28:29I can't answer that question very well. All I can say is software is a worse business now than it was five years ago. I mean, look at Constellation Software. Their stock has been cut in half. Like that, that's crazy. Like I, and I don't know if that's correct because they own like the most boring software businesses ever, but, um, yeah, dude, it's like a brave new world. And it's so depressed. I hate to give such a depressing answer because your show is all about entrepreneurship. And like, I just want to say I'm more excited than ever. There's never been a better time to be an entrepreneur and build stuff.