YT YouTube2Text

YouTube2Text → Transcripts → AI Agents run my business and life (Andrew Wilkinson)

AI Agents run my business and life (Andrew Wilkinson) — Transcript

by Greg Isenberg · 8,588 words · 1,253 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Today's episode is an inside look to how
0:02Andrew Wilkinson is using AI agents to
0:05run his personal and business life. This
0:08is a guy who has a great setup. He shows
0:10us exactly how he's using things like
0:13Open Claw and AI agents to be more
0:15productive and to make more money. How
0:18he uses it with his family office. How
0:20he's using it to start a startup. How
0:23he's using it to think about where the
0:25next generation business is and where he
0:28should be putting his money. This is an
0:30inside look and I don't think he's
0:31shared this anywhere with [music] how
0:33he's using AI agents to run his life and
0:36his businesses. By the end of this
0:38episode, you're going to have your
0:40creative juices flowing. I certainly did
0:42and I can't wait to see you in there.
0:47>> [music]
0:52>> Andrew Wilkinson, I think third time on
0:54the pod. Welcome back. Good to have you.
0:57Andrew, by the end of this episode, what
0:59are people going to learn? Oh man, um
1:02they're going to learn that Open Claw is
1:05uh hard um but amazing when it works. Um
1:08they're going to learn
1:10uh whether they have ADHD or OCD.
1:13Uh they're going to learn how to build a
1:15crazy vector database about their
1:17complex company.
1:19Um
1:19and
1:21they're going to learn a few prompt
1:22tricks uh and whatever random other
1:24stuff we end up stumbling into. Okay, so
1:28for those of you who don't know, Andrew
1:30is a very curious guy and he gets deep
1:33into rabbit holes. So, it sounds like
1:35what we're going to learn is a lot about
1:38how you're using AI in your day-to-day.
1:42The the good, the bad, the ugly,
1:45uh and everything in between.
1:47Yeah, let's do it.
1:49All right, let's get into it. I would
1:51say I had this crazy moment like so many
1:54other people in December of last year
1:56where I played with Claude Code and all
1:59the AI tools like Replit and stuff, and
2:02they could kind of do stuff. It was
2:04interesting, but it wasn't uh it wasn't
2:06there yet. And in December of 2025, I
2:10just literally like I I feel like I
2:13started shooting heroin. Um I
2:17started waking up at 3:00 or 4:00 in the
2:19morning, rolling out of bed with a big
2:21smile on my face, and literally just
2:24sitting in terminal and Claude Code with
2:2610 tabs open. Uh and I've been doing
2:28that ever since. And my goal, um you
2:31know, I've had a bunch of different
2:32things I've been trying to do uh on the
2:34productivity side. Like I was basically
2:36trying to build an open claw agent uh
2:39before open claw came out and doing a
2:41very poor job of it. So, I was excited
2:44um when I was able to do that. But, I've
2:45mostly have been excited as a business
2:47person um for those of you uh the
2:51listeners that don't know, I basically
2:53buy and start businesses. I mostly start
2:56businesses as a hobby for for fun. Um
2:59but, I buy a lot of businesses and I'm
3:02always astounded by the amount of
3:05administrative burden in uh you know, a
3:08SaaS company. If you think about how it
3:10scales, um you know, support,
3:13accounting, there's all these like very
3:15boring wrote um types of tasks that are
3:18now able to be automated. And so, as a
3:21test before I started doing this in our
3:23operating businesses, I basically built
3:26a business um a SaaS business, and then
3:29I've been running it entirely
3:31autonomously using open claw. Um so,
3:34that's probably a cool place to start.
3:36And I I just want to say like
3:38with open claw, um I basically have been
3:42we we referenced heroin a moment ago.
3:46Um you know, I have never done heroin
3:47just in case anyone's concerned about
3:49me. But, um apparently when you do
3:52heroin, the first time is the best time.
3:54And they call it chasing the dragon.
3:56You're trying to get back to that
3:58special moment. And I had this special
4:01moment. I was traveling. I went to
4:03Arizona on a trip, and I forgot to bring
4:05my laptop. Um, but I'd set up an
4:07OpenClaw agent on my computer, and I
4:11basically was able to run my entire
4:13business using OpenClaw in the back of
4:16Ubers. Like, I was going to this
4:17conference. It was super, super busy.
4:19And I was astounded by how competent it
4:23was able to be, um, and that nobody
4:25picked up on the fact that, you know,
4:26every single email that I wrote was
4:28being written by OpenClaw. And so,
4:31I've been chasing that moment ever
4:33since.
4:34Uh, and the problem is that I feel like
4:36I'm spending 50% of my time debugging
4:39it, and 50% of my time, maybe 30% of my
4:43time making it better, and then 20% of
4:45my time being productive. So, it's like
4:47the classic productivity treadmill.
4:49Um, but we have been able to build some
4:51pretty interesting stuff. Um, so, let me
4:54let me screen share with you, and I will
4:57show you this app that I built. Um, so,
5:02this all started because I
5:04uh I wanted to build a custom GPT uh or
5:08ChatGPT project for me and my
5:10girlfriend's relationship. The idea
5:12would be that we would be able to query
5:14it, "Hey, we're having a fight about
5:15XYZ. Uh, what should we do?" And I asked
5:19it, I said, um, "Hey, if I was to do a
5:21bunch of psychological tests, which ones
5:24would you want to see?" And it gave me
5:25this list of like 15 different screens
5:28that it wanted me to do. And so, I went
5:30to Claude Code, and I said, "Hey, can
5:31you just put all these into a
5:32multiple-choice test, set up scoring,
5:35and then put them in a JSON file?" And
5:37so, both of us did that. It took about
5:3940 minutes, and we put the JSON files
5:42into ChatGPT, and we said, "Without
5:44knowing anything about this this couple,
5:47um, tell us about their relationship.
5:49And we read it line by line and like our
5:52jaws just like dropped. Like it nailed
5:55every single fight that we have, every
5:57single issue at home.
5:59And so I saw the power of these multiple
6:02choice personality tests. And so I you
6:05know saw the product opportunity and I
6:07was like I'm going to build a business
6:08out of this. And so I started this thing
6:10called Deep Personality. Let me just
6:13share my screen here. Quick break in the
6:16pod to let you know about a free
6:17workshop I'm doing that answers some of
6:20your biggest questions. The first is in
6:23the AI age, what are some categories
6:25that are ripe for startup ideas? I'm
6:28going to outline a bunch of those
6:30different categories, the ones that I
6:31think the most opportunity is. Second is
6:34there's a million AI tools out there,
6:36which are the ones that matter when I'm
6:38trying to build a business? I'll walk
6:40you through that. And lastly, how do you
6:43actually build a business with some of
6:45these AI tools? So I'm actually going to
6:47show you live how I do this with my
6:49ideabrowser.com team, and I can't wait
6:52to see you there. It's this Thursday at
6:5412:00 p.m. Eastern. And if you go to the
6:56show notes, if you go to the
6:57description, you can click a link, RSVP,
7:01and learn a few things. I hope it'll get
7:03your creative juices flowing and I'll
7:04see you back at the pod. And you know
7:07you could have you know you you own an
7:10agency called MetaLab. You could have
7:11called out MetaLab to go and build this
7:13for you. But you decided I'm going to do
7:16this all myself. I'm going to actually
7:18productize this myself. I'm going to
7:19design it myself. I'm going to build it
7:20myself. Why did you make that decision?
7:23Well, I think um
7:26I love people. Like I'm very
7:27extroverted, but the worst part about
7:30business is people.
7:32Um you know, if you think about using a
7:34let's use a screenwriting analogy. So in
7:36Hollywood there's all these brilliant
7:38screenwriters and they write these
7:39incredible scripts. And so they have
7:42this idea and they know where they want
7:44it to go, right? So let's say they start
7:46here and they want it to go here. They
7:47want to make an incredible feature film.
7:49In between
7:51all in between the script and the
7:54feature film are 100 people that they
7:56have to convince, $50 they have to
7:59raise,
8:00a million middle men and all that. And I
8:02think that creativity is just
8:05compromised based on how the number of
8:08people between your vision and
8:10execution. And so
8:13I think like writing for me has been a
8:14very like pure pursuit where I can think
8:18about it and I can write it and I can
8:19work with AI to make it better if I need
8:21to.
8:22But it feels it feels like so
8:25such a clear execution of your vision.
8:28And I feel like
8:30Vibe coding is
8:32that. Like finally I can do absolutely
8:37every part of it and I can do it at my
8:39own pace. You know, if I was using
8:41employees,
8:42they're unreliable, they don't see my
8:44vision, they don't understand every
8:46aspect of design. So you can have a
8:49great designer who sucks at front end or
8:52you can have or it doesn't know how to
8:53design something that will work in a
8:54front end. You can have a great designer
8:57who doesn't know how to write copy. So
8:58there's just all these pieces that come
9:00together with Vibe coding and basically
9:02like in a few manic days, I was able to
9:05build this pretty crazy app. So let me
9:08let me show you um how it works. So
9:10basically
9:11you do a personality test and it takes
9:14about 40 minutes. You're just clicking
9:15multiple choice answers and stuff and
9:17then you get this report. And so here's
9:20my archetypes. It'll be like you're the
9:22blazing architect. Here's your
9:24superpowers. Here's your kryptonite.
9:27This is just like the basic. And then
9:30as you scroll, you can see the AI has
9:33taken all of your answers and it's
9:35written like a 100-page report. So, this
9:39report literally goes on and on and on
9:41and on and it's written like Robert
9:43Greene. So, it's like super impactful
9:45and deep and hits you
9:48um and it basically walks you through
9:50what you're great at and what you're
9:52terrible at, what job you should have,
9:55um is your relationship working for you?
9:57Do you have ADHD, OCD, are you on the
10:00spectrum? Like all these different
10:02things and so you can see
10:04Let me get all my god, it's so big. All
10:06the way to the bottom. Um
10:08so, you can see like these are my kind
10:09of personality traits, my attachment
10:12style, um my internal family systems,
10:17like how much I people please, all this
10:19stuff.
10:20So, I built this thing and usually what
10:23happens for me is I start something and
10:26then I don't want to finish it because I
10:28don't want to hire the people and deal
10:29with it. And so, what I ended up doing
10:32is in this case, um just building agents
10:36to do everything. And so, I can show you
10:39where we were using open cloud. We're
10:40currently using this thing called Harbor
10:43that a friend of mine built um and it's
10:45basically a harness for agents. So, you
10:49can see all the different agents that
10:51are running here. So, we have a dev, we
10:54have a marketing agents and then we have
10:56a support agent. And basically the way
10:58it works is um somebody emails support
11:02and uh either support just tackles the
11:05ticket uh or it will actually fix the
11:08ticket. So, it'll sell send it to the
11:09dev agent. If it's a P0, so like a super
11:12scary security breach, it'll just
11:14immediately fix it and merge the PR.
11:17Otherwise, it'll uh we just wake up and
11:19have a bunch PRs.
11:21Um and then it'll email the person back
11:23and be like, "Hey, I fixed your issue."
11:24or whatever it is. That works basically
11:27perfectly. Like it's shocking. Like
11:29support to me is not a job very, very
11:31soon.
11:32Um the one that's really exciting is the
11:34marketing agent. So, we basically have
11:38it uh it's hooked into PostHog, so it
11:41has all the data for the app, and it
11:43actually manages a meta and Reddit ads
11:47account, and so it'll
11:49um do multivariate testing, it'll create
11:51ad creative, it will set budgets, and we
11:54basically just message it and we'll say
11:56um you know, "Hey, can you increase the
11:58budget by $1,000?" Or we approve like a
12:01big SEO project or something like that.
12:04So, where we're at with it is like we've
12:07done
12:08maybe like $20,000
12:10of revenue, like it's still small, and
12:13where I'm really excited um to go next
12:16is like what happens when we give it
12:18like a $100,000 a month ad budget and we
12:21find ads and creative that actually
12:23work.
12:24Um so, that's been that's been insane.
12:27Um and I think that we're probably
12:303 to 6 months out from being able to
12:32just
12:33hand businesses off to AI to run at
12:36least basic businesses like this.
12:39Harbor, I've never heard of.
12:42And I see I see a lot of stuff.
12:44Let me um let me just ask. So, it's my
12:47friend Gavin made it. Yeah. Yeah, link
12:50me to the Harbor website. Um yeah, it's
12:54Harbor. It reminds me a little bit of
12:57Paperclip. Have you seen Paperclip?
13:00Yeah, I have.
13:02So, this is at its
13:03github.com/geekforbrains/harbor.
13:06Yeah.
13:07And uh he's he's a friend of mine that
13:10I've been working on this with, and uh
13:12it's it's working really well. We just
13:14found that Open Claw wasn't quite
13:16deterministic enough, and the other
13:18issue with Open Claw is because it's a
13:19text-based interface, it was a little
13:22bit hard to keep track of all the agents
13:24and what's actually happening.
13:26Um so, it's worked really well for us.
13:28Yeah, I mean, I think what's cool about
13:30it is like if you're starting a company
13:33and you're thinking about org charts,
13:35roles, responsibilities,
13:38you know, you don't think about it
13:40within within text, you know, back and
13:42forth, right? You think about it as
13:44like, "Okay, I'm going to spin up this
13:45agent or employee to do XYZ, and then
13:49how can I monitor what that XYZ employee
13:52is doing?" Yeah, like so basically like
13:55it's just a it's just a GUI basically
13:57for something like Open Claw, and it's
13:58still running on Claude.
14:00Um, but you know, you have all these
14:01documents with uh kind of like a
14:03knowledge base, you've got databases,
14:06you have all your environmental
14:07variables, etc. It's like it's very
14:09similar to Paperclip. Um, I think the
14:12problem it reminds me a little bit of
14:14crypto where
14:16crypto is not crypto is built by nerds,
14:18so they make it nerdy. And when I saw
14:20Paperclip, I got so excited because I
14:22was like, "Oh my god, I can this makes
14:24sense to me visually." You like have
14:26this org chart, but it's clearly built
14:28by nerds because it's very confusing.
14:30And so
14:32my I mean, my kind of belief is like
14:34pretty soon Anthropic and uh OpenAI are
14:37going to launch like basically like
14:40CEOs. So, you just be like, "I'm just
14:42going to hire a CEO for my company, give
14:45me your data, I'll figure it out, I'll
14:47run it, I'll run the whole business."
14:48And it right now, the problem I feel I
14:50feel like um you know, 2 years ago it
14:53was like uh
14:55we're you get you get 5 minutes with a
14:58super genius who has Alzheimer's. Like
15:00the context window is so small that you
15:02just have these brief moments.
15:05Now I feel like with the the 1 million
15:07context window, um they can like maybe
15:10remember a day, but it's like Memento,
15:12right? Um, and I feel like once the
15:14context window is like 5 million, 10
15:17million, then they'll be able to run
15:18entire companies with this. Yeah.
15:20There's some people now selling
15:22autonomous companies. I don't know if
15:24you've seen this. Pulsia is one of those
15:26companies.
15:27I think Pulsia, p o l
15:29>> work? Like I I saw that New York Times
15:32story about that guy who like started
15:34the autonomous
15:35uh you know, what is it? Like
15:36billion-dollar business or whatever. But
15:39and obviously that's real and New York
15:40Times is writing about it. But I I can't
15:43parse what's real versus
15:47I think
15:49and uh you know, we're recording this
15:52April 29th, 2026.
15:54You cannot run a an fully autonomous
15:57company
15:58in right now
16:00that it yeah, that is going to get
16:03better that you can just like
16:06you know, go hang out on the beach 24/7.
16:09That being said, I think there's parts
16:11of the business that can be autonomous
16:12like support is like a good part you
16:16know, area. But like you're not going to
16:18outsource like product at this point in
16:22my opinion autonomously to to something
16:25like Pulsia today. You know, maybe that
16:27changes in 6 months, 12 months,
16:29whatever. But like my take is the people
16:32that are selling autonomous companies
16:34today are just selling a dream. Yeah, I
16:36kind of agree. Like I think um
16:39open claw agents are still like Zap.
16:42They're basically like Zapier zaps but
16:45that can make basic decisions and have
16:47intelligence. But you still have to tell
16:49them step by step, this is how I want
16:52you to think. This is how I want you to
16:54operate. Like let's say you hired an
16:55intern and you gave them an email
16:57address. You would never have to say,
16:59you should check your email every 15
17:01minutes and you should reply. And when
17:03you do reply, you should you know, think
17:05about the following things. Like it
17:07feels very It's like you're dealing with
17:10a baby. A a genius baby but you have to
17:12teach them how to do every single thing.
17:15Where I'm excited is when I can just
17:18have a model that's just like a CEO
17:20model or or and just like delegate to
17:23them and they'll just figure it out,
17:25which I think is kind of what everyone
17:27acted like Open Claw was, but it's
17:29totally not. Yeah.
17:31I want to go back to go to the deep
17:32personality test. Can I give you
17:34unprompted feedback? For sure.
17:37So, when you were going through this, I
17:38was like like the content looked
17:41exceptional.
17:43My my you know, my issue in general with
17:47like vibe-coded projects is that
17:50because anyone could, you know, create
17:53apps and products like this,
17:55it's almost like you need to have some
17:58credibility layer in order to listen to
18:01the content, you know what I mean? So,
18:03like
18:04um
18:06my you know, if I was like the you know,
18:09PM or CEO on this, I would be like, how
18:11could I partner with
18:13you know, someone who's credible in the
18:15space
18:16such that when I get this you know,
18:18output, this report, I'm going to take
18:20this seriously.
18:23I think part of the problem is it's
18:25anyone who is
18:27a psychologist or um someone who would
18:30be like legit on this would never do it
18:34because it's a bit cheeky. Like I'm
18:36using um psychological screens that are
18:39supposed to be interpreted by a
18:40psychologist,
18:42but I'm not a psychologist. I'm using AI
18:43to do it. And I think it's incredibly
18:46accurate. Like I've I've hammered it on
18:48a million different models and I've
18:49looked at um example tests and stuff,
18:52but that is definitely an issue. If
18:54there's someone listening who's
18:55connected to like a Jay Shetty type
18:58person, definitely let me know because I
19:00think it's perfect for that. Um it
19:02really is one of those like 1 + 1 = 5
19:05kind of things. But yeah, right now it's
19:08what I realized is like I can when I
19:10post this in my newsletter, it goes
19:12crazy viral
19:14um for a bit.
19:15But I don't have enough pull in that
19:17space where my audience is a bunch of
19:19like tech bros. They're not going to,
19:21you know, they're not going to share
19:22this with all their friends and stuff.
19:24It does definitely need something like
19:25that. Yeah, and I I I also wonder like
19:29you know, I think about like who
19:31like what are influencers, celebrities,
19:33micro-celebrities that people look up
19:35to, like a couple, like even just a
19:37couple that like has gone through this
19:40publicly maybe and have come out of the
19:42other side and they're just happier and
19:44when people think of like
19:46you know, a couple that's, you know, a
19:48dream couple, like this is their, you
19:50know, they're involved in this in some
19:52capacity. That's smart. So, you and your
19:54wife, you want to do Should we do it?
19:56>> Yeah, yeah. I mean, my my wife is like
19:58very like
19:59uh
20:02private, you know? So, I'm more of the
20:05public one. Um but I'll see what I can
20:07do. The same as my girlfriend, she
20:09hates publicity and stuff.
20:11>> Yeah. Um let's see what else can I show
20:13you. Um I mean, another big piece is
20:17just data. So, I, you know, I feel
20:21really frustrated because at different
20:24times I've kind of standardized on
20:26different LLMs, like I'll get into the
20:27Claude ecosystem and I'll use the
20:29desktop app, I'll use ChatGPT and I'll
20:32build out a bunch of projects. I've
20:34basically just accepted that I'm going
20:36to live in Claude code until somebody
20:38builds her um which I'm sure is coming
20:41in the next couple of years. Um and so,
20:44my strategy is basically been
20:46um to take all the different data
20:48pipelines that I have and put them into
20:51a centralized place. And so, I have um
20:54Fireflies record all my meetings. Um
20:57every night a cron goes in to the API,
21:00builds an
21:01a markdown file, puts it into G brain.
21:04Um I've built out vector databases um
21:07for my businesses. This is a really
21:10cool,
21:11um, thing. So, you know, a vector
21:13database can store, um, you know,
21:15millions and millions and millions of
21:17tokens, and you can basically, um,
21:20you can basically have like a service
21:22like Pinecone go and crunch through a
21:25large amount of data, and it basically
21:27makes it highly searchable for an LLM.
21:29And so, I can show you an example of
21:32that. Let me just switch the screen
21:34sharing. So, okay, so here's an example.
21:37So, like, I did, um, I trained a vector
21:40database on my family office, Foley
21:43Partners, which is basically just my
21:45like personal holding company. And here
21:48I just asked it, uh, what did I say?
21:51Um,
21:52break down how many minority venture
21:54investments I've made and how many are
21:56in the money, bankrupt, or declined. And
21:59basically it went off and did this
22:00query, and then boom, I can see that
22:03I've made 132
22:05direct investments. I invested 16
22:07million, it's now worth 36 million, and
22:10then I can break down all the different
22:11write-offs, everything. And so, this has
22:14been, um,
22:15really, really powerful. So, like, if
22:18you think about, I did it for Tiny as
22:20well, and in Tiny I have, uh, man, 24
22:26businesses or something like that, tons
22:28of historical data. And the challenge
22:31when you run a conglomerate is like, I
22:33don't know it, like, are any of the
22:35companies spending too much money? Are
22:37any of the CEOs full of Um,
22:40are any of the businesses trending up,
22:42down, expenses are going up? It's just
22:44too much data for one person to
22:46understand. And so, I can't show it cuz
22:49it's all public company data, but
22:50basically what I can do in Tiny now is I
22:52can say, "Hey, review the last quarter,
22:54and I want you to give me any icebergs
22:56or issues." Or I can ask a simple
22:58question like, "How how accounting staff
23:01do we have in head office?
23:03Um or whatever. And I've found this so
23:06useful. It's not perfect. Like it'll
23:08it'll often like say the wrong person is
23:10CEO of a business or the numbers are a
23:12bit off. But in terms of being able to
23:15like be the eye of Sauron inside your
23:17company, I think it's like incredibly
23:20powerful. Mhm.
23:22And and curious like within Tiny, like I
23:26see how you're using it to interface
23:27with Tiny and and to you know, be a
23:29better chairperson and give advice. But
23:34within Tiny, I imagine that there's like
23:36some efforts to make the company's AI
23:38first and stuff like that, right? Yeah,
23:41and that's one of those challenges that
23:42I think everyone can relate to where
23:45what is the saying? It's um never expa-
23:48expect someone to understand something
23:50that their paycheck depends on them not
23:51understanding. So, you've a lot of
23:54people that are living in the old world
23:57and need to be upgraded um in their
24:00understanding. And often they're very
24:03attached to the old way of doing things.
24:05And so, to be honest, like the challenge
24:07in Tiny is that we are generally pretty
24:10hands-off. And where we're really
24:12leaning in is in a lot of the software
24:15businesses because those are the ones
24:17that are at the most risk. I think we've
24:20talked about this before, but
24:22it's not like software is dead. If
24:23anything, software is going to thrive
24:25way more. But it used to be like the
24:27newspaper industry where it's like in
24:30order to compete, you have to assemble
24:33this incredibly expensive group of
24:35people that's very rare to find, right?
24:38You need like great programmers, great
24:41designers, etc. And that that was like a
24:43very finite resource. So, it's kind of
24:44like the newspaper industry where you
24:46had to be rich, you had to buy like $2
24:48million printing press. And then once
24:50you have it, you actually have a
24:51competitive advantage. Now, I just think
24:54software is free. Anyone can make it.
24:56Anyone can rip it off. Like my
24:58personality app, it's cool.
25:00Does it have a moat? Absolutely not.
25:03Like maybe if to your point I had John
25:05Legend and Chrissy Teigen backing it or
25:08something, maybe there'd be something
25:10there. But I think pricing pressure is
25:13just going to go crazy because it's like
25:16the restaurant industry. Like, you know,
25:18restaurants are not a bad business,
25:20they're bad because there's too much
25:21competition.
25:22So, if you're listening to this and and
25:25you know, someone's listening to this,
25:26they're watching this actually and
25:27they're saying like, "Okay, I like what
25:29Andrew's doing here with like
25:31you know, this Foley advisory thing. I
25:34like what he's doing with personality
25:35test. He I feel like he has a good sense
25:38of where the world's going.
25:40You know, where where what you know,
25:42what would you be building now if you're
25:44you know, if you're
25:45you want to be building something?
25:47It right?
25:49>> [sighs]
25:49>> Where's your opportunities?
25:50>> I think we talked about this a year ago
25:53and if anything, I've become more
25:55concerned.
25:57Um
25:58when I look around, I'm
26:00don't get me wrong, I'm not not
26:01concerned. I'd just say it's getting
26:03harder and harder and harder. Like
26:07I think we've all been astounded seeing
26:10um how good Claude has gotten and how
26:12much it can do
26:13um and how quickly it can do it. And
26:16from my perspective, I think that if I
26:19was somebody who's obsessed with vibe
26:21coding and is 20 and let
26:24I would say the goal should be like make
26:26one to two million dollars by building a
26:28product like deep personality or
26:29something like that, which is pretty
26:31straightforward. Try and make a million
26:33or two million bucks and this is such a
26:35sad answer, but I'd be like invest in
26:37data center stocks or invest in TSMC
26:40because to be honest, that's what I'm
26:42doing. Like I am I'm either buying like
26:46something with a moat, which is very
26:47very hard to find certainly in
26:49technology right now.
26:51Uh home services businesses like very
26:53very brick and mortar kind of stuff.
26:56But even that I think is at risk with
26:57robotics over the next 10 20 years.
27:01But I honestly I all I can really
27:03recommend someone does is like buy Iron
27:06and TSMC. Mhm.
27:08Yeah, I think uh
27:10with you know, with respect to like the
27:12GUI, the graphical user interface, like
27:16it just feels like where the you know, I
27:19always I think about it as like the
27:21hour of progress, like the hour of
27:23progress or deep you know, deep progress
27:25is that
27:26the UI is going away. And
27:30like I don't know about you, but I'm
27:32spending more and more of my time in
27:33Codex. I'm spending more and more of my
27:34time in Cloud Code. Like those have
27:36become my operating systems. So I'm
27:38spending less and less time in apps.
27:40And it's easier than ever to create
27:42apps. Therefore, like yeah, it's pricing
27:44pressure, competition, harder to get
27:47mind share.
27:48Well, here I mean it's an interesting
27:49like someone asked me like is Adobe
27:52Yeah.
27:53>> And that's an interesting one because
27:54like I'm a pretty avid user of
27:56Lightroom. And I love I've got a Leica.
27:59I love editing photos and stuff. And
28:02while I do put my photos into Gemini
28:05sometimes for like recoloring or
28:07retouching or something, I still really
28:10enjoy the process of like moving the
28:12sliders and making it exactly the way I
28:14want it. And so I think you know, it's a
28:17question of like does this being the
28:19standard matter or will someone just
28:22vibe code an even better Lightroom and
28:23everyone will switch to that because
28:25they're you know, a big slow incumbent
28:27and someone else will do it faster. I
28:29can't answer that question very well.
28:32All I can say is software is a worse
28:34business now than it was 5 years ago. I
28:36mean, look at Constellation Software.
28:39Their stock has been cut in half. Like
28:40that that's crazy. Like I and I don't
28:43know if that's correct because they own
28:44like the most boring software businesses
28:46ever. But um yeah, dude, it's like a
28:48brave new world. And it's so depress I I
28:50hate to give such a depressing answer
28:51cuz your show's all about
28:53entrepreneurship. And like I just want
28:54to say like I'm more excited than ever.
28:56There's never been a better time to be
28:58an entrepreneur and build stuff. It's
29:00just you are picking up pennies in front
29:02of the steamroller in most industries.
29:06It's also hard to predict what's going
29:08to happen in 12 months from now cuz
29:10you know, uh or 18 months from now or 24
29:1424 months from now. So like
29:16my advice to people is just build now
29:19knowing that um you know, you're a big
29:22Buffett guy, right? So like, you know,
29:24what does he say? Uh
29:26you know, there's only so many puffs you
29:27get on the cigarette or whatever. What
29:29is it? Yeah, you can um get Well, he's
29:32he's more talking about cigar butts,
29:33right? So you find a cigar butt on the
29:35street and you can get one puff out of
29:37it for free. And I think a lot of
29:40previously high-quality businesses that
29:41would be like a gourmet cigar from a
29:43cigar shop are now cigar butts. Right.
29:46Yeah.
29:46>> So my take my my you know,
29:48glass-half-full take is you
29:51you can get some puffs out of it and and
29:55it might turn in like your personality
29:56test might be like, oh, a $1 to $2
29:59million a year business for 3 years or
30:01it might turn into like a media
30:03conglomerate.
30:05So how
30:05>> built it. I built it for, you know, I
30:08was pretty irresponsible cuz I'm like
30:09using fast mode and going all out with
30:11all the cost, but I could probably build
30:14it for 10 grand, 20 grand of tokens or
30:17whatever to get to where we got to. I
30:19probably spent like 80 or 100 honestly
30:21cuz I've been so Dude, I'm I'm spending
30:23like our family office um
30:26So basically like I have Tiny and then I
30:29have my personal holding company um
30:31and you know, family office the just the
30:33rich person word for uh personal holding
30:36company. So I I'm sorry for being a
30:38douche. But um but basically like we
30:41have been instead of scaling employees,
30:44um, um we've literally just been scaling
30:46Claude. And so, I'm so lucky. Like, my
30:49CFO and president have both become
30:51obsessed with Claude code, but they're
30:53actually just, um, scaling the API cost.
30:56So, instead of a payroll, we just have a
30:57$40,000 a month Claude bill. And it's
31:00doing it's doing everything. Like, I
31:01love to show you some of the stuff that
31:03they've built. It's just crazy.
31:06Yeah, I would love to see it.
31:08Here's here's an example of like
31:10software being completely So,
31:13there's this product called Addepar.
31:15Have you ever heard of it? No. So,
31:17Addepar is like, um,
31:19basically like software for rich people
31:22to track their portfolios, and they
31:24charge 50 to $100,000 a year.
31:27Um, and basically what it is is like, it
31:29pipes into your bank, it pipes into your
31:31accounting, it pipes into the public
31:32market data, and it just tells you like,
31:35here's how much, um, here's how much
31:37money you have. Here's your personal net
31:39worth. Here's all your assets. Here's
31:41your balance sheet, etc.
31:43And so, what we did is my CFO just said,
31:46"Fuck that. I don't want to pay, um,
31:48those guys all that money, and we can
31:49actually build something that is way
31:51more customized to us, and, um,
31:53integrated into all of our data." And
31:55so, he said that, and like the, you
31:57know, the
32:00You remember like 5 years ago when one
32:01of your engineers would be like, "Why
32:02are we paying Slack all this money?
32:04Let's just build it ourselves." We all
32:06know that's like a fool's errand, right?
32:08Like, that's insane. And so, when he
32:09told me he was just going to build it, I
32:11was like, "First of all, you're a CFO.
32:13Like, I'm a little skeptical. Um, but
32:15two, I was like, "Look, do we really,
32:17you know, is that actually going to
32:18work?" In like, 2 weeks, he built this.
32:21Has this AI assistant. So, this I can
32:24query,
32:25um, let's see like what's, um, stress
32:28test. What is this?
32:30So, basically it's like the basically
32:32like a, um,
32:33Anthropic API, but it has access to all
32:36of our data.
32:38So like okay, here's one.
32:41I screwed it up. Bad demo. Steve Jobs
32:43would be yelling at me right now. Yeah.
32:45So this is going to go off and query
32:47this. Let's see what it does.
32:49But anyway, this was built by my CFO.
32:52He's never coded in his life. Like when
32:54I told him about Cloud Code and that he
32:56should check it out, he's like, "I don't
32:57know how to code. Like I'm not
32:59technical." And literally within like a
33:01week, I'm so lucky that he's the kind of
33:03person that would do this. But he built
33:05this and it's just it's insane.
33:08Going back to the glass of half full of
33:10entrepreneurship, you know, I've been
33:12seeing a lot on Twitter lately of like
33:14services is the new software. So I could
33:16imagine, I know this is you built this
33:18for yourself, but I can imagine a
33:20product like this
33:23and then that you sell services on top
33:25of this, right? Like if you if you
33:27actually wanted to sell this to other
33:28family offices and stuff like that.
33:30Whereas like, right? Like a weekly call
33:32or maybe it's like all, you know,
33:35add skills every single week or help you
33:37do XYZ. Yeah, I mean I think it would
33:40the way I look at it is
33:42I always give the example of funeral
33:44homes. So funeral homes are very unsexy
33:47industry that nobody wants to work in.
33:49And so you don't have that much
33:51competition. And so if you build
33:52software for let's say like some guy who
33:55owns a funeral home has like a web wiz
33:57nephew like you or me in like 2010
34:01and he's like, "Hey, can you build
34:02software that's custom for funeral homes
34:04for me to manage mine?" And then you
34:06start the software business and you're
34:08the only game in town. Like no other no
34:11other person owns this funeral home and
34:14has a web wiz nephew who's going to do
34:15that. And so you have a monopoly and you
34:18become the default. And I think the new
34:20world is, you know, if 400 people in the
34:24United States own funeral homes, let's
34:26say like 20 of them are smart and know
34:29AI. They're all going to build their own
34:31software. And so what does that do? That
34:33means it's great for the consumer.
34:35There's now 20 different software
34:37choices that they can use. Someone can
34:39build their own custom thing if they
34:40want to, but pricing comes down big
34:44time. So like, you know, Addepar, the
34:46company I mentioned before, they're
34:48charging 50 to 100,000 dollars a month.
34:50I don't think that they're going to be
34:51able to do that for too much longer
34:53because there's going to be a lot of
34:54competition that comes out. Hopefully,
34:56somebody watches my video and says, "I'm
34:58going to build this, too." and then go
34:59crush Addepar cuz like it's not
35:02particularly good software in my
35:04opinion.
35:05Mhm.
35:06So you can see, okay, so for some reason
35:07I can't scroll still Vibe Coded, but you
35:10can see it's like giving me an analysis
35:12of like how I could rebalance my
35:13portfolio. Um it can do all sorts of
35:16stuff like it can track risks and be
35:18like, "Oh hey, you have XYZ risk in your
35:21portfolio." Um it's very cool. Awesome.
35:25Um we only have a few minutes left. What
35:27uh anything Well, I actually have a
35:29question. You mentioned uh
35:31G Brain before, and I haven't gotten
35:34around to using it.
35:36Um
35:38how have you found it? And and for, you
35:39know, can you explain what G Brain is
35:41for for people who don't know? Yeah, I
35:43will. So Gary Tan is a friend of mine,
35:45and he basically built this um Vibe
35:48Coded project where he wanted a vector
35:50database um
35:52for personal kind of knowledge base. So
35:55really quick um ability to draw from a
35:58large base of markdown files on your
36:01computer. So basically trying to solve
36:02the problem of memory. And because he's
36:04a friend, I was like, "Oh, I'll install
36:05this and try it out." And actually, it's
36:07really, really awesome. Um there's a few
36:10different people kind of competing
36:12in [clears throat] the open-source world
36:13for this.
36:14Um but I've been really impressed with
36:15it.
36:17I don't I don't want to do a demo cuz
36:19it's like actually a really boring demo.
36:21But basically, what I did is I ingested
36:23all of my email, and I said, "Hey, I
36:25want you to um go through it my email
36:28thousand at a time and I want you to
36:29draw connections and I want you to build
36:31pages for everybody that I know." So,
36:33basically I have a page on every person
36:36that I know and then I can say something
36:38like um
36:39"Hey, I'm raising money for our yerba
36:41mate business. Who do I know that would
36:43be good for that round?" And then I can
36:45say, "Okay, now go and draft a powerful
36:48email to them and make a deck and send
36:50it to them."
36:51Um and so I just find that
36:54Oh, what what is what this has enabled
36:56me to do is like
36:57have a task that would otherwise take
36:59like eight hours. Like I I do these like
37:02men's groups and we rent board rooms and
37:05I need to charge everyone because I pay
37:07I end up paying for dinner in the board
37:08room. And so I was like, "Okay, I got to
37:10I got to send everyone Stripe links. I
37:12have to get everyone on a subscription.
37:14I have to send all the emails. I have to
37:15hector them." And I just had that as a
37:17single task to my open claw agent and it
37:20went out went in the Stripe API, sent
37:22the emails, did everything. So, like
37:24there are these magic moments using
37:26stuff like G brain and and open claw.
37:30I want to I want to show you two really
37:32cool actually three really cool things
37:33before I go. So, one So, I mentioned
37:35this guy Gavin Vickery who built Harbor.
37:38So, we were doing a vibe coding retreat
37:40at my my cabin and I was talking about
37:43how sad I was that
37:46the Limitless pendant sold to Meta. So,
37:48it's like this little pendant. I think I
37:50showed it last time I was on. You just
37:51wear it and it would record your entire
37:53day and build transcripts of it.
37:56And I was saying,
37:58"I'm so annoyed because the iPhone has a
38:00microphone on it. Like why can't the
38:02iPhone just record all the time?" And he
38:04just went, "Well, I can build that." And
38:0724 hours later he had this app called
38:10Hearsay and basically
38:13can't really see it, but what you do is
38:15you say what times of day you want to
38:17trigger it. And so you can be like, "Oh,
38:18I'm I'm in meetings in the afternoon or
38:21or you know, I want it to record all
38:23day. And then it literally just records
38:25your entire day and builds transcripts
38:28and then sends them to your iCloud. And
38:30so you just have context on your life.
38:32And so I record my entire day and then G
38:35brain ingest it and it just knows
38:37context and then my open claw agents
38:39have full context. Um, let me show you
38:43also a couple cool things I've been
38:46doing with
38:48open claw.
38:50You've been busy, Andrew. I have been
38:53very busy. I mean, this is all I do.
38:55Like this is just like a complete
38:57obsession
38:58and it is um, kind of sad because I'm
39:03like
39:03let's see, cuz I'm always spending my
39:06time um,
39:08I'm always spending my time
39:10uh, tweaking my stuff versus doing
39:12actual work. So, um,
39:14basically like it does it reads all my
39:17email, triages all my email and then it
39:20figures out automatically what projects
39:22I have going on. We might have to blur
39:25some of this. Um, but here's so it's,
39:27you know, I'm on your podcast and it's
39:30telling me, oh, you need to um, prepare
39:32for it.
39:33Uh, you know, it randomly was like, oh,
39:35your airtable account's going to go out.
39:37It's my brother's birthday. I'm bidding
39:39on a building. Like it just figures out
39:41all those things. And so every day I get
39:44this report and then you can see that it
39:46um, gives me next steps. So, I can say
39:49like, don't worry about it or I can say,
39:52hey, send an email to my um, my admin
39:55asking her to um, remind this person
39:58that. So, that's pretty cool. I also get
40:01um, emails. So, this is like it's
40:03identified a high priority email. Again,
40:05we're going to have to blur this because
40:07this is someone's comp, but this is
40:08somebody saying, oh, hey, I'm owed a
40:10bonus, right? So, and then it drafts in
40:13my voice different response options. So,
40:15I can just respond inline in Telegram
40:18and say 1B. Uh, Um and so, it turns your
40:20entire business into a multiple choice
40:24um quiz, basically. So, every day I just
40:26get like 20 different messages. They're
40:29emails or it goes in my iMessage and
40:31identifies things I need to respond to
40:33and I can just say like 1A, 2B, 3C.
40:38And um and that's saved me a shitload of
40:40time. Another cool thing I've doing I
40:43built a brief. So, basically this goes
40:45into my Readwise reader and my email and
40:48it looks at which newsletters came in
40:51and then it based on what it knows about
40:53me. So, I prompted it. I kind of said
40:55like I'm interested in AI and health and
40:58certain things. Um and then it chooses a
41:00few stories and then check this out. Did
41:03you ever listen to The Daily on New York
41:05Times?
41:05>> Yeah. So, listen to this.
41:07Here's what landed in the inbox and
41:09reader today, Tuesday, April 28th. We
41:12have a special forces soldier who bet on
41:14an operation. So, basically makes a
41:16custom podcast using Gemini voice and I
41:20just listen to that every day in the
41:21shower and it's so cool because one of
41:23the reasons I stopped listening to
41:25podcast like that was because they would
41:27be depressing or I you know, just stuff
41:30I don't care about. Um so, I have it
41:32prompted to like only tell me things
41:35that are going to make my day better or
41:37are relevant to to my own life or my
41:39city or my businesses. And then I also
41:42have it do a countdown. So, it says you
41:45have um
41:46this many more summers with your kids.
41:48You have this many more days to live.
41:50Here's a quote from Seneca, you know, to
41:53give you like a sense of purpose today.
41:55So, it's just incredibly cool how you
41:58can basically make your media diet
42:00whatever you want in a custom way with
42:02this. I mean, that's a product in
42:04itself, right? Like honestly. For a bit
42:07until Open AI does it.
42:09You think Well, you think they would do
42:11it? like the 7-minute podcast? Like I
42:14would subscribe. Like if you if it was
42:16$10 a month
42:18to
42:20the 7-minute podcast that's going to
42:23make you more productive, happier, and
42:24healthier.
42:26Uh and that is personalized to me. Like
42:30I would pay for that.
42:32Yeah. I think I think it's kind of like
42:35um
42:36you the other day I taught my open claw
42:39agent. I said, "Whenever I tweet, I want
42:41you to create an Instagram story." So it
42:44creates an image that's perfectly scaled
42:46for Instagram, and I want you to um put
42:48it into the um Instagram uh creator
42:51thing, right? So basically schedule it.
42:53And that was a product before. So
42:55there's this company called Tweet Shot,
42:57and they basically automate that. It's
42:59just dead, you know? It's like anything
43:01that could be an API call is just
43:03cooked. So to me, that's like an API
43:06call business. That's like the first
43:07thing to get run over. But for like
43:09normies, I agree. Like if I told my dad
43:12about it, and I could be like, "Hey,
43:13what topics are you interested in?
43:15Create me a daily podcast and text it to
43:17me." Like I I do agree that is a cool
43:19idea.
43:20Someone should build that, by the way.
43:22And me and Greg will back you.
43:24Um so here's another one. So so I have
43:27two open claw agents running on a VPS,
43:30and
43:31um Ava is kind of my personal assistant
43:33that I just showed you. This is Mara.
43:35This is my um my doctor um doctor agent.
43:39It also helps I've made them very
43:40attractive women, so I always want to
43:42talk to them. Um and I'm always
43:44flattered when they message me. Um but
43:46this this is really cool. So I have it
43:48um I have all my Apple Health data from
43:51my Apple Watch, my eight sleep, etc. And
43:54it uploads into a Google Drive. So I
43:56have a JSON file of all my um health
44:00metrics. And then every morning, it
44:02sends me a health summary. So it looks
44:04at my HRV, resting heart rate,
44:07respiratory rate, tells me about my
44:09sleep, and then I have this weird
44:12viral nerve pain that I get, and it
44:16looked through my data over the last 5
44:18years and said, "Oh, I've correlated it,
44:21and I see that every time you get that
44:22flare, your wrist temperature changes
44:25for 3 days before." And so, it tells me
44:28in advance that
44:31and then it also tracks like my
44:32medication adherence and all that kind
44:34of stuff. So, it's just stuff like
44:37I And then the other thing is like if I
44:38ask it a question, it's prompted to use
44:41Max Intelligence. So, let's say I said
44:43um
44:44"Hey, should I take this medication?" It
44:46will spin up a uh rheumatologist, a um
44:51you know,
44:53uh internist, like every type of doctor
44:56as a Claude expert with extra high
44:57thinking, and it will do a team of
44:59experts for like 10 minutes across my G
45:02brain and all my medical data, and it'll
45:04give me a really informed answer. So,
45:06I'm like just blown away by some of this
45:09stuff. Andrew, we covered a lot of
45:11territory, like way more territory than
45:14I even thought was remotely possible,
45:16but
45:17uh that's why I love having you on. I
45:19appreciate it. We did We did pretty much
45:21everything. We did agents, we did
45:23running your business, we did family
45:25office, we did health, we did
45:27personalized health, we did living
45:29happier and healthier.
45:31So, I want to thank you for coming on.
45:32I'll include links on where to find and
45:34follow Andrew on the internet, where to
45:37take uh the deep personality test, where
45:40to find Tiny if you're ever interested
45:41in
45:42selling a company.
45:44And uh Andrew, is there anything you
45:46want to leave people with? Yeah, I have
45:48one prompting tip I want to give people
45:51um that's changed my use of Claude a
45:53lot. So, I will say um
45:56"This is my goal." So, let's say uh real
45:58estate example, "I'm thinking about
46:00buying this building.
46:02I want to interview me. Ask me I'll
46:04literally write ask me a shitload of
46:06questions to determine your prompt and
46:09use question tool. So, it'll pull up
46:11multiple choice questions. It'll
46:12interview me for sometimes five or 10
46:15minutes and what I found that it does is
46:17just give me like an incredible breadth
46:21of basically building perfect prompts
46:23and I think this is like the thing that
46:25everyone misses cuz people are always
46:26like, "Oh, like you know, there's always
46:29going to be jobs because someone needs
46:30to build the prompts." And it's like,
46:32"No, nobody needs to build the prompts.
46:34The AI should just interview you." It
46:36OpenAI just doesn't want to spend all
46:38the computer at this point. So, it is
46:40fully capable of just interviewing you
46:41to figure out what you need.
46:43So, I love that and I also highly
46:45recommend using agent teams. So, I'll
46:48say always use a team of agents of eight
46:51sub agents and I find the answers are
46:53just incredible.
46:54I like that. Good tips.
46:56Andrew Wilkinson, thanks for coming back
46:58on the pod. Thanks, dude. Always fun.

About this transcript

This page contains the full transcript of AI Agents run my business and life (Andrew Wilkinson) by Greg Isenberg, generated from the public captions YouTube serves with the video. The transcript has 8,588 words across 1,253 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.