May 4, 2026 · 47m · startup-ideas
AI Agents run my business and life (Andrew Wilkinson)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Investor and entrepreneur Andrew Wilkinson joins Greg Isenberg to reveal his personal workflows for managing business empires, family office operations, and health using autonomous AI agents. He shares live demonstrations of vibe-coded applications, bespoke executive assistants, and explores how generative AI is fundamentally commoditizing traditional software moats.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 20.4% of the talking time here. How this is scored →
Greg as informed peer 3.9 Guest teaching 5.2 Guest disagreement 1.5 Greg pushing back 1.7
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 42:32 Andrew dismisses API wrapper startups
Andrew firmly rejects Greg's enthusiasm about commercializing custom daily audio briefs, stating that anything reducible to an API call has zero defensibility.
Hardest push from Greg ▶ 15:48 Greg calls fully autonomous companies a fantasyGreg directly challenges the hype around autonomous software startups like Pulsia, arguing that anyone selling fully autonomous businesses today is selling an illusion.
Biggest teaching moment ▶ 20:50 Andrew breaks down conglomerate vector databasesAndrew gives a masterclass on using Pinecone vector databases to query complex private equity and holding company financials across dozens of subsidiaries.
Greg holds their own ▶ 17:36 Greg identifies the core distribution flaw of vibe codingGreg demonstrates sharp product strategy expertise by pointing out that vibe-coded apps become worthless without a credible distribution partner to legitimize the output.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| Welcoming Andrew Wilkinson and Setting Expectations | 2 | 5 | 1 | 0 | Greg introduces Andrew and sets up the episode's focus on AI workflows. Andrew details his intense workflow using Claude Code and OpenClaw to autonomously run SaaS tasks. | |
| The Origin Story of Deep Personality | 4 | 4 | 1 | 2 | Andrew recounts the origin story of Deep Personality from a relationship test prompt. Greg interjects with an ad read and asks why Andrew built it himself instead of delegating to his agency MetaLab. | |
| The Philosophy of Vibe Coding and Deep Personality Demo | 2 | 6 | 2 | 0 | Andrew explains his philosophy on vibe coding, arguing that intermediary employees dilute creative vision, and presents a live walkthrough of Deep Personality's 100-page psychological report. | |
| Autonomous Business Operations with Harbor Agent Harness | 2 | 6 | 0 | 0 | Andrew demonstrates Harbor, an agent harness managing dev, marketing, and customer support sub-agents running ad campaigns and auto-merging PRs. | |
| Agent Interfaces and the Evolution of AI Context | 5 | 4 | 1 | 2 | Greg compares Harbor to Paperclip and discusses visual org charts. Andrew critiques overly complex interfaces and discusses the evolution of context windows from 'Memento' to full operational memory. | |
| Assessing the Reality of Fully Autonomous Companies | 7 | 2 | 2 | 7 | Greg pushes back strongly against claims of fully autonomous companies like Pulsia, declaring that founders promising hands-off businesses are selling a dream. Andrew concurs and grounds the current limitations. | |
| Distribution, Credibility, and Partnerships for AI Products | 7 | 3 | 2 | 5 | Greg provides critical feedback that vibe-coded apps lack credibility moats without domain authority partnerships. Andrew agrees and acknowledges his audience limitations. | |
| Centralized Data Pipelines and Vector Databases | 3 | 7 | 1 | 1 | Andrew shows how he connects Fireflies transcripts and company financials into Pinecone vector databases to oversee 24 portfolio businesses in Tiny. | |
| Overcoming AI Inertia and the Commoditization of Software | 3 | 6 | 2 | 1 | Andrew discusses organizational inertia against AI adoption and compares modern software commoditization to the printing press and competitive restaurant margins. | |
| Strategic Capital Allocation and Investment in the AI Age | 5 | 5 | 2 | 2 | Andrew gives a sober outlook on tech moats, suggesting capital is best deployed into hardware/TSMC. Greg notes that graphical UIs are rapidly disappearing in favor of terminal environments like Claude Code. | |
| Cigar Butt Businesses and Replacing Headcount with Compute | 5 | 6 | 1 | 2 | Andrew details replacing headcount with a $40k/month Claude compute bill and demonstrates how his non-technical CFO built an internal Adapar competitor in two weeks. | |
| Personal Memory and Autonomous Execution via GBrain | 3 | 6 | 1 | 0 | Andrew introduces Garry Tan's GBrain project and illustrates automated workflows that read email context to autonomously invoice and manage event attendees via Stripe. | |
| Ambient Life Transcription with Hearsay App | 2 | 6 | 1 | 0 | Andrew demos Hearsay for 24/7 background audio logging and explains his daily Telegram workflow where OpenClaw formats all communications into multiple-choice triage prompts. | |
| Curated Daily Audio Podcasts and Media Automation | 5 | 5 | 4 | 3 | Andrew plays a customized Gemini-voiced news briefing. Greg suggests commercializing it as a $10/month consumer subscription, but Andrew counters that thin API wrappers get crushed immediately. | |
| Mara: Personalized AI Health and Medical Diagnostician | 3 | 7 | 1 | 0 | Andrew shares Mara, his AI doctor agent analyzing longitudinal Apple Health biometric data and running multi-specialist diagnostic panels, before concluding with prompt engineering techniques. |
Statements from this episode (36)
Wilkinson: Built and Runs a SaaS Business Entirely Autonomously via OpenClaw
“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.”
Wilkinson: AI Agent Productivity Tooling Sinks 80% of Time into Maintenance
“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 and then 20% of my time being productive. So it's like the classic productivity treadmill.”
Wilkinson: Creativity is compromised by the number of intermediaries in execution
“And I think that creativity is just compromised based on how the number of people between your vision and execution.”
Wilkinson: Employees are unreliable and fail to understand founder vision
“You know, if I was using employees they're unreliable. They don't see my vision. They don't understand every aspect of design.”
Wilkinson: Customer support will not be a human job very soon
“Like support to me is not a job very, very soon.”
Wilkinson: Basic businesses will be fully run by AI in 3-6 months
“And I think that we're probably three to six months out from being able to just can businesses off to AI to run at least basic businesses like this.”
Wilkinson: Anthropic and OpenAI will soon launch AI CEO products
“My, I mean, my kind of belief is like pretty soon Anthropic and OpenAI are gonna launch like basically like CEOs. So you just be like, I'm just gonna 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.”
Wilkinson: 5M to 10M context windows will enable AI-run companies
“And I feel like once the context window is like 5,000,010 million, then we'll be able to run entire companies with this.”
Isenberg: Fully autonomous companies are still an unrealistic dream
“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. Like support is li…”
Wilkinson: OpenClaw agents still require micromanagement, not true delegation
“Like I 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. L…”
Isenberg: Vibe-coded AI apps require a credibility layer to be taken seriously
“My, you know, my issue in general with like vibe coded projects is that Because anyone could, you know, create apps and products like this, it's almost like you need to have some credibility layer in order to listen to the content.”
Wilkinson: AI is incredibly accurate at interpreting psychological screening tests
“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.”
Wilkinson: A 'Her'-style AI assistant will be built within years
“I've basically just accepted that I'm going to live in Claude code until somebody builds her which I'm sure is coming in the next couple of years.”
Wilkinson: Family office invested $16M across 132 venture bets, now worth $36M
“I can see that I've made 132 direct investments. I invested sixteen million. It's now worth thirty six million.”
Wilkinson: Tiny Intervenes in Software Portfolio Businesses Over AI Risk
“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.”
Wilkinson: Software Creation Is Effectively Free and Highly Reproducible
“Now I just think software is free. Anyone can make it. Anyone can rip it off.”
Wilkinson: Software Will Face Extreme Pricing Pressure Like Restaurants
“I think pricing pressure is just gonna go crazy because it's like the restaurant industry. Like, you know, restaurants are not a bad business. They're bad because there's too much competition.”
Wilkinson: Home services and brick-and-mortar face robotics risk in 10-20 years
“I'm either buying like something with a moat, which is very, very hard to find certainly in technology right now home services, businesses, like very, very brick and mortar kind of stuff. But even that I think is at risk with robotics over the next 1020 years.”
Isenberg: Traditional user interfaces are disappearing as AI environments replace apps
“The UI is going away. And, like, 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. Like, those have become my operating systems. So I'm spending less and less time in apps, and it's…”
Wilkinson: Software is a worse business now than five years ago
“All I can say is software is a worse business now than it was five years ago.”
Wilkinson: Constellation Software's stock has been cut in half
“I mean, look at Constellation Software. Their stock has been cut in half like that.”
Wilkinson: Many previously high-quality businesses are now 'cigar butts'
“And I think a lot of previously high quality businesses that would be like a gourmet cigar from a cigar shop are now cigar butts.”
Wilkinson: Spent $80k to $100k on AI tokens building personality app
“I probably spent like 80 or a hundred honestly.”
Wilkinson: Holding company spends $40k/month on Claude instead of hiring
“So instead of a payroll, we just have a 40,000 dollar a month Claude bill. And it's doing everything.”
Wilkinson: Non-technical CFO built portfolio tracking tool in one week via Claude
“But anyway, this was built by my CFO. He's never coded in his life. Like when I told him about cloud code and that he should check it out, he's like, I don't know how to code. Like I'm not technical. And literally within like a week, I'm so lucky that he's the…”
Wilkinson: AI-built custom software will collapse vertical SaaS pricing power
“And I think the new world is, you know, if. 400 people in the United States own funeral homes, let's say like 20 of them are smart and no AI, they're all going to build their own software. And so what does that do? That means It's great for the consumer. There…”
Wilkinson: Addepar will not be able to sustain premium pricing much longer
“I don't think that they're going to be able to do that for too much longer because there's going to be a lot of competition that comes out. Hopefully somebody watches my video and says, I'm going to build this too, and then go crush Adapar.”
Wilkinson: Garry Tan built GBrain as a personal vector database
“Gary Tan is a friend of mine and he basically built this vibe coded project where he wanted a vector database for personal kind of knowledge base. So really quick ability to draw from a large base of Markdown files on your computer. So basically trying to solv…”
Wilkinson indexed his email archive in GBrain to profile contacts
“Basically what I did is I ingested all of my email and I said, Hey, I want you to go through my email a thousand at a time and I want you to draw connections and I want you to build pages for everybody that I know. So basically I have a page on Every, every pe…”
Wilkinson Records Daily Audio to Contextualize AI Agents via GBrain
“And so I record my entire day and then G brain ingests it and it just knows context. And then my open claw agents have full context.”
Wilkinson: AI agent turns executive communications into multiple-choice Telegram prompts
“So, and then it drafts in my voice, different response options. So I can just respond in line and telegram and say one B. And so it turns your entire business into a multiple choice quiz basically. So every day I just get like 20 different messages, their emai…”
Wilkinson: Any software product that is just an API call is cooked
“It's like anything that could be an API call is just cooked. So to me, that's like an API call business. That's like the first thing to get run over.”
Wilkinson: AI Health Agent Detects Nerve Pain Flares via Wrist Temperature
“And then I have this weird viral nerve pain that I get. And it looked through my data over the last five years and said, oh, I've correlated it. And I see that every time you get that flare, Your risk temperature changes for three days before. And so it tells …”
Wilkinson: AI Health Agent Simulates Multi-Doctor Team for Medical Inquiries
“If I ask it a question, it's prompted to use max intelligence. So let's say I said Hey, should I take this medication? It will spin up a rheumatologist, a you know internist, like every type of doctor as a clod expert with extra high thinking, and it will do a…”
Wilkinson: Prompt engineer jobs are unnecessary because AI can interview users directly
“Nobody needs to build the prompts. The AI should just interview you. OpenAI just doesn't want to spend all the compute at this point, but it is fully capable of just interviewing you to figure out what you need.”