📺 Today’s recommended deep-dive video: https://www.youtube.com/watch?v=65IAqRUxg3c

The Autonomous Empire: How Andrew Wilkinson Runs His Life via AI Agents

Entrepreneur and Tiny co-founder Andrew Wilkinson reveals his shift from managing human teams to overseeing a fleet of AI agents that handle everything from SaaS support to personal health diagnostics. By replacing traditional administrative overhead with a $40,000 monthly Claude API bill, Wilkinson is pioneering a new model for the “vibe-coded” autonomous business.

Core Question: Can AI agents realistically replace the administrative and executive functions of a multi-million dollar conglomerate today?

Highlights

• The transition from “traditional business” to “vibe coding,” where vision and execution happen simultaneously.
• How a CFO with zero coding experience built a custom wealth-management platform to replace $100,000 software.
• The “Eye of Sauron” strategy: Using vector databases to spot financial “icebergs” across dozens of portfolio companies.
• Advanced prompting techniques, including “Agent Teams” and “Self-Interviewing” to bypass manual prompt engineering.

⏱️ Reading time: approx. 7 minutes · Saves you about 40 minutes vs. watching.

Want to take notes while watching? Click the image below and let AI Notebook capture the key points for you 👇

The Rise of the Vibe Coder

From Human Overhead to API Calls

The traditional distance between a founder’s vision and the final execution has collapsed entirely due to the rise of “vibe coding.” For years, the creative process in business was compromised by the sheer number of people standing between an idea and a finished product, leading to diluted results and slow iteration cycles.

Wilkinson describes a “heroin-like” addiction to the new workflow provided by tools like Claude Code and Replit. Instead of hiring a team of designers and front-end developers, he now spends his early mornings in a terminal window, directing agents to build entire SaaS products like “Deep Personality” in a matter of manic days. The administrative burden of a company—traditionally handled by support staff and accountants—is being swallowed by agents that never sleep and require no management overhead.

When a support ticket arrives, an agent doesn’t just reply; it analyzes the codebase to fix the bug itself. If the issue is critical, the agent merges the pull request and notifies the user of the fix automatically. This shift turns the business owner into a high-level strategist who manages tokens rather than personalities, allowing for a level of precision that was previously impossible without a massive headcount.

💡 Digging Deeper

Q: What is Harbor, and why use it over standard Claude?
A: Harbor is a GUI harness for agents that provides a visual “org chart” for AI. It allows users to track what different agents (Dev, Marketing, Support) are doing in real-time, providing a more deterministic environment than a simple text interface.

Q: Why build products yourself instead of using your agency, MetaLab?
A: Wilkinson argues that every person added to a project compromises the vision. Vibe coding allows for a “pure pursuit” where the founder can handle design, copy, and code at their own pace without the friction of unreliable human intermediaries.

The Vectorized Holding Company

Building the “Eye of Sauron”

Managing a conglomerate of over 20 businesses creates a data fog that no single human executive can fully penetrate. By training a vector database on every internal meeting, email, and financial statement, Wilkinson has created a searchable brain for his holding company, Tiny. This allows him to query his entire empire for specific insights, such as identifying which CEOs are overspending or which businesses are trending toward a crisis.

The system uses tools like Pinecone to store millions of tokens, making historical data instantly accessible to a Large Language Model.

This isn’t just about search; it’s about automated synthesis and oversight. A CFO who had never written a line of code used these tools to build a custom portfolio tracker that replaced Addepar, a service that costs up to $100,000 per year. The AI assistant now stress-tests their portfolio and suggests rebalancing strategies based on live market data and internal bank pipes.

💡 Digging Deeper

Q: Is software still a good business to start?
A: Wilkinson is bearish, suggesting that “software is free” now. Because anyone can vibe-code an app, moats are disappearing, and pricing pressure is skyrocketing, making many high-quality businesses look like “cigar butts.”

Q: How does the “Eye of Sauron” help with accounting?
A: It can instantly identify anomalies, such as how many accounting staff are employed across the head office or if a specific business unit is drifting from its budget, without needing a manual audit.

The Personalized Life Assistant

Health, Media, and Memory

Beyond business, Wilkinson uses agents like “Ava” and “Mara” to curate his reality and monitor his physical well-being. By ingesting Apple Health data, sleep metrics, and historical medical records into a private database, his “doctor agent” can predict illness flares before they happen. It identifies subtle correlations—like a rise in wrist temperature three days before nerve pain—that a human doctor would likely miss.

His media diet has also been entirely reconstructed. Instead of listening to generic podcasts, he has a custom daily briefing generated by Gemini’s voice engine. The agent scans his newsletters and emails, summarizing only the topics relevant to his life and businesses, while omitting depressing or irrelevant news.

To solve the problem of human memory, Wilkinson uses an app called “Hearsay” to record his entire day. These transcripts are fed into G-Brain (a personal knowledge base), allowing his agents to have full context on every conversation he’s had. This creates a feedback loop where the AI knows his relationships, his goals, and his schedule better than he does.

💡 Digging Deeper

Q: How does the “Team of Experts” prompting work?
A: For medical or complex questions, Wilkinson prompts Claude to “spin up” eight sub-agents (e.g., a rheumatologist, an internist, a researcher) to debate the topic for 10 minutes before providing a final, synthesized answer.

Q: What is the “Self-Interview” trick?
A: Instead of writing a long prompt, ask the AI to “interview me with a shitload of questions” to determine the goal. The AI then asks 5-10 minutes of questions to build its own perfect prompt context.

Key Takeaways

The era of the “Administrative CEO” is coming to an end. Wilkinson’s experience suggests that we are moving toward a world where the most successful entrepreneurs are those who can effectively orchestrate “agent teams” rather than manage large groups of people. The ability to “vibe-code” allows a single individual to act with the force of a mid-sized corporation, provided they can master the data pipelines and prompting logic required to keep the agents aligned.

However, this new reality brings significant risks to the traditional software moat. When software becomes nearly free to produce and replicate, the value shifts toward physical assets (data centers, chips) or businesses with deep physical-world moats, like home services or funeral homes. The “puffs” left in the software “cigar butt” are valuable, but the long-term game is increasingly about controlling the intelligence layer and the data it feeds upon.

Q&A

Q1: How much is Wilkinson spending on AI currently?
He mentioned his family office has shifted from human payroll to a Claude API bill of approximately $40,000 per month.

Q2: Can a non-technical person really build these tools?
Yes. Wilkinson’s CFO, who had no prior coding experience, built a custom wealth-management platform in two weeks using Claude Code.

Q3: What is “G-Brain”?
A project by Gary Tan that creates a vector database for personal knowledge, allowing a user to search through every email and transcript they’ve ever generated.

Q4: How does he automate his social media?
He uses an agent that monitors his Twitter; whenever he tweets, the agent automatically creates a scaled Instagram Story and schedules it via the Instagram API.

Q5: What is the main drawback of current AI agents?
Wilkinson notes they are not yet fully autonomous “CEOs.” They still require step-by-step instructions, similar to managing a “genius baby” or a high-functioning intern with Alzheimer’s.

Q6: What does he recommend for investors right now?
He suggests investing in the infrastructure of AI, specifically data center stocks and TSMC, as software moats continue to erode.

Q7: How does he manage email triaging?
An agent reads his emails and presents him with multiple-choice responses (1A, 2B, 3C) via Telegram, allowing him to run his business through a series of simple taps.