Startup Ideas Podcast Host — How Will People Make Money in the AI Age?

In one sentence

The "agentic era" creates a world of abundance where profit is found only in owning scarce resources, specialized human experiences, and the infrastructure (harnesses) that directs AI behavior.

Why read this (and who it's for)

This summary is for entrepreneurs, investors, and builders who are feeling overwhelmed by the rapid pace of AI development and are struggling to find a "moat" in a world where anyone can prompt a startup into existence. It solves the problem of "commodity creation"—the realization that if an AI can build the product, the product itself is no longer the value. After reading this, you will be able to identify which business models are becoming obsolete and which 13 categories offer genuine, defensible opportunities in an AI-saturated economy.

The big idea

The host argues that we are entering an "agentic era" where AI agents will handle the bulk of digital labor. In this world, the cost of generating content, code, and basic services will trend toward zero. Consequently, the traditional "software as a service" model—where a company simply provides a convenient interface for a common task—will collapse because the "convenience" will be provided for free by personal agents.

To survive and thrive, businesses must pivot toward scarcity. The host identifies 13 specific categories where value remains high because the AI cannot easily replicate the underlying asset. This includes physical reality (real estate, robotics), scarce inputs (energy, proprietary data), and "harnesses" (the rules and workflows that govern how an AI behaves in a specific industry). The logic is simple: when the output of AI is infinite, the value of the input (the data, the power, the physical space) becomes the only thing people will pay for.

Key ideas

AI Native Service Firms

• What it is — Businesses where agents perform 90% of the work, but a human remains in the loop to review "exceptions" and finalize the output.
• Why it matters — It allows for extremely high-margin businesses (like bookkeeping or creative agencies) because you can scale without a linear increase in headcount.
• In the book — The host suggests a bookkeeping firm where agents prepare the books and humans review the final monthly reports, charging for the completed job rather than the hour of work.

Offline Businesses

• What it is — Businesses that provide physical experiences, community, and wellness in the real world.
• Why it matters — As digital life becomes saturated with AI-generated content, people will pay a premium for physical presence and human connection.
• In the book — The host cites "Othership," a neighborhood sauna club in Toronto/New York, as a model for selling a membership to a curated physical experience.

Distribution (The ACP Funnel)

• What it is — The process of building an Audience, then a Community, and finally a Product.
• Why it matters — In an era where anyone can build a product instantly, the hardest thing to build is a loyal audience. Owning the distribution channel is the ultimate moat.
• In the book — The host argues that owning "meme pages," YouTube channels, or Instagram audiences allows you to get paid by advertisers to build an audience before selling them a product.

Proprietary Data Sets

• What it is — Identifying and acquiring data that the big AI labs do not currently have access to.
• Why it matters — Data is the "new oil." If you own the data that trains the models, you hold the keys to the kingdom.
• In the book — The host points to companies like Mercore and Micro One, which are selling data to big labs for massive valuations.

Domain-Specific Harnesses

• What it is — A "harness" is the system of tools, workflow rules, memory, and checks that sit around an AI model (the "engine").
• Why it matters — While "GPT wrappers" were easily replicated, "harnesses" that understand the specific nuances of an industry (like legal or freight) create lasting value.
• In the book — The host compares the model to a car engine and the harness to the body, steering wheel, and floor mats. A harness is what makes a general model useful for a specific job.

Robotics and Physical AI

• What it is — Businesses that deploy physical machines to perform real-world tasks.
• Why it matters — There is a massive opportunity to "sell software by selling hardware."
• In the book — The host mentions "Osmo," a company doing window cleaning as a service, and suggests that even non-technical people can find success in the sales or rental side of robotics.

Physical Products with a Fan Following

• What it is — Creating niche, high-quality physical goods for a specific "fan" community.
• Why it matters — AI can help design products that were previously too niche to be profitable, but the value lies in the "fan" loyalty.
• In the book — The host uses "Fellow" (specialty coffee kettles) and "The Minimal Company" (phones for people who don't want to scroll) as examples of capturing a specific niche.

Compute and Energy

• What it is — Owning the "scarce inputs" of the AI era: chips, electricity, and cooling.
• Why it matters — Every agent requires power and processing; therefore, the infrastructure of the AI era is a foundational business.
• In the book — The host cites Crusoe (AI infrastructure) and Michael Dell’s son’s energy company as examples of the massive capital flowing into these scarce inputs.

Health, Longevity, and Care

• What it is — Products and services for the aging population and the "don't die" movement.
• Why it matters — This is a massive, underserved demographic with high demand for high-quality, high-NPS services.
• In the book — The host suggests looking for "high-quality" service businesses, such as caregiver agencies or longevity-focused products.

Marketplaces and Social Networks

• What it is — Platforms that facilitate "liquidity" between supply and demand.
• Why it matters — While hard to build because they require both sides of a market to show up, they offer a massive moat and provide incredible data.
• In the book — The host mentions Mark Liu's "Trustrr" (a marketplace for buying/selling startups) as a modern example of this returning to the forefront.

Real Assets

• What it is — Owning physical property like industrial warehouses or small workshops for trades.
• Why it matters — The "new world" still requires physical locations and property.
• In the book — The host suggests renting out small industrial workshops to local trades like plumbers or cabinet makers.

Vertical Agents

• What it is — Giving an agent responsibility for a recurring, specific job in a niche industry.
• Why it matters — It moves beyond "chatting" with an AI to "delegating" a workflow.
• In the book — An example is an agent for "purchasing for indie restaurants"—checking stock, comparing offers, and placing orders.

Security

• What it is — Protecting data and workflows from the increased risks of the agentic era.
• Why it matters — As agents gain access to sensitive data (like accounting invoices), the risk of hacks and leaks increases exponentially.
• In the book — The host suggests two types: software security (like 1Password) and "agent access control" to ensure data doesn't leak during automated workflows.

Frameworks & tools

The ACP Funnel (Audience, Community, Product)

Use this model to build a business that is "anti-fragile" to AI competition.

1. Audience: Start by creating content that provides useful, niche information. Get paid by advertisers to build this initial reach.
2. Community: Turn that audience into a community where people interact and share interests.
3. Product: Only after you have a community do you build the product they actually want.

The Harness Framework

When building an AI application, do not just build a "wrapper." Build a "harness":

1. Identify the Industry: Choose a specific niche (e.g., Freight Brokerage).
2. Define the Workflow: Map out the exact steps a human currently takes to do the job.
3. Add Memory: Ensure the system remembers past interactions and specific client preferences.
4. Add Rules: Create the "checks" and "guardrails" that ensure the AI follows industry-specific regulations.
5. Integrate Tools: Connect the AI to the actual software tools used in that industry.

Memorable examples & stories

The Backyard Sauna (Othership)

The founder of Othership started by creating a communal sauna in his own backyard to test the model before scaling to New York. Lesson: Start with the smallest possible version of an offline business to prove the model before committing to high-cost real estate.

The "Car" Analogy for AI

The host compares an LLM to a car engine (the power) and the "harness" to the rest of the car (the body, steering, and seats). Lesson: The value is not in the engine (the model) because everyone has access to the engine; the value is in the car (the system) you build around it.

The "Don't Die" Movement

The host references Brian Johnson and the "don't die" movement as a cultural shift toward longevity. Lesson: There is a massive, underserved market in health and longevity that can be tapped into by "niching down" into specific needs for older adults.

Key quotes

"In a world where agents exist, abundance are going to exist... owning these assets... are going to put you just in a better position." Why it matters: This frames the AI era not as a "scarcity of opportunity" but as a "scarcity of assets."

"I think that the best way to sell software is by... selling hardware." Why it matters: Physical presence and hardware provide a moat that pure software cannot replicate.

"Harnesses are the new GPT wrappers." Why it matters: It highlights the shift from simple UI-over-LLM to complex, rule-based industrial systems.

How to apply it

This week

• Identify your "Scarcity": Look at your current business or skills. Which parts of your work are "commoditized" (could be done by an agent) and which are "scarce" (require your specific data, your physical presence, or your unique community)?
• Audit your "Harness": If you have a tool or service, map out the "rules" and "workflow" that make it work. If you can't describe the rules, you don't have a harness; you have a wrapper.
• Research a "Fan Following": Identify a niche you are obsessed with (like the host's coffee example) and look for a physical product that doesn't yet exist for that niche.

Ongoing

• Build the ACP Funnel: Every time you share information online, ask: "How does this build my audience?" and "How does this lead to a community?"
• Think in "Verticals": When looking for AI opportunities, stop asking "What can AI do?" and start asking "What specific, recurring job in [Industry X] can an agent own?"
• Monitor the "Scarce Inputs": Keep an eye on energy, compute, and proprietary data markets as these will be the foundational layers of the new economy.

Common pitfalls

• The "Wrapper" Trap: Many people think building a simple UI over a model is a business. The host warns that 90% of these will fail because they lack a "harness" or a "moat."
• Underestimating Physicality: People often assume the AI age means everything becomes digital. The host argues the opposite: the more digital the world becomes, the more valuable physical assets (real estate, robotics, community) become.
• Ignoring the "Human" Element: Don't assume agents will do 100% of the work. The most profitable "AI Native" firms will still have humans reviewing the "exceptions" and high-level strategy.

Connects to

• The Age of AI (Kissimmee, et al.) — Explores the technical shifts in how we interact with machines.
• Zero to One (Peter Thiel) — Relevant to the "moat" and "proprietary" thinking regarding data and distribution.
• The Almanack of Naval Ravikant — Specifically the concepts of leverage (code and media) and owning equity in the things you build.

If you remember only three things

• Focus on Scarcity: In an era of AI-generated abundance, profit is found in owning scarce inputs (data, energy, physical assets).
• Build Harnesses, not Wrappers: Value is found in the rules, workflows, and industry-specific memory you build around the AI engine.
• The ACP Funnel: Your greatest moat is your audience. Build a community first, then a product.

Internalize it

• Reflection questions
  • If an AI agent could do 90% of my current job today, what is the 10% of "exceptions" or "strategy" that only I can handle?
  • Which "scarcity" (data, physical space, community, or specialized knowledge) do I currently own or have access to?
  • If I were to build a "harness" for my