13 Businesses for the Age of AI, According to Greg Isenberg

Greg Isenberg uploaded a 25-minute video yesterday with a claim designed to annoy you: there are only 13 businesses left to build.

Not 13 ideas. 13 categories. His full map of businesses for the age of AI, the ones he says are still worth owning once agents can do most knowledge work.

I watched the whole thing, pulled every number out of it, then went and verified the companies he name-drops. The numbers are wilder than the claim: a data-labeling startup at a $10B valuation, and a sauna club that started in a guy's backyard and now charges $333 a month.

Here's the full breakdown, with my take on where a bootstrapped founder actually fits in.

1. AI-native service firms

This is the one Greg says he's "ideating actively around," and honestly, me too.

An AI-native service firm is an agency where agents do most of the delivery. Not all of it. Greg's bar is around 90% agent work with a human reviewing the exceptions.

His example: a bookkeeping firm for one specific type of business. Agents pull the client records and prepare the books, a human checks the weird transactions, and you charge for the completed job. High margin, and you don't need many clients to make it a real business.

The versions he floats are "logos as a service" and "ads as a service." I broke down how one of these firms actually runs in my piece on the AI-native service firm model if you want the deeper playbook.

2. Offline businesses

The counterintuitive one. In an AI video, Greg's second pick is businesses with a physical front door.

His example is Othership, the sauna and ice bath club. The first version was literally founder Robbie Bent's backyard in 2019: an ice bath, plus a sauna built into his converted garage. Friends came, then a community, then real locations. Today they run four clubs (two in Toronto, two in New York), charge $65 for a drop-in class, and memberships start at $333 a month.

People will pay a premium to do things with other humans. Agents can run the admin, but the room is the product.

Greg's filter for the high startup cost: find the food truck version of your idea first. Othership proved the whole concept in a backyard before anyone signed a lease.

3. Distribution

You can prompt your way to a product in a weekend now. Getting anyone to see it is the hard part, and it's getting more expensive every year.

Greg's data point: when Facebook's ad platform launched around 2009, clicks cost about 5 cents. Today you're paying $1 to $5 depending on the audience and country.

An organic audience gets more valuable every year those ad prices climb. His framework for it is the ACP funnel: audience → community → product. Build the audience first (advertisers literally pay you while you do it), form the community, then build what they ask for.

This is the category I picked, by the way. The podcast and this blog are my ACP funnel, running in public.

4. Proprietary data sets

"Data is the new oil" is the line Greg uses, and the valuations back him up.

Mercor, which connects AI labs with domain experts to train models, raised a $350M Series C at a $10B valuation and has been reported in talks at $20B. Micro1, same lane, raised $100M+ at $4B. These companies are a couple of years old.

You don't need to compete with them. The solo founder version is smaller and more interesting: collect useful niche information that someone's personal agent can't scrape or ask ChatGPT for. Greg pegs that kind of vertical data business at a few million a year in revenue, no lab contracts required.

5. Domain-specific uses

Greg stops and defines this one properly because everyone throws the word around.

The model is the engine. The use is everything around the engine: the tools the model can access, the workflow rules, the memory, the checks. A use takes a general model and turns it into a system that understands how one industry actually works. His example is an operating system for freight brokerage.

His sharpest line in the video: "uses are the new GPT wrapper." Everyone mocked wrappers for 12 months, then the top 10% quietly added industry tools, rules, and memory, and became uses. Harvey and Spellbook did exactly this in legal.

6. Robotics and physical AI

Most people file this one under Jetsons, decades away. It's already hanging off buildings in Manhattan.

Greg points at Ozmo, the window-cleaning robot from Skyline Robotics, which raised $6.5M, then another $9.8M, and cleans high-rise windows in New York today. Building owners pay for clean windows, the company deploys and operates the robots. Beautiful model.

And you don't need the PhD. Greg's point for normal founders: someone has to buy a few of these machines and rent them out, and someone has to sell them. He quotes his friend Brian Norgard's tweet that the best way to sell software might be to sell hardware.

7. Physical products with a fan following

Greg is a coffee nerd, so his example is Fellow. Before Fellow, kettles were just kettles. They designed one for coffee people, who happily pay high prices for well-designed gear, and grew it into a whole product suite.

The other example: the Minimal Phone, built for people who don't want to scroll. Its latest Kickstarter pulled in over $650,000 from more than 1,000 backers. Apple is never making that phone, which is exactly the opening.

The new unlock is that AI makes niche hardware design actually feasible. You can prototype on a Bambu printer and manufacture a product that was too niche to exist in 2020.

8. Compute and energy

Every agent needs chips, electricity, cooling, and somewhere to run. Whoever owns those scarce inputs gets paid no matter which model or app wins.

The numbers here are the biggest in the video. Crusoe raised a $3.9B Series F for its vertically integrated AI infrastructure platform at a $30.9B valuation. Zach Dell (Michael Dell's son) raised $1B for Base Power at a $13B valuation, with 23,000+ home batteries already installed across Texas and Illinois.

Not exactly a bootstrapper category. But Greg's point stands: if you can touch this supply chain anywhere, even on the sales side, you're positioned well.

9. Health, longevity, and care

The underrated ICP here: older adults. Young founders build for other young founders, and almost nobody builds genuinely good products and services for people over 65.

Greg sees two lanes. Products with a high NPS for older adults, and care businesses: agencies for caregivers, services that make aging easier. Hard businesses to run, real demand, and you're doing something that matters.

On longevity, Huberman became the guy for health and Bryan Johnson tapped the "don't die" movement. Greg's point is that the niches underneath them are still unclaimed.

10. Marketplaces and social networks

Nobody builds these anymore. That's exactly why Greg likes them.

When he landed in Silicon Valley around 2013, matching supply and demand was the thing: Uber, Thumbtack, Instagram. Then everyone decided liquidity was too hard (it is more art than science, he admits) and moved on.

But the comeback is visible. Marc Lou built TrustMRR, a marketplace for buying and selling startups, as basically a solo founder. Personal agents add a second reason to care: they'll need marketplace access to get things done for their owners. I wrote about what that means for builders in selling to AI agents.

11. Real assets

The least sexy item on the list: small industrial workshops rented to local trades. Electricians, cabinet makers, plumbers.

Agents can't rent themselves a workshop. The physical world still needs locations, warehouses, and property, and the trades that keep everything running need somewhere to work from. Owning that real estate is a quiet, durable position while everyone else fights over software margins.

12. Vertical agents

A vertical agent owns one recurring job in one specific industry, end to end.

Greg's example: purchasing for independent restaurants. The agent checks stock, compares supplier offers, asks the owner for approval, places the order. Every week, forever.

People have been building these for the past 12 months and he still thinks the opportunity is wide open, because every industry has a dozen of these recurring jobs. The pattern is the same one behind most business automations that actually stick: pick one painful recurring workflow, automate 90% of it, keep a human on the approval step.

13. Security

Greg's closer, and his darkest prediction. He expects a major personal-agent startup to get hacked, with user data dumped on the internet. His words: "it's going to happen more and more."

Two lanes here. Software security: companies like 1Password and Cloudflare, plus new stuff like agent access control (his example is letting an accounting firm's agent read invoices and draft entries, and nothing else). And physical security: he mentions an Uber-for-bodyguards type service with bulletproof cars.

People already pay 1Password to protect their logins. After the first big agent breach, they'll pay a lot more, for a lot more. Grim, but probably right.

How I'd actually pick one

Thirteen categories is a map, not a to-do list. You pick one.

My honest read for a bootstrapped founder at $5K to $50K MRR: start with number 3. Distribution feeds every other category on this list. An audience makes your service firm cheaper to launch and your next product easier to validate, because the buyers are already watching you build it.

Greg says the same thing in the video, almost in passing: build the organic audience before the offline business, before the service firm, before everything.

Then stack a delivery business on top. For most solo founders reading this, that's number 1 or number 12 (service firm or vertical agent), because both can start with one client and zero capital.

The worst move is sampling four categories at once. I've watched founders in the Club try that. The ones who hit $100K MRR picked one lane and stayed in it for 18 months.

FAQ

What are the 13 businesses for the age of AI?

Greg Isenberg's list: AI-native service firms, offline businesses, distribution, proprietary data sets, domain-specific uses, robotics and physical AI, physical products with a fan following, compute and energy, health and longevity and care, marketplaces and social networks, real assets, vertical agents, and security.

What is a domain-specific use?

A use is everything built around an AI model: the tools it can access, the workflow rules, the memory, and the checks. The model is the engine, the use is the rest of the car. A domain-specific use turns a general model into a system that understands one industry, like freight brokerage or legal work. Harvey and Spellbook are the reference examples in legal.

Do I need to be technical to build in these categories?

No. Several categories need no code at all: offline businesses, distribution, real assets, and the sales side of robotics. For the technical ones, Greg's point about Ozmo applies everywhere: every robotics and AI company needs people to sell, rent, and operate the product, not just build it.

Which one should a solo founder start first?

Distribution, then a service business on top. An organic audience costs time instead of money, pays you through advertisers while it grows, and de-risks whatever you build next. An AI-native service firm or a vertical agent can then launch with one client and near-zero capital.