A solo podcast host scrolls through one whiteboard doc, thirteen cards deep, arguing that every durable business left to build either delivers work through agents or owns the thing agents still can't replace.
Posted
today
Duration
Format
Listicle
educational
Views
5.3K
295 likes
57 · 43
Big Idea
The argument in one line.
Thirteen business categories split into two camps: some use agents to deliver the work faster, and the rest own the scarce inputs, like data, compute, trust, and physical presence, that no agent can manufacture.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A solo founder trying to decide which side of the agent economy to build on before committing capital or time.
Someone evaluating a specific business idea who wants a concrete example and a named flow diagram instead of abstract AI hype.
A builder open to both software and physical businesses, a sauna club, a workshop rental, a fleet of window-cleaning robots.
SKIP IF…
You want one detailed business plan handed to you; this is thirteen one-page sketches, not a single blueprint.
You're only interested in pure software plays with no physical, regulatory, or trust-building component.
TL;DR
The full version, fast.
The video sorts thirteen viable business categories for an agent-filled economy into two groups. Some use agents to deliver the work itself: AI-native service firms, domain-specific harnesses, and vertical agents that handle one recurring job like restaurant purchasing. The rest own what agents cannot replace: proprietary data, organic audiences, physical products with a devoted following, robotics, compute and energy, real estate, and security. Every category gets the same format, one named example, a simple flow diagram, and an honest admission of the hard part, from manufacturing headaches to long infrastructure buildouts. The throughline is that reaching an audience and owning scarce inputs matter as much as the AI layer itself.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
States the premise outright: only thirteen businesses are worth building in a world full of AI agents, and the goal isn't escaping some 'permanent underclass' so much as understanding which assets are worth owning.
01:15 – 02:57
02 · AI-Native Service Firms
Defines the category as a service business where agents handle most of the delivery and a human reviews exceptions, using a bookkeeping firm billed by the completed job as the running example.
02:57 – 05:41
03 · Offline Businesses
Uses the neighborhood sauna club Othership to argue people still pay to physically go somewhere and be with others, and that the cheapest way to test the concept is a backyard-sized MVP before a full lease.
05:41 – 07:34
04 · Distribution and Organic Audiences
Tracks rising ad costs since 2009 to argue for building an owned audience first, introducing the ACP funnel: audience, then community, then the product the audience already asked for.
07:34 – 09:08
05 · Proprietary Data Sets
Argues the valuable thing is information a competitor can't scrape, using a 'what actually fixed this broken dishwasher' dataset as the example of licensing verified repair outcomes to service networks.
09:08 – 12:16
06 · Domain-Specific Harnesses
Explains a harness with a car analogy, the model is the engine, the harness is everything else, the body, the steering wheel, the rules, then argues GPT wrappers survived by turning into harnesses.
12:16 – 14:01
07 · Robotics and Physical AI
Points to Osmo's window-cleaning robots to argue the real business can be deploying and operating someone else's equipment rather than inventing the hardware.
14:01 – 16:34
08 · Physical Products With a Fan Following
Uses the Fellow coffee kettle and the Minimal Company's distraction-free phones to argue a well-designed niche object can earn a devoted, repeat-buying fan base.
16:34 – 17:31
09 · Compute and Energy
Cites Crusoe's $3.9 billion raise for AI infrastructure to argue that chips, electricity, cooling, and physical servicing are scarce inputs worth owning a piece of.
17:31 – 19:30
10 · Health, Longevity, and Care
Argues older adults are an underserved ICP, sketching an in-person strength and mobility studio tested on a small paid cohort before a dedicated location.
19:30 – 21:04
11 · Marketplaces and Social Networks
Cites Mark Liu's TrustMRR.com as a fresh marketplace example while admitting that liquidity, having both sides of a market present at once, is more art than science.
21:04 – 21:46
12 · Real Assets
Uses small industrial workshops rented to electricians, cabinetmakers, and plumbers to argue an agent can find the right property but can't generate more of that location.
21:46 – 22:18
13 · Vertical Agents
Sketches an agent responsible for one recurring job, restaurant purchasing, that starts with draft orders in one category and expands only after its decisions are verified.
22:18 – 24:08
14 · Security
Argues demand for security rises with AI adoption, using agent access controls for an accounting firm, read invoices allowed, draft entries logged, moving money gated behind human approval, as the example.
24:08 – 25:17
15 · Closing thoughts
Recaps the full map of thirteen businesses as the ones worth paying attention to and closes by thanking the audience and asking for comments.
Atomic Insights
Lines worth screenshotting.
Facebook ad clicks cost about five cents in 2009 and run one to five dollars today, which is the stated reason to own an organic audience instead of renting reach.
A domain-specific harness is not the AI model itself, it is the industry tools, workflow rules, memory, and approval checks wrapped around a general model.
The startups that survived the GPT-wrapper era did it by adding industry-specific tools and memory until a thin wrapper became a full harness.
Proprietary data sets are valuable because they hold information a founder cannot simply scrape, not because the pile of documents is large.
A window-cleaning robotics company can be built around deploying and operating someone else's equipment rather than inventing the robot.
Niche hardware with a devoted following, like a specialty coffee kettle, can take years to earn broad awareness while still selling well inside its niche.
Compute and energy are scarce inputs every agent depends on, which makes owning cooling capacity or specialized servicing a smaller, viable entry point than building a data center.
Older adults are described as an underserved customer base for health and longevity products specifically because younger founders rarely build for them.
Marketplaces and social networks are hard to start because liquidity, having both supply and demand present at once, is treated as an art rather than a formula.
A vertical agent earns trust by starting with draft orders in one category and expanding only after its decisions are verified.
A security business can start with one narrow, logged workflow, like reading invoices, before it's ever trusted to move money.
A small industrial workshop is framed as valuable because an agent can find the right property but cannot generate more of that location.
Takeaway
Thirteen Ways To Own What Agents Still Need
AGENT-ERA BUSINESS MAP
Across all thirteen categories, the winning move is either operating the agent that delivers the work, or owning the data, trust, physical space, or scarce input that no agent can replace.
02AI-Native Service Firms
An AI-native service firm doesn't need agents to handle all the delivery, letting them do most of it while a human reviews exceptions still counts.
Billing by the completed job instead of by the hour works once agents absorb most of the production time.
Start in one niche with a handful of clients so you can learn the recurring exceptions before trying to scale the service.
03Offline Businesses
People still pay to physically go somewhere and be around other people, which AI cannot replicate no matter how good it gets.
Prove a communal or experience-based concept in the smallest possible footprint, a backyard or a single room, before signing a lease.
Agents can run the admin side of an offline business, but the room, the staffing, and the experience itself are still what customers are buying.
04Distribution and Organic Audiences
Ad costs have climbed from pennies a click to multiple dollars, making an owned, organic audience more valuable than renting reach.
Build the audience first through something like a newsletter, let it tell you what to build, then sell what it already asked for.
An audience that comes back on its own is a hedge against depending on any single ad platform or algorithm.
05Proprietary Data Sets
The valuable thing isn't a bigger pile of documents, it's information a competitor genuinely cannot scrape from the open web.
Partner with the people who generate the raw information, like technicians doing real repairs, instead of trying to generate it yourself.
Test data collection at a small, paid scale before committing to the harder problems of coverage and keeping records current.
06Domain-Specific Harnesses
A harness is the industry tools, workflow rules, memory, and approval checks wrapped around a general model, not the model itself.
Narrow scope wins early: build one workflow on top of an existing agent framework rather than a whole industry platform.
The hard part of a harness business is handling rare exceptions and proving the system holds up on real, messy data.
07Robotics and Physical AI
You don't have to invent the robot to build the business; deploying and operating someone else's equipment is a legitimate company on its own.
Prove the model on one physical location, like one building's windows, before trying to scale the deployment.
Hardware businesses carry real-world constraints software doesn't: safety, equipment economics, and site access all have to work before the business does.
08Physical Products With a Fan Following
A product can be relatively easy to design and still take years to earn the kind of following that drives repeat sales.
Target one object a niche community already obsesses over rather than a broad, generic product category.
AI can help design the product, but community trust and perceived quality are what actually earn the buyer's choice.
09Compute and Energy
Every agent ultimately needs chips, electricity, cooling, and somewhere physical to run, which makes those inputs scarce by default.
A smaller entry point than building infrastructure outright is servicing one specific physical bottleneck, like cooling capacity, that customers will pay to remove.
Capital requirements and long delivery timelines are the real barrier in this category, not the underlying technical idea.
10Health, Longevity, and Care
Information about health and longevity is getting easier to access for free, but hands-on delivery of care still has to happen in person.
Older adults are a large, underserved customer base specifically because younger builders rarely design products or services for them.
Test a new care or coaching concept with a small paid group in rented space before committing to a dedicated location.
11Marketplaces and Social Networks
The matching logic behind a marketplace is replicable, but the people and transaction history on both sides of it are not.
Start manually in one region or one process before trying to automate matching at scale.
Having both supply and demand present at the same time is closer to an art than a formula, which is why liquidity is the hard part.
12Real Assets
An agent can identify the right property, but it cannot create more of a specific physical location, which is why location-based assets hold value.
Validate real demand from tenants or trades before committing to purchasing or renovating a physical space.
Customers are paying for access to a location that works for their specific business, not for the property itself.
13Vertical Agents
A vertical agent takes ongoing responsibility for one recurring job in a specific industry rather than acting as a general assistant.
Start with draft orders or draft decisions in a single category, and only expand the agent's authority after its judgment is verified.
Changing prices, rare substitutions, and costly errors are the operational risks that make a vertical agent harder to trust than it looks.
14Security
Demand for security rises alongside AI adoption because agents increasingly get access to money and sensitive systems.
Start with one sensitive workflow, like reading invoices, and expand only after trust is established through the client's existing IT provider.
Customers are paying to delegate real work without handing over unlimited authority, a narrower and more sellable promise than general security.
Glossary
Terms worth knowing.
AI-native service firm
A service business where agents handle most of the delivery work, like preparing books or drafting content, while a human reviews exceptions and still sells the completed job.
Harness
The industry tools, workflow rules, memory, and approval checks built around an AI model that let it reliably do a specific industry's job, as distinct from the model itself.
ACP funnel
A build sequence of audience, community, product: grow an audience first, turn it into a community, then build the product that audience is already asking for.
Vertical agent
An AI agent given ongoing responsibility for one recurring job inside a specific industry, such as restaurant purchasing, instead of acting as a general-purpose assistant.
GPT wrapper
An early-era AI startup built as a thin interface on top of one model, with no proprietary data or industry-specific workflow logic layered on top.
Resources
Things they pointed at.
03:03productOthership (neighborhood sauna club)
11:21productHarvey and Spellbook (legal AI harnesses)
22:28tool1Password and Cloudflare (security tooling examples)
Quotables
Lines you could clip.
07:28
“Data is just so valuable. Data is the new oil.”
short, punchy claim that reframes proprietary data as the scarce resource of the AI era→ TikTok hook↗ Tweet quote
06:27
“I think that owning organic audiences is so, so, so important.”
the repetition lands it as a standalone thesis on distribution→ IG reel cold open↗ Tweet quote
10:36
“I believe that harnesses are the new GPT wrapper.”
contrarian-sounding reframe of a trending term, works without setup→ newsletter pull-quote↗ Tweet quote
13:11
“The best way to sell software is by selling hardware.”
borrowed one-liner with a clean reversal structure→ TikTok hook↗ Tweet quote
24:21
“These are the only businesses that are left.”
closing thesis statement, works as a title card line→ IG reel cold open↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphoranalogystory
Today, we're gonna talk about the only businesses left to build in the agentic era. Yes, I think that there's only 13 businesses to be building in the world of agents. Now, I don't subscribe to the people who say there's gonna be this permanent underclass and you just gotta create businesses or else you're kind of messed up for the future.
I don't think it's gonna be as dire as that, but I do think that in a world where agents exist, abundance are gonna exist, owning these assets i mean are gonna put you just in a better position so today what are we gonna do we're gonna go through the 13 businesses that are left to be built by the end of this you're going to understand all 13 pretty well why why there's an opportunity why they're going to exist in a world of agents a couple ideas so but you know Grab your coffee, grab your tea, strap in because hey, this is the stuff that if you're a teenager, if you're in 20s, you're 30s, 50s, 70s, this is good stuff to know because the world is changing and it is changing fast.
The first type of business that you'd want to own in an AI agent world is AI native service firms. What is that? Well, an AI native service firm is when the agents do more of the delivery.
Now, I've seen people say AI native service firms is when they do all of the delivery. I don't believe that's the case. I think you can still have the agent doing 90 % of the work and still be an AI native firm.
So as an example, you can create a bookkeeping firm for one type of business where you have client records, the agents prepare the books, and then the human reviews exceptions, and then you complete monthly books. And you charge for the completed job. It could be a monthly fee, but it could also be, hey, every time we...
complete these books, it's going to cost XYZ. Why is owning this business good? Well, it's high margin.
You don't need that many clients. The why now exists when you have Astra, 5 .5, some of these great models that are coming out soon.
they could get this incredible amount of output. So I'm really interested in AI -native service firms. I did a whole episode on one recently.
I'll include it in the description around creating a creative agency. So you can imagine doing logos as a service or ads as a service, these sorts of things. Really like these businesses, super high margin, and something I'm ideating actively around.
The second is just generally offline businesses. People... Yes, it's an AI age.
But people actually want to pay to go and do something with other people. So one example of this is this company called Othership. I heard about them from the founder actually in Toronto where they started.
And they've got this neighborhood sauna club where you walk into the sauna and it's like... huge and there's other people and there's music and there's someone who can help you do cold plunges and it's just this incredible environment and tea. They basically leveled up the experience of a communal sauna.
And I think now more than ever, people want this sort of thing. And it's not cheap. It's really not.
The visit itself could be, I don't remember exactly what it is, but it's not cheap. But it's an incredible experience and they sell a membership. So I think businesses like this that are onto trends.
Here, the trends obviously are wellness, community, and you're just seeing them scale. They've been scaling in New York. And I love these types of businesses, especially ones that have a membership model.
So I think you're going to see a lot more like this. And you can run, of course, you can have agents do a lot of the admin, but the room, the experience is what people are. Now, the hard part about offline businesses, obviously, is that it costs a lot more money to start an offline business than an AI -native service firm, especially if you're creating an offline business in a place like Manhattan or Toronto, a place with a high amount of real estate.
So how I think about offline businesses is, what is the food truck version of your business. What do I mean by that?
If you're going to start a restaurant, sometimes the best thing to do is to start a food truck version of it because it's just so much cheaper and you can prove out that model that way. So the MVP, what does that look like? And I actually think, now that I'm remembering, I remember speaking to the founder about this, the first version of Othership was literally in the founder's backyard.
He literally made this in his backyard. And obviously that didn't cost too much money. He started inviting friends, community members, and then he decided to get a place.
So take it one step at a time. Third is distribution. So we live in this world now where you prompt a few times and all of a sudden you basically have a startup.
But getting people to the products that you're building or people are building is... harder than ever, honestly. And my belief is that the price of reaching people is going to get higher and higher.
And actually, if you take a step back and look at what people used to charge for, for example, a Facebook ad. In 2009 or whatever, when the Facebook ad platform came out, they were literally charging five cents a click. Now the average click, you know, I don't have the exact data in front of me, but we're talking a dollar, two dollars.
even more than that, $3, $4, $5, depending on who you're trying to target and which country you're trying to target. So I think that owning organic audiences is so, so, so important. And I think that even before you end up...
If you end up wanting to create an offline business, if you end up wanting to create an AI native service firm, starting with the distribution, building organic audiences, the cool part about this is you can get paid. to generate useful niche information. You get paid by advertisers to basically build your audience and then you go and build the products that they ask for.
This is something I call the ACP funnel, audience community product. First you start with the audience, then you build the community, then you build the product. This is something that is a business model.
I think people who own multiple meme pages, multiple YouTube channels, Instagram audiences, those people are going to do well in this era. So I think number three, distribution.
The fourth, which is actually not talked about that often, is proprietary data sets. So I think that what you're starting to see is that data is just so valuable. Data is the new oil.
all these companies like Mercore, Micro One, who are selling data to the big labs and are getting insane valuations, billions of dollars, hundreds of millions of ARR. It's absolutely insane. That's at the big idea level, which is basically pay people for data, get data that the big models don't have, and then...
basically mark it up to the big labs. But if you're just a solo founder and you don't want to go build something that big and you want to create a business that might generate a few million dollars a year, how do you just get very useful information that people cannot get by using their personal agents or going to chat GPT, that sort of thing.
And I think thinking about that is a really, really important thing. I think there's going to be this gold rush for getting and acquiring data.
I think we're in the first inning of that. And this is just a business that if you own data in this AI world, you're looking really good. The fifth is domain -specific harnesses.
I feel like a lot of people throw the word harness around and honestly people don't really understand what it is. So I'm going to take a second to explain to you just as clear as I can what a harness really means and how is it different than a LLM model. So you can think of the model or I think of the model as basically the engine of the car.
how fast it goes. And I think of the harness as the tools it can access, the workflow rules, the memory, the checks, basically everything around the engine. And if we use the car analogy, we're talking about the body, we're talking about the steering wheel, we're talking about the floor mats, we're talking about the color of the car, the paint.
the electrical, everything else. That's the easiest way to explain it. It's not a perfect way of explaining it.
I am missing a few things, but that's the general way to think about it. So if you can take a general model and turn it into a system that understands how an industry works, for example, the example I was using is the operating system for a freight brokerage agent, you're going to be able to have a lot of value. We talked about the gold rush around proprietary data sets.
I think you're also witnessing now this gold rush around harnesses. I believe that harnesses are the new GPT wrapper. Two years ago, we saw a lot of people create GPT wrappers.
Initially, for about 12 months, people were saying, GPT wrappers were useless. People weren't going to be able to create any sort of lasting value with it or any value with it.
And we saw quite the opposite. Well, we saw a bunch of GPT wrappers go nowhere in the same way that startups 90 % fail. But I think the top 10%, a lot of what happened is they turned into harnesses.
So they were... They started as a wrapper focused on one model. And as that model got better, that service got better.
But they started adding things like here. They started adding industry tools and workflow rules and memory. And then they became harnesses.
An example of that would be, you saw a lot of that in the legal space, Harvey, Spellbook, I think it's called. So I expect to see a lot more of that. I think there's a lot of opportunity.
And by the way, comment on YouTube for if you ever want me to go deeper into any one of these, just let me know. And I can create dedicated videos with some ideas and that sort of thing. And also, while you're at it, like this.
And if you're on Spotify or Apple, please rate it five stars because we've got a few low stars. So from back in the day. So let's try to get it up.
The sixth is robotics and physical AI. So this one, a lot of people think that it is so far away in the future that it's like the Jetsons, just like really in the future. But I don't think people realize that we are like almost there.
I saw this new product called Osmo. I think they raised like six or seven million dollars recently. And they are literally doing window cleaning as a service.
So building owners pay for clean windows and then you deploy and operate the equipment or they deploy and operate the equipment. And it's just a beautiful business, right? The hard part is obviously building these companies, right?
But I think that you're going to see a lot more robotics and physical AI.
I saw a really good tweet by my friend Brian Norgard and he said, I think the best way to sell software is by selling hardware. And I think you're going to see a lot of that. So a lot more businesses like that.
And for the people listening who are like, I'm not going to go and create a robotics company. What do I know about all this? PhD stuff.
Well, I think there's going to be an opportunity for you to maybe buy a few and rent it to people. Or if you're into sales, you can go and sell them. There's a lot of opportunity.
It doesn't necessarily mean you need to be an Elon Musk character to go and figure out what these things are and build them. You can also be on the sales, marketing side, and I think there's going to be a ton of opportunity. Number six.
Number seven. Physical products with a fan following.
I'm sure a lot of you know that I love coffee and I'm a bit of a nerd like that. Specialty coffee equipment is an example of people get so nerdy about it and they're willing to pay high prices for well -designed things.
And when Fellow came out, like there was nothing like this. A lot of the, this is a kettle. So a lot of the kettles didn't really feel like they were designed for coffee people.
And then this came out and it started, you know, they started crushing it. Now they have a huge product suite of different products. Like I got a Fellow grinder back there.
By the way, not affiliated with Fellow. Just, again, like a fan. And I think there's a ton of opportunity to just create these niche products.
I also think with just AI, like for example, Astra and the latest new models, you can actually design hardware and physical products that were too niche before. You can go and do that now. So if you have the insight for a particular fan following or a particular ICP, you can go and buy, I think the device is like the bamboo.
It's a Chinese device and you can go and like prototype some of these new products and then, you know, obviously mass produce it, sell it. Another friend of mine or a friend of mine started something called the Minimal Company. And those are phones that are focused on people who don't want to scroll.
And I think they did millions of dollars on their Kickstarter. Again, that's like a fan following. Apple's not creating that product.
So there's a lot of opportunity. I think there's a lot of pent -up demand for specific products here. And especially the ones that can go viral in that niche.
If you really understand a niche, now's a really good time. I really do want to do an episode around physical products, how to use AI to create them. So hopefully you guys are into that.
Number eight, compute and energy. We can't forget that every agent needs chips, electricity, cooling, and somewhere to run. So if you own that scarce input, hey, that is a good place to be.
Saw just the other week. Crusoe raised $3 .9 billion for its vertically integrated AI infrastructure platform. So you're just seeing that.
I think Michael Dell's son is raising billions of dollars for his energy company. You're just seeing people need compute, people need energy. These are the scarce inputs of our time.
So if there's any way that you can go and be involved in some of these businesses, probably do pretty well. Number nine, health, longevity, and care. I've talked a lot about it on the Startup Ideas podcast, on this podcast, about older adults.
I think older adults are missing. It's just like an ICP that a lot of people, especially younger people, are not creating for. But they need products, they need services, and any way we can make their lives happier and healthier I think would be a great thing.
So I think there's tons of ideas, products, services that are actually high quality, that have a high NPS, that are exceptional, that you could be creating. And then also from a service business perspective, care. There's a lot of opportunity there.
Helping older adults and just owning agencies in that space. For example, caregivers and things like that. A lot here.
And obviously hard businesses to do. A lot of those service businesses to run. But great businesses and also you're doing a good thing by caring for folks.
And then the longevity piece, I feel like Brian Johnson really kicked off this whole don't die movement. Well, he didn't kick it off. He kind of tapped into it.
He was like, I'm the guy. So health, maybe that was Huberman. Longevity, maybe that was...
Brian Johnson. There's still opportunity to go and be the gal or guy in that space if you just go down a niche, niche down. But point is, there's a lot of opportunity here, a lot of business to be created, and I think a lot of demand to tap into.
The tenth is marketplaces and social networks. Now, a lot of people are not creating these products and startups anymore.
When I first came to Silicon Valley, it was just the thing. We're talking 2013, 2014, 2015, that era. Creating things like Uber, which matched drivers and customers.
Creating things like Thumbtack, which was like services to find a handyman and things like that. And social networks, Instagram.
TikTok, things like that. The thing is, on one hand, I love these businesses because if you're able to create it, you do truly have a moat. But liquidity, meaning making sure you have both the demand and supply at the same time, is an art.
I feel like marketplaces and social network is more art than science in that way. But I think there's tons of opportunities right now. Mark Liu, who some of you know, created a marketplace around selling startups and buying startups, TrustMRR .com.
So you're starting to see people come back to marketplaces and social networks. But I feel like it's a great thing to own, co -own, be a part of if you can. Personal agents are going to want to access marketplaces.
They're going to want to access social networks. The data that comes from both these things are really interesting. I love these businesses, so had to put it in there as well.
just in general, just like real assets. So the example I have here is small industrial workshops rented to local trades. And trades in general, by the way, like electricians, cabinet makers, plumbers, physical assets, warehouses, industrial warehouses, things like that.
We can't forget that you need a place to have a location. a property for a lot of the new world that we're entering in. So literally physical locations and real assets, real estate, I think still going to be a really interesting business.
Vertical agents, number 12. So just giving an agent responsibility for a recurring job in a specific industry. The example I have here is purchasing for indie restaurants.
So you have checking stock, comparing supply offers, ask for approval, place the order. I feel like people have been building vertical agents for the last 12 months, but I still think there's a ton of opportunity to build, like we talked about, harnesses and to build agents. And last but not least is security.
People need security more than ever in the AI age. Now, I'm talking about two types of security. I'm talking about...
Software companies like 1Password. I'm talking about companies like Cloudflare who have a lot of security AI features. I'm talking about the example I have here, agent access control for accounting firms that allow you to just read invoices and draft entries that would allow you to do that.
But basically thinking about sensitive workflows and then working. through firms' existing IT providers and just having access controls so that data doesn't get leaked and things don't get hacked and things like that. I think that's one piece of it.
Just security in the agentic era. Things are going to get hacked. This is going to be more and more of a problem.
I wouldn't be surprised if one of the personal agent startups or just agent startups. There's a big hack, security breach, and things get put on the internet.
I think that might happen. I think it's going to happen. I think it's actually going to happen more and more.
So people are going to want security and owning those assets and businesses are going to be great. And then actually security in general. I think there's Uber.
I forget the name of it, but it's like Uber for like, you know. bodyguards and you get a car and it's bulletproof and stuff like that. I think there's going to be more of those businesses.
People feel less and less secure. So they're just going to want to pay for security in general. A lot of security businesses I'm really interested.
And that's what I'd want to own in this world. I mean, I'm not saying, like I said in the beginning, I'm not saying to escape the permanent undercross, you need to own a piece of this. But I am saying, These are the only businesses that are left.
Owning a piece of these businesses are going to set you up. And this is where it's at. This is a map of the ones that I would pay attention to, the 13 ones, the best ones.
And I hope that this has gotten your creative juices flowing. Please do let me know. I read every single comment.
I reply to most on YouTube. And thank you for giving me your time. I don't take that lightly.
I think it's amazing. I can just come on here and be like, hey, I think that these are the 13 businesses, the only businesses left to build. And some of you are going to take that in and it's going to impact your lives.
And that's really cool. So thank you again. You're the best.
I'm rooting for you. I'll see you next time.
The Hook
The bait, then the rug-pull.
A solo podcast host claims there are only thirteen businesses worth building once agents are doing a meaningful share of the work, then spends twenty-five minutes defending that number with a named example and a hand-drawn diagram for each one.
Frameworks
Named ideas worth stealing.
01:15model
AI-Native Service Firm
Client records
Agents prepare the books
Human reviews exceptions
Completed monthly books
A service business where agents handle most of the delivery and a human only reviews exceptions before the job is billed as complete.
Steal forany existing service business deciding how much of its delivery pipeline to hand to agents while keeping one human checkpoint before the client-facing output
02:57model
Offline Business MVP
First visit
Weekly ritual
Test a communal or experience-based offline concept on a single visit before asking customers to commit to a recurring membership.
Steal forany physical-location business validating demand cheaply before signing a lease
05:41model
ACP Funnel
Useful niche information
An audience that comes back
Products they actually ask for
Build an audience with useful information first, let it become a community, then build only the product that community is already asking for.
Steal forany founder deciding what to build next by listening to an existing audience instead of guessing
07:34model
Proprietary Dataset Loop
Symptoms + machine + attempted repair
Verified outcome: did the fix work?
Licensed dataset for service networks
Collect real-world outcome data that documents whether an attempted fix actually worked, then license the verified dataset.
Steal forany technician-adjacent or outcome-verifiable industry sitting on data that's valuable precisely because it can't be scraped
09:08model
Domain-Specific Harness
Model
Industry tools + workflow rules
Memory + checks + human approval
A general AI model wrapped in the industry tools, workflow rules, memory, and approval checks that make it understand how a specific industry runs.
Steal forany vertical SaaS or service business layering judgment and process on top of a commodity model instead of competing on the model itself
12:16model
Robotics Deployment Model
Deploy equipment
Operate equipment
Building owner pays per job
Own the deployment and operation of someone else's physical robot rather than trying to invent the hardware.
Steal forany physical-AI idea that feels blocked on hardware R&D but could start as a service business operating existing equipment
14:01model
Physical Product With a Fan Following
One object
One obsessive niche audience
Repeat recommendation
Design one well-made physical object for a niche audience that already cares enough to recommend it to others.
Steal forany niche hardware idea choosing a single hero product and a single devoted audience instead of a broad product line
16:34model
Scarce Physical Input
Power
Cooled capacity
Workloads running
Own or service one scarce physical input, like cooling capacity, that every AI workload ultimately depends on.
Steal forany infrastructure-adjacent idea looking for a smaller entry point than building a full data center
17:31model
Guided Care Model
Qualified coach
Regular practice
Personal support
Pair a qualified human with a regular, structured practice and ongoing personal support, tested first with a small paid cohort.
Steal forany service business proving a care or coaching concept in rented space before committing to a dedicated location
19:30model
Two-Sided Marketplace Match
Factories with available capacity
Trusted matching + coordination
Buyers with a deadline
Match underused supply with time-pressured demand, starting manually in one region before automating the match.
Steal forany marketplace idea that needs to prove liquidity in a narrow slice before expanding
21:04model
Real Asset Rental
Electrician
Cabinetmaker
Plumber
Rent a physical workshop space near customers to multiple local trades, validating tenant demand before acquiring the property.
Steal forany real-estate-adjacent idea choosing a location-locked asset precisely because an agent can't manufacture more of that location
21:46model
Vertical Agent
Check stock
Compare supplier offers
Ask for approval
Place the order
Give an agent ongoing responsibility for one recurring job in one industry, starting with draft decisions before it's trusted to act alone.
Steal forany narrow, repetitive back-office job in a specific industry that could be handed to a supervised agent one step at a time
22:18model
Scoped Security Access
Read invoices: allowed
Draft entries: allowed + logged
Move money: human approval
Grant an agent narrow, logged permissions for a sensitive workflow while keeping the highest-risk action gated behind human approval.
Steal forany business giving agents access to financial or sensitive systems that needs a tiered permission model instead of all-or-nothing access
CTA Breakdown
How they asked for the click.
VERBAL ASK
11:50subscribe
“comment on YouTube for if you ever want me to go deeper into any one of these... also, while you're at it, like this, and if you're on Spotify or Apple, please rate it five stars”
a soft, mid-video engagement ask with no product or link pitch; the video's only other call is the closing thank-you, there is no sales CTA in the body of the video itself
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Greg Isenberg watches sixty minutes of Sam Altman's Dev Day keynote and picks the four launches, out of more than twenty, that he thinks hand outside builders a real business.
A two-person breakdown of OpenAI's top-tier model that skips the game demos and goes straight to code audits, nine money-making agent prompts, and a Raspberry Pi speaker built and shipped in about 30 minutes.
A solo walkthrough of five open-source GitHub repos getting traction right now, an AI writing editor, an agent-run CRM, a video-editing agent, a skill security scanner, and a phone-controlling harness, each with the exact install command and the first small workflow to try.
Greg Isenberg and developer Vinny break down WebMCP, the experimental browser feature that lets any AI agent search, compare, and buy on a website without scraping the page, then price two startup ideas built on it.
00:39
01:09
01:25
02:06
02:26
02:46
03:07
03:27
03:48
04:29
04:49
05:10
05:30
05:50
06:09
06:46
07:05
07:24
07:43
08:21
08:39
08:58
09:19
09:43
10:06
10:53
11:17
11:40
12:04
12:24
12:42
13:17
13:34
13:52
14:10
14:29
14:48
15:27
15:46
16:05
16:24
17:02
17:21
17:40
18:20
18:40
19:00
19:20
19:39
20:17
20:35
20:54
21:25
21:39
22:02
22:12
22:27
22:45
23:22
23:40
23:58
24:16
24:33
25:08
18:56
28:24
22:52
38:46
24:43
28:58