24:56Greg Isenberg · Tutorial
Claude Fable 5 is BANNED. What to do?
A 25-minute field guide to local AI models, written the weekend a government letter erased the world's most powerful model overnight.
June 13thThe argument in one line.
When AI makes software, content and answers nearly free, durable businesses are built on things AI can't copy overnight: physical work, permission, audience, trust, community, liquidity, proprietary data and results.
Read if. Skip if.
READ IF YOU ARE…
- An employee or aspiring founder who wants to start a business in 2027 and worries that AI will clone whatever they build.
- A founder or agency owner with a software product who wants to know which moats still hold when anyone can copy a product pixel for pixel.
- A creator or operator with an audience who is deciding whether to turn it into a community, a marketplace or a data business.
- Someone brainstorming startup ideas who wants a category map to mix and match rather than a single prescribed idea.
SKIP IF…
- You want step-by-step build instructions or tool tutorials. This is a high-level idea map with brief examples.
- You already run a business in one of the eight categories and want deep operating detail on it.
The full version, fast.
Anyone can now spin up a software company in seconds, so the question is what still has value. Isenberg offers eight categories: atoms (physical businesses), permission (licenses, regulation, partnerships), attention (media machines), timing (early trends), belonging (paid communities and events), liquidity (marketplaces), unique data (data brokers and curated directories) and outcomes (customers paying for results). Your first product may be copyable, but the data, brand and trust it collects are not. He ranks atoms, permission and attention near the top and outcomes last, and urges mixing categories.
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.
Create a free account →Where the time goes.
00:00 – 01:17
01 · Intro
Sets up the anxiety around AI-proof businesses and promises a map for building in 2027 and beyond.
01:17 – 02:26
02 · Atoms: physical businesses
Coffee shops and plumbers can't be cloned by AI, and tech money may push their valuations up.
02:26 – 04:10
03 · Permission: licenses, regulation, partnerships
Licensed professions, regulated industries and preferred-vendor deals are hard to copy on day one; partner with a lawyer if you need a license.
04:10 – 06:28
04 · Attention: building a media machine
Pick a micro niche and use AI to run a content machine, building a trusted brand rather than one page.
06:28 – 08:24
05 · Timing: spotting trends early
Monitor trends in micro niches with agents, build first, and collect data and trust even if the product is copyable.
08:24 – 10:59
06 · Belonging: paid communities and IRL events
Paid communities and in-person events built around a shared goal, with Hampton as the example.
10:59 – 14:38
07 · Liquidity: marketplaces
Marketplaces are sticky, and agents let a tiny team run one at millions in GMV.
14:38 – 18:31
08 · Unique data: data brokers and curated directories
Sell data nobody else has to AI labs or to people, including curated directories like better restaurant reviews.
18:31 – 20:44
09 · Outcomes: customers paying for results
Automated AI services such as a creative agency or a Fiverr and Upwork bidding machine, ranked the weakest category.
20:44 – 22:08
10 · Closing thoughts
Ranking of the eight types and the advice to mix categories instead of choosing one.
Lines worth screenshotting.
- Physical businesses like a coffee shop or a plumber can't be cloned overnight by AI, and their valuations may rise as tech money looks for moats.
- Licenses, regulation and preferred-vendor deals are permission that a copycat has a hard time getting on day one.
- If you can't hold a license yourself, partner with someone who can, and run the marketing, growth and AI side while they run the firm.
- Attention is underrated: people follow people, and AI lets one person run a content machine around a micro niche.
- Aim to build a media company with an irreplaceable, trustworthy brand in your niche, not a single page.
- Your first product at an early trend may be the least AI-proof part. The data, brand loyalty and trust you collect around it are what last.
- People are more connected and more disconnected than ever, which makes paid communities and IRL events a strong AI-proof play.
- The hard parts of a paid community are choosing the niche, defining the goal, delivering a great experience and setting a no-brainer price.
- Marketplaces are among the stickiest businesses ever built, and personal agents now let a team of one to five run one doing millions in GMV.
- Look for an underserved niche that is screaming for a marketplace, or partner with a creator who already has the attention.
- The big AI labs are hungry for new types of data, so data nobody else can get can be sold to them, to people, or both.
- As LLMs and social feeds give everyone infinite content, curated high-quality directories become worth paying for.
- Automated AI services where customers pay for results are real but early, and Isenberg ranks outcomes the weakest of the eight on their own.
- Defense in an agent business comes from feeding agents the best data and insights, plus permission like an exclusive deal.
- You don't have to pick one category. Mixing attention with belonging, or permission with liquidity, is part of the game.
Build on what AI can't copy overnight
WHAT TO LEARN
When software and content cost almost nothing, value lives in physical work, permission, trusted audiences, communities, marketplaces, unique data and results, and the strongest plays mix several.
02Atoms: physical businesses
- Physical businesses resist overnight cloning, and their valuations may rise as tech money looks for moats.
03Permission: licenses, regulation, partnerships
- Licensed and regulated businesses are hard to enter, and you can partner with a license holder while you run marketing, growth and AI.
- A preferred-vendor deal with a medium or large company is permission a competitor can't copy on day one.
04Attention: building a media machine
- AI lowers the cost of making content, so pick a micro niche and build a trusted media brand instead of posting randomly.
05Timing: spotting trends early
- Build first on early trends, because the data, brand loyalty and trust you gather outlast a product that is easy to copy.
06Belonging: paid communities and IRL events
- Paid communities and IRL events work when you choose the niche, set a clear shared goal and price it as a no-brainer.
07Liquidity: marketplaces
- Marketplaces are sticky, and agents handle support and operations so one to five people can run one doing millions in GMV.
- Look for an underserved niche or a creator audience that needs a marketplace, rather than a new mechanic for its own sake.
08Unique data: data brokers and curated directories
- Data the big labs can't get is sellable to them, to people, or both, and curated directories gain value as feeds fill with noise.
09Outcomes: customers paying for results
- Automated services paid on results are real but early, and they only hold up when fed unique data and backed by permission.
10Closing thoughts
- Mix categories, such as attention with belonging or permission with liquidity, instead of limiting yourself to one.
Terms worth knowing.
- Atoms
- Physical, real-world businesses such as coffee shops or plumbing, contrasted with pixel (software) businesses.
- Permission
- Advantages that need approval or relationships to get: licenses, regulation, and exclusive or preferred-vendor deals with larger companies.
- Preferred vendor
- A supplier a large company has formally approved and promotes to its own customers, such as a Salesforce or Microsoft partner agency.
- Liquidity
- In a marketplace, having enough buyers and sellers already present that supply and demand actually meet.
- GMV
- Gross merchandise value: the total dollar value of goods sold through a marketplace.
- Data broker
- A business that collects data others can't easily get and sells it, for example to AI labs that train models.
- Micro niche
- A very narrow audience or topic area, narrower than a normal niche, where a focused brand can become the trusted source.
Things they pointed at.
09:30productHampton (Sam Parr)
11:40productFinalLap (collectible car auctions)
12:50productCars & Bids
15:00productMercor
15:00productmicro1
20:10productFiverr and Upwork
Lines you could clip.
00:31
“Pretty much everyone has a little bit of anxiety around how to create an AI-proof business right now.”
01:47
“Your toilet gets clogged, AI is not going to be able to fix that leak.”
07:41
“Your initial product might not be an AI-proof idea. What is AI-proof is the data and the brand and the people that you end up collecting.”
12:07
“Marketplaces are one of the most sticky businesses of all time.”
18:10
“How do you create data that someone else can't get?”
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.
We need to talk build an asset, who want to create income for themselves, who want to build a business, scale a business, have that whole experience. What do AI -proof businesses actually look at?
So by the end of this episode, I'm going to cover all of that. And I just want to get your brains thinking about where the world is going. How do you think about AI -proof businesses?
What does that even mean? And just get your gears turning on it because let's be honest. Pretty much everyone has a little bit of anxiety around how to create an AI -proof business right now.
Even when you're speaking to incumbents, companies that have billions of dollars of revenue, they're kind of like, oh my god, you know, is, you know, big Frontier Lab going to come after me? So today's episode is just going to be thinking about how you should be thinking about this. So this is for anyone who's interested in building a business, 2027 and beyond.
Let's go.
We'll start actually with, you know, Adams businesses. So what do I mean by Adams? You have pixel businesses, the software businesses, then you have Adams.
These are physical businesses. And when you think about a physical business. Coffee shop, you got to come in, people want coffee.
Or a plumber, your toilet gets clogged, AI is not going to be able to fix that leak. These are the sorts of businesses that aren't going anywhere. And you can actually make the argument that these businesses' valuations are going to go up from here.
Because you're going to have a lot of that AI tech money trying to get in here once they see. the moat and the opportunity. I think you're starting to see a little bit of that.
I don't know about you, I just see tons of content around boring businesses. People are trying to get into it.
It's harder to do physical businesses, but those physical businesses, AI can't clone overnight. The second is, Businesses that require some permission.
What do I mean by that? I mean licenses. I mean regulation.
And I also mean if you have the ability. If you're likable or you have connections or you can use Grokbot or whatever to actually get you in front of the right people to go and create a deal with a larger company and they help sell your product, that is a partnership that can't be copied on day one. So what do I mean by that?
Maybe you have an agency and you partner with Salesforce and you're one of their preferred vendors or a preferred vendor of Microsoft or it doesn't even need to be that big, just a preferred vendor of a medium -sized company, ideally a large -sized company, but a medium to large -sized company. That is something that people can clone.
Licenses, what do I mean by that? I mean that there's some businesses that you need a license to operate. There's some businesses that are regulated and not everyone could go and do that.
What's an example? If you want to own a bookkeeping firm, you have to, in some countries, have been educated a certain way. If you want to own a law firm, in some countries, you actually have to be a lawyer.
So maybe you partner with a lawyer. And you handle the marketing and growth side and the AI side. And they handle running the law firm.
These are the sorts of things to help. I'm just trying to get your creative juices this entire episode flowing so that you can start thinking about, okay, if I am going to start a new business, how do I make it as AI -proof as possible? Attention.
Somehow attention is underrated right now. People follow people. And in a world where AI exists, if you have the right systems there, they can make your organic content.
You can go and create more organic content. I was one of those people that hated short form video. I mean, I like to consume it, but I hated it from a...
creation standpoint. Never liked doing it. I hated the editing around it.
Now I have Opus 5 .5 as my editor. And I can go and create short form on the fly when it works for me in my schedule. And I've actually been quite enjoying it.
And just over the last week I've gotten... Literally hundreds of thousands of views on that content, which is really cool. And people saying it's valuable.
So there is literally no excuse in 2026, 2027 to create a machine that is creating media. And I believe that media is somehow underrated right now. I'm talking about newsletters, I'm talking about meme pages, I'm talking about...
founder -led content. We all know that distribution is king.
I'm not talking about just creating one page. I'm talking about creating a media company. I think that those businesses are going to do tremendous over the next few years.
To a certain extent, they're somewhat AI -proof.
There's some versions of media that could be replaceable, but the idea is if your goal is to create a good media company, you are trying to create an irreplaceable, trustworthy brand that people come to for your niche. I think the way to think about attention is, what is your micro -niche? What is your niche?
How do you create a machine around it? How do you use AI to amplify? And you don't necessarily need to be the face.
You don't necessarily need to do founder -led content. But it does help when you have a human being in front doing it. The next is timing.
So timing, what did I mean by this? I just basically mean that you're seeing more and more trends. four different micro niches and niches now than ever.
And I think that there's an opportunity to use something like ideabrowser .com, use something like a Grokbot, use something like a Cloud Duo. any personal agent platform that you want to use to go and monitor for trends in your niches, in your micro niches, and go and build something as the first mover advantage. Sometimes I am shocked when you're early on a trend how few people have actually gone and built something for these people.
Okay, what am I talking about with building? It could be a software product. It could be a physical product.
It could be a media company. It could be an Atom's here, a physical thing. We're going to talk about other things too, like marketplaces and data and stuff like that.
It could be one of those. But the point I'm trying to make around trends is being a first mover advantage allows you to collect data, to collect... insights around a particular micro -niche or niche.
From that, your initial product might not be AI -proof. It might not be an AI -proof idea. In fact, it might be the least AI -proof.
But what it is AI -proof is the data and the brand and the people that you end up collecting into this, that trust that you build, that brand loyalty that you build. I think there's a ton of opportunity to go and look at trends. I just believe that everyone should be doing this.
Everyone should be creating machine systems around finding trends, acting on trends, and ones that they believe that they can add value to. To me, this is a no -brainer. Belonging.
Now more than ever, people are as connected as ever, but they're feeling as disconnected than ever. People want to be with people, and there's tremendous opportunities to build AI -proof, around that sense of belonging.
That could be something like paid communities. That's something like a circle or a school or just a weekly virtual event where you're keeping people accountable. When you think about belonging, you're thinking about a goal that this community has and shared interests that they have.
They want to hit that goal. around that shared interest, fast, with someone else, with that accountability, having fun, entertainment. There's so many niches and micro -niches where there's an opportunity to charge people to help them go and do that.
That could be IRL. You can just do a full IRL experience. My friend Sam Parr has his company Hampton, which is not affiliated at all.
what they do is, from what I understand, meetups around North America for entrepreneurs. Sort of like AA for entrepreneurs that do at least, let's say, $3 million a year in revenue. I think that's really cool and that's a really big opportunity.
What is the Hampton for XYZ? What is that paid community that people would be willing to spend hundreds of dollars, thousands of dollars? And then you, as the host, are helping them.
Well, you're helping curate the group, and you're also helping them move along their journey. So virtual and also IRL events, there's just so much opportunity in IRL events. We just threw an IRL event, was it last week, in Nashville, bringing around founders who are interested in deploying agents to build their businesses.
We had so much demand, we didn't really know even how to meet it. I think there's tons of demand for these sorts of things. The hard part for you is going to be thinking about what is the niche, What is the goal?
How can you actually produce a really good experience? And what is the price point that makes it a no -brainer? The next category is around liquidity.
I often hear people say, Greg, software is done. It's over. Don't even consider thinking about building anything in software.
One example I always give to people is marketplaces.
There's so much still opportunity to build marketplaces. And what is a marketplace, actually, before I get into it? A marketplace is just a fancy word for saying a place where there's supply and demand and you're connecting that supply and demand on this website, on this app.
A friend of mine launched a marketplace last week, I think. It's called FinalLap .com. Basically what it is is high -end collectible cars.
Every week you come to the site and you can bid on this high -end collectible cars thing. They curate the cars and there's a lot of hype around it. I think he has a big audience on Instagram so he uses his Instagram audience to market it.
It's just an example of a little bit of a new mechanic car sales, collectibles. So there's that niche there.
And all of a sudden, there's a new marketplace for this sort of thing. And I think what's really cool about marketplaces is they are one of the most sticky businesses of all time. If you're able to get people, get the attention and bring that attention into a marketplace, you've got a really good...
good business there. In the car space, we all know cars and bids sold for a ton of money, many millions of dollars. There's so many examples of marketplaces that have gotten acquired for ridiculous amounts and ridiculous sums of money.
Why? Because again, we all know that it's so hard to go and create. Now I think there's an opportunity to go and actually find creators, find media companies, and build their marketplaces for them, partner with them.
Or you end up just incubating them. and tension yourself, and then you go and build the liquidity. I think there's a lot of opportunity here, a lot of unmet demand.
The hard part here is about thinking about what is a new mechanic that doesn't exist currently? What is a new niche or micro -niche that just doesn't need to be new, actually? What is an underserved niche or micro -niche that is just screaming that they need a marketplace?
So I like this one a lot. Also what's cool about the marketplace and liquidity space here is you can go and use a Grokbot, a Cloud Code, one of these personal agent platforms to go and actually run a lot of the tough... parts of running a marketplace.
In the past, running a marketplace was pretty intense from a management perspective. Support tickets, you have a lot of people and everything needs to be perfect. But with AI now and personal agents, we've gotten to the point where you actually don't really need a big team.
You can create a team of one, two, three, four, five people it could support a marketplace that does many millions in GMV. I like this one a lot.
Unique data. This has gone viral over the last few months. You see data brokers making millions of dollars a week.
I'll just give a high level, what is that? What is that business? The business is basically, on one hand, and maybe I'll do a whole breakdown on a separate episode.
if people are interested. But basically, on one side, you have unique data around something. Initially it was coding.
A lot of these platforms like Mercore and MicroOne and things like that, they were getting data, coding data, and they were selling it to... the big models now there's the the big models just are hungry for these new types of data that they don't have access to so the game is what is what is that data And then how can you sell it to them?
That's basically the game. There's still a tremendous amount of opportunity there. And that's what I'd call a data -rich business.
Another data -rich business is actually if you have access to some amount of data that isn't easily accessible, you can go and have a curated directory of some sort that people will pay for. And I think you're going to see a huge explosion of curated websites and apps over the next few years. Why do I think that?
In a world where LLMs basically give you infinite content, social feeds basically give you infinite content, you're going to want to go and access curated databases of content that you believe is high quality. I'll give you the example that comes to mind just off the cuff.
Google reviews or Yelp reviews, just reviews of restaurants in general, have gotten to a place where, I don't know if you've noticed this, it's either five -star reviews or one -star reviews.
It's a three on five, right? Because you've got some fives and ones or a four on five. And it's just very polarizing with reviews.
And also I think a lot of people have gamed the review sections on places like Yelp and stuff. So I think that for me it would be really cool. to pop into a website or an app and really give me new restaurant recommendations that were actually high, high, high quality reviews that weren't just like love it or hate it.
Yes, that's an example of a consumer business. Consumer business have more risks, less likely to work. But I'm using it as an example to share that there's probably some reviews.
that are high quality. And there's probably a way to curate that into a database and someone might be willing to pay something to access that. I think there's a ton of opportunity in enterprise.
I think there's a ton of opportunity in prosumer. I think there's a ton of opportunity in the SMB space. I think that there's some consumer opportunities as well.
But overall, the game here is... How do you create, because we're talking about AI -proof startups, what does that look like? How do you create data that someone else can't get?
And then you either sell it as a website or app to people monthly or whatever, or you sell it to the labs, or you do both. Lastly, I think you're starting to see this, but it's so early. Customers paying for results.
I believe that there's going to be more and more automated AI services. What do I mean by that? Basically where you have a bunch of agents going and doing the work.
The example I've given in the past on the pod is how do you create a creative agency that is fully automated by different creative models. It creates logos on the fly or landing pages on the fly. The unfair advantage that you can have, is around context, is around data, is around the system that you've created.
If you can create the best system for a particular output, then there's some amount of moat or some amount of defense that you can have against that. If I was ranking these, I wouldn't put the outcomes one as like... the best one.
It's probably the least best. I think physical stuff, license and regulation stuff, huge.
I don't want to rank them all, but I would say at the bottom would probably be outcomes.
I'll end on this one here just to finish my thought here. Automated AI services, I just feel like you don't see this very often and there's an opportunity. The example I gave on the podcast before is how do you look at platforms like fiverr and upwork see what people want to pay for and then create a essentially machine that goes and bids on some of this work there's a ton of opportunity there and i think agents are going to fulfill a lot of that work the way to make it ai proof is if you have the best possible And the way you make the best possible agents is if you feed them with the best possible data, the best possible insights, the best possible inspiration, and you have something like we talked about around permission, like you have a deal with a particular company that no one else can have and stuff like that.
This is basically the map of how to think about a world where when AI makes software and content and answers, pretty much free. How do you build something with value?
Is it everything in here? No. This is just some of the main ones, the main categories to get your creative juices flowing.
I'll end with, I think there's a ton of opportunity. You do have to be creative. you do have to think about how many, do you want to mix some of these?
Do you want some attention stuff with some belonging stuff? Do you want some permission stuff with some liquidity stuff? Do you want some outcome stuff with some unique data?
So there is a bit of a mixing. It's not like you just go into one category and that's it. But I do believe that I wouldn't look at the world and be like, it's done and we can't create startup ideas and we can't create assets.
In fact, sometimes when there's the most amount of chaos, create the most amount of opportunity. Thank you for listening.
Hope it's been helpful. If you want more of this stuff in your feed, like, comment, and subscribe. I'm here to get your creative juices flowing, here to help you navigate this crazy world we live in with ideas that could help you build a business.
Thank you for watching the Startup Ideas podcast and I'll see you next time.
The Hook
The bait, then the rug-pull.
Everyone building a business right now carries a little anxiety about AI cloning it. Greg Isenberg answers with a map of eight categories where value survives when software, content and answers cost almost nothing.
Named ideas worth stealing.
02:26list
The 8 AI-proof business categories
- Atoms
- Permission
- Attention
- Timing
- Belonging
- Liquidity
- Unique data
- Outcomes
A map of where value survives when AI makes software, content and answers nearly free. Isenberg ranks atoms and permission near the top and outcomes at the bottom, and says to mix categories.
Steal forbrainstorming a new business or stress-testing an existing one
07:59model
What stays AI-proof after you build first
- Data
- Brand
- Trust
Your first product at an early trend may be the least AI-proof part, but the data, brand loyalty and trust you collect in that micro niche are what last.
Steal forany early-trend product launch
How they asked for the click.
VERBAL ASK
21:41subscribe
“If you want more of this stuff in your feed, like, comment, and subscribe.”
Short and low-key, tacked on after the closing thought.
MENTIONED ON CAMERA
Visual structure at a glance.
