The $5 Trillion AI Roll-Up Opportunity for Solo Founders
Greg Isenberg breaks down how Thrive Capital and General Catalyst are buying small service firms and running AI agents inside them, then lays out the exact folder structure, agent pipeline, and weekly routine for running a one-person version.
Posted
today
Duration
Format
Essay
educational
Views
2.6K
262 likes
57 · 43
Big Idea
The argument in one line.
A $5 trillion wave of retiring-owner service businesses is too small for billion-dollar funds but perfectly sized for a solo founder who buys one firm, builds a shared AI-agent layer with a human-approval rule, and repeats.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You already run or want to run a small holding company and want a concrete AI-agent operating structure to apply to it.
You're a solo founder or small operator looking for an acquisition strategy that doesn't require venture-scale capital.
You want an actual folder and file structure for running AI agents inside a business with a human-approval safety rail.
SKIP IF…
You're looking for a passive-investment pitch; this is an operating playbook for buying and running service businesses, not a stock tip.
You have no interest in acquiring or operating small businesses like accounting, bookkeeping, or property management firms.
TL;DR
The full version, fast.
About a million small service businesses, roughly $5 trillion in value, are expected to change hands by 2035 as their owners retire, and AI agents can now handle the back-office busywork inside them well enough that a person just checks the draft. Thrive Capital and General Catalyst have already proven this at firms like Larson Gross and Long Lake, buying trusted firms, running agents in the background, then moving the back office over, pushing profit from 5-10% toward 30-40%. Because a $2 million bookkeeping firm is too small for a billion-dollar fund, a solo founder can buy one firm, build a shared layer of agents and rules, install a GM with real upside, and repeat, using the same folder structure, intake-preparer-reviewer pipeline, and weekly corrections-log routine the funds use.
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.
Cold open on the $5 trillion figure and the claim that billion-dollar funds are already buying up small businesses and running AI agents inside them. States the opportunity is open to solo founders and previews the three things the video will cover.
01:52 – 04:52
02 · Why the $5 Trillion Shift Is Happening Now
Three forces line up at once: owners who built firms in the 80s and 90s are retiring with no succession plan, AI agents can now reliably do the back-office work, and service firms remain priced at 5-10% profit as if margins can never improve.
04:52 – 08:38
03 · Examples: Thrive Holdings & General Catalyst
Thrive Holdings has bought close to 50 accounting firms in 24 months, including Larson Gross where Codex-based AI cut one accountant's workload from 180 to 15 hours a year. General Catalyst has put $750M+ into roll-ups like Long Lake (property management) and Crescendo (support centers, 90% of tickets AI-handled).
08:38 – 10:07
04 · The Fund Playbook
The five-step playbook: build the agent platform first, buy a trusted firm, run agents in the background, move the back office over, then buy the next firm and plug it into the same platform, pushing profit from 5-10% toward 30-40%.
10:07 – 10:51
05 · The Small-Deal Gap
A billion-dollar fund can't justify time on a $2M-a-year bookkeeping firm, leaving most of the roughly one million businesses expected to sell too small for the big players and open to solo buyers using the same AI models.
10:51 – 14:23
06 · The One-Person Holdco
Lays out the structure: the founder owns the holding company, each business has a GM with real equity upside, and every business shares one layer of agents, rules, back office, and dashboards, making each new acquisition easier than the last.
14:23 – 16:47
07 · The Folder Structure
The shared system is organized into a thesis folder, a shared folder (global rules, test cases, runbooks), and one folder per business (clients, people, business-specific rules, corrections log). If you only set up three files: global rules, business rules, and the corrections log.
16:47 – 18:38
08 · The Agent Pipeline
Work passes through intake, preparer, and reviewer agents before a human approves it. The hard rule: the reviewer can block a draft, but no agent can send anything to a client.
18:38 – 19:41
09 · Inside a Reviewer Agent File
Shows the actual plain-English job-description file for the reviewer agent: what it checks, what it can and can't do, and when it must stop and escalate to a human. Most agent problems start when nobody tells the agent where its job ends.
19:41 – 21:49
10 · My Week Running the Holdco
Monday is numbers (profit margin, human minutes per job, fix rate, retention, staff happiness); Tuesday is one call per GM; Wednesday is the corrections log, turning fixes into rules; Thursday and Friday go to finding the next business.
21:49 – 22:24
11 · How to Land the First Business
Recommends picking one industry, selling owners a single AI-powered service to earn trust, and being the obvious buyer after 6-12 months of good work. Marketplaces like BizBuySell rarely produce good deals unless a listing is unusually curated.
22:24 – 27:43
12 · Arguments Against AI Roll-Ups
Answers five objections in a lightning round: roll-ups always fail, margins get competed away, regulated work needs human checking, people hate change, and this is just a fancy way of saying layoffs.
27:43 – 29:49
13 · Closing Thoughts
Reiterates the size of the opportunity for one-person or small holdcos, introduces the term 'multi-preneurship,' and asks viewers who start one of these businesses to share their results.
Atomic Insights
Lines worth screenshotting.
McKinsey expects about a million small businesses, worth roughly $5 trillion combined, to sell by 2035 as their owners retire.
A normal service firm runs at 5-10% profit, a number that hasn't moved in 30 years because buyers price it as if it never will.
Thrive Holdings has bought close to 50 accounting firms in 24 months and just committed another billion dollars to buy more.
At Larson Gross, AI built on OpenAI's Codex cut one accountant's 180-hour-a-year job down to 15 hours.
General Catalyst has deployed more than $750 million into at least 10 roll-up companies as part of a $1.5 billion strategy.
General Catalyst's Long Lake has bought 18 property management companies and reached $100 million in EBITDA in under two years.
General Catalyst deals typically pay sellers 60-70% in cash and roll 30% into equity, giving the seller a reason to stay through the handoff.
A billion-dollar fund can't justify spending time on a $2 million bookkeeping firm, which leaves most of the million businesses for sale open to solo buyers.
The fund playbook is build the agent platform first, buy a trusted firm, run agents in the background, then move the back office over before buying the next firm.
If the thesis holds, this playbook can take a business from 5-10% profit to 30-40% profit on the same revenue.
Every business in a one-person holdco shares the same agents, rules, back office, and dashboards, so each new acquisition is easier to bolt on than the last.
The agent pipeline runs intake, preparer, and reviewer agents, but only a human approves work before it reaches a client.
The reviewer agent can block a draft, but no agent, including the reviewer, is allowed to send anything to a client.
The corrections log, where every human fix to an agent's work gets recorded and turned into a rule, becomes the most valuable asset in the business after a few hundred jobs.
The best way to land a first acquisition is to sell one AI-powered service to owners in an industry for six to twelve months, earning trust before they're ready to sell.
Isenberg treats staff and client attrition, not AI failure, as the risk he takes most seriously in this model.
Takeaway
The folder structure and agent rules that let a solo founder run a mini roll-up.
WHAT TO LEARN
A wave of retiring-owner service businesses is too small for big funds but perfectly sized for a solo operator who copies the same agent pipeline, folder structure, and human-approval rule the billion-dollar players use.
02Why the $5 Trillion Shift Is Happening Now
Three forces are converging at once: owners built in the 80s and 90s are retiring, AI agents can now do the back-office work, and service firms are still priced at 5-10% profit as if that will never change.
McKinsey estimates about a million of these small businesses, roughly $5 trillion in value, will change hands by 2035.
What makes this different from the old private-equity roll-up story is that AI agents are now genuinely good enough to trust with the work, not just capable of a rough draft.
03Examples: Thrive Holdings & General Catalyst
Thrive Holdings has bought close to 50 accounting firms in 24 months and just committed another billion dollars to buy more.
At Larson Gross, AI built on OpenAI's Codex processed 7,000 tax returns in one season and cut one accountant's 180-hour job down to 15 hours.
General Catalyst has put more than $750 million into at least 10 roll-up companies, including Long Lake (18 property-management acquisitions, $100 million EBITDA) and Crescendo (AI handling 90% of frontline support tickets).
General Catalyst's deal structure pays sellers 60-70% cash and rolls 30% into equity, which gives the seller a reason to stick around for the handoff.
04The Fund Playbook
The five-step playbook is: build the agent platform first, buy a trusted firm, run the agents in the background, move the back office over, then buy the next firm and plug it into the same platform.
If the thesis holds, this turns a business's profit margin from 5-10% into 30-40% on the same revenue, because clients and invoices stay the same while agents absorb the labor cost.
05The Small-Deal Gap
A billion-dollar fund can't justify spending time on a $2-million-a-year bookkeeping firm, so most of the roughly one million businesses expected to sell are too small for the big players.
A solo founder uses the same AI models the funds use (Codex, Claude Code, Gemini) and gets to be the integration person in the room, instead of hiring a manager to learn the business from scratch.
06The One-Person Holdco
The structure is: you own the holding company, each business underneath has its own GM (often the person already running it, given real equity upside), and every business shares one layer of agents, rules, back office, and dashboards.
Because the shared layer is built once, each new acquisition gets easier: the second business is easier than the first, the third easier than the second.
Sellers who spent decades building a business often prefer handing it to a person over a large venture capital firm, which is an edge a solo buyer has that a fund doesn't.
07The Folder Structure
The system is organized as three folder types: a thesis folder (what industries and what a good business looks like), a shared folder (global agent rules, test cases, runbooks), and one folder per business (clients, people, business-specific rules, corrections log).
If you only set up three files, set up global rules, business rules, and the corrections log — those are what let you actually trust an agent with a client.
08The Agent Pipeline
Work moves through three agents in sequence: intake (collects and chases documents), preparer (writes the first draft), and reviewer (checks the draft against the rules and can send it back).
The hard rule: the reviewer agent can block a draft, but no agent, reviewer included, can send anything to a client. A person always approves first.
09Inside a Reviewer Agent File
Every agent gets a plain-English job description file covering what it can do, what it can never do, and when it has to stop and ask a human.
Most agent failures come from nobody telling the agent where its job ends, not from the model being incapable.
10My Week Running the Holdco
Monday is numbers day: check profit margin, human minutes per job, how often drafts need fixing, client retention, and whether key people are staying, for every business.
Tuesday is one call per GM to hear what's working and which clients need attention; Wednesday is the corrections log, turning every human fix into a new rule; Thursday and Friday go to finding the next business.
The corrections log hour is described as the most valuable hour of the week, because after a few hundred jobs that list of rules becomes the thing nobody else can copy.
11How to Land the First Business
The recommended path in is to pick one industry, sell owners a single AI-powered service (like cleaning up month-end for their messiest clients), and use that work to learn the business from the inside.
After 6-12 months of doing good work for a handful of firms, one of those owners becomes a natural acquisition when they're ready to step back.
Marketplaces like BizBuySell rarely produce good deals unless a listing is unusually curated; direct outreach works better.
12Arguments Against AI Roll-Ups
Objection 1, 'roll-ups always fail,' is true of private equity broadly, but the failures come from overpaying and integrating too fast, not from AI itself — buying slowly and doing the integration yourself avoids it.
Objection 2, 'margins get competed away once everyone uses AI,' may eventually be true, but 20-year client relationships and a business's own accumulated rules list take years for a competitor to copy.
Objection 3, 'regulated work needs human checking,' is true, but checking a draft takes far less time than producing one from scratch, which is exactly where Larson Gross's 31% time savings came from.
The objection taken most seriously is that people hate change and staff or clients might leave; the mitigation is 30 days of no visible change to clients, agents running in the background before touching anything real, and turning the person who knows every client into a GM with real upside.
On the 'this is just layoffs' objection, the honest answer is that some jobs will change: people shift from doing the work to checking it, and some staff won't be suited to the new role.
Glossary
Terms worth knowing.
Holdco
A holding company: a business whose only job is to own other operating businesses, sitting above a portfolio of small firms.
GM (General Manager)
The person who runs day-to-day operations at one business inside the holdco, usually someone already working there, given real equity upside.
Corrections log
A running record of every fix a human makes to an AI agent's draft, used weekly to write new rules and tests so the agent stops repeating the mistake.
Intake agent
The AI agent responsible for collecting documents from a client and chasing down anything missing before work starts.
Preparer agent
The AI agent that produces the first draft of the actual work, like categorizing transactions or drafting a month-end close.
Reviewer agent
The AI agent that checks a preparer's draft against the business's rules and either passes it to a human or sends it back for revision.
EBITDA
Earnings before interest, taxes, depreciation, and amortization; a common measure of operating profit used when valuing acquisitions.
Roll-up
An acquisition strategy that buys multiple similar businesses and combines them under shared infrastructure to cut costs and raise margins.
“Over the next 10 years, there's going to be millions of small businesses that are going to get sold because their owners are retiring.”
cold-open hook with a concrete stake→ TikTok hook↗ Tweet quote
03:28
“My human beings on my team are making more mistakes than my AI agents.”
blunt, quotable claim about trusting agents over people→ IG reel cold open↗ Tweet quote
06:19
“The people who used to do the work are becoming the people who check it.”
tight thesis statement on the labor shift→ newsletter pull-quote↗ Tweet quote
17:38
“The reviewer can block. No agent can send.”
single-sentence safety rule, easy to screenshot→ TikTok hook↗ Tweet quote
19:29
“Most agent problems start when nobody tells the agent where its job ends.”
generalizable insight beyond this niche→ newsletter pull-quote↗ Tweet quote
21:31
“After a few hundred jobs, that list of rules becomes the most valuable thing you own.”
strong closing line for the corrections-log section→ 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
Over the next 10 years, there's going to be millions of small businesses that are going to get sold because their owners are retiring. We're talking about $5 trillion worth of these businesses. Now, some of the biggest investors in the world like Thrive and General Catalyst are already buying them up and they're using AI agents to make them way more profitable.
Now, most people think that billion dollar funds can only do that and there's just no room for solo founders or small teams. I don't. I think there's a huge opportunity here for solo founders and almost nobody's talking about it.
I posted about it on X yesterday and it got over a million impressions. The funniest part was that my DMs and replies were filled with people messaging me to stop talking about it publicly because they think there's so much alpha in this. So.
Obviously, I'm making a whole episode of the podcast about this and you're listening to it. If you stick around until the end, you'll learn three things. You'll learn why this is happening right now and how Thrive and General Catalyst are actually doing it.
You'll see exactly how I'd run one as a solo founder, down to the folders, the agent files, and what my average week would look like. And you'll hear the biggest arguments against this. including the one I take most seriously.
I had hundreds of tweets of people who were like, yeah, here's what's wrong. And I speak to those arguments. If you're new here, I'm Greg Eisenberg.
I run a holding company of my own. I host the Startup Ideas podcast, and I spend most of my time thinking about which kind of businesses are worth building right now. If this episode gets 5 ,000 likes, comments, subscribes, I'll give away more of my notes on this whole topic.
It'll be in the pinned comment. Now let's get into the episode.
So why is this $5 trillion shift happening right now? And why is there an opportunity? I think three things happen at the same time.
And when that happens, you usually get a new kind of business. The first thing is owners are retiring. So that $5 trillion number actually comes from McKinsey.
And they expect about a million of those businesses to actually sell by 2035. These are people who started an accounting practice or an insurance agency or property management company. Maybe it's the 80s or in the 90s.
But they've built this loyal client base over 30 years. And now they want to retire. Maybe their family doesn't want to get into the business.
They don't have people lined up to take it over. Their kids went off and did something else. And their employees usually don't want to run the whole thing.
The second point is AI can do the work now. And I say that. Like, you know, I don't think I could have said this eight months ago, but I'm saying it now.
Think about what happens inside those businesses all day. Someone types numbers from a PDF into a spreadsheet and someone chases a client from the same missing document for the third time. Someone writes the same status update like 50 times a week.
Not long ago, if you pointed AI at that work, you got a rough draft and someone had to redo it and it was like hallucinating all the time. Now, if you set it up correctly, you get a draft that somebody checks. And I'm getting to the point, I don't know about you, but I'm getting to the point where I can trust my AI agents and my AI agent systems sometimes way more than my human beings.
My human beings on my team are making more mistakes than my AI agents. The third piece is service businesses are priced like their margins are stuck. So a normal service firm runs at 5 % to 10 % profit, something like that.
And everyone assumes that it'll just sort of stay that way because that's been that way for the last 30 years. And that's how buyers value those businesses. But when you put all three points together, you get a huge wave of businesses about to change owners full of the exact kind of work agents are good at and priced.
as if nothing about them can change. And I think that's one of the biggest opportunities about right now. So yes, people have talked about boomers and the $5 trillion number I've seen before, but they've talked about it in the sense of like, okay, I'm going to buy this company and then maybe I'll optimize it a little bit.
The black swan is the AI agents are actually really, really good right now. So I think that the funds, the big funds, have figured this out first. So let me tell you exactly what they're doing.
Let's learn from them. And some of these stories are actually really amazing. Okay, so what are the big funds doing?
Thrive. Thrive Capital is Josh Kushner's firm. So they were early in companies like Instagram, OpenAI.
And a while ago, they spun out something called Thrive Holdings. And when Thrive Holdings started buying, surprised a lot of people because they went out and they bought local accounting firms. Now, they've bought close to 50 of these accounting firms over the last 24 months, and they just committed another billion dollars to buy more.
So obviously it's working, right? One of them is a firm in Bellingham, Washington called Larson Gross. It's been around since 1949.
They've got five offices and they've got 200 people. After Thrive got involved, the firm started using AI built on OpenAI's codecs. I'm pretty sure Thrive has some sort of partnership with OpenAI.
This tax season, it processed 7 ,000 returns. And accountants saved 31 % of their time on average. One accountant had a job that used to take her 180 hours a year, and she got it down to 15.
So it's really, really working. And the interns who used to prepare tax returns started training to review them instead. And that's the part I can't stop thinking about.
The people who used to do the work, are becoming the people who check it. And keep that in mind because it comes back when we get to the arguments against all of this at the end.
The other firm is General Catalyst. So they're one of the biggest venture capital firms in the world. They set aside $1 .5 billion for this strategy.
And from what I heard, they put more than $750 million into at least 10 companies. One of them is Long Lake. which buys property management companies.
They say that they've bought 18 businesses and gotten to 100 million in EBITDA in under two years with margins doubling, which is hard to even comprehend, right? Another is Crescendo, which runs customer support centers. They say AI now handles about 90 % of frontline tickets.
And there's one I really like the structure of. So when general catalyst companies buy a business, they usually pay about 60 to 70 % in cash. And the founder rolls about 30 % equity into the new company.
So the person who built the relationship for 20, 30 years has a real reason to stay and make the handoff go well, which is like a big issue, right? One honest note, by the way, a lot of these numbers, when I was doing my research, are self -reported by young companies that are raising money and none of them have been through a recession yet.
So I take them as a strong signal of where this is going, not as proof, right? It's early days. But the thing that I take away from this is when you have Thrive and General Cadlis, both venture firms who I look up to and I think that they're really smart, they're putting billions of dollars into the same idea.
You got to pay attention. And if you look at how they're running it, there's a pretty clear playbook and a pretty big gap they're leaving for everyone else. And that's what I care about.
Of course, I hope they do well, but I really care about you listening person, founder, who wants to make their first few million dollars. And I want to give you that opportunity or at least get your creative juices flowing to connect the dots. So I'm trying to analyze their whole playbook, General Catalyst and Thrive.
And the simplest way I can describe it is first they build the platform, meaning the agents and the software, before they buy literally anything. Then they buy a firm people already trust. With clients who've been around for years, a lot of the time they like brand names, like good brands.
they run the agents in the background. So the agents do the work alongside the people and nobody outside sees anything change. If anything, they're more timely, it's more optimized, the customers are hearing back faster and they're getting hopefully better work.
Once the agents are reliable, they move the back office over. Things like data entry, collecting documents, and first drafts. And then they buy the next firm.
And then they plug it into the same platform. So look at what that does to the business. Revenue stays the same because the clients are the same and they're paying the same invoices.
Costs go down because agents are doing a big share of the work. So profit goes from anywhere around 5 % to 10 % to, if the thesis holds, to 30 % or 40%.
It's basically the same business making three to four times profit. Now, the thing I think most people miss is that the funds need big deals. If you're a billion -dollar fund, you can't spend your time on a bookkeeping firm doing $2 million a year.
It's just too small for them, and it can't move the needle. That's how they think. And the McKinsey number that I gave earlier, about a million businesses are expected to sell, Those are mostly small firms.
So Thrive and General Catalyst, I don't think that they're going to call them. I think they're going to call the bigger firms or they're going to call the mid -market firms. So the question I kept asking myself is, what does the one -person version of this look like?
And let's explore that. So how would I set up a one -person holdco doing these AI roll -ups? I would set it up as...
You, the founder, you're at the top. You own a holding company, which is just a company whose job is to own other businesses. That's the simplest way to think about a holding company.
Under it, there's a few small businesses. Each one has a GM, a general manager, and usually that's someone who's already there. Think of it as a senior bookkeeper who's been with the firm for 15 years, knows every client.
Then you make her the GM and give her a real piece of the upside. And this bar at the bottom is the part that makes the whole thing work. Every business uses the same agents, the same rule system, the same back office, and the same dashboards.
So you build the layer once. So the second business is easier than the first because half of what it needs already exists. And then the third one is easier.
Again, fourth, fifth, you get the idea. At the start, I said there's a huge piece of this pie for solo founders. And there's a few reasons why I believe that.
So the tools that you're going to use are the same tools that the funds are using. So Thrive agents run on Codex. But yeah, you can use, you know, Codex, Cloud Code, you know, Gemini.
You have access to the same models. You also get to be the integration. So the hardest part of any roll up is getting each new business onto the system without breaking anything.
So the funds have to hire managers who learn every business from scratch while you're the one in the room. You know the clients and you read the agent drafts yourself. So a lot of owners who would also rather hand their business to a person, if you spent 40 years building something, you care about who takes it over.
Some of them are just not going to want to sell to a large venture capital firm, but they might want to sell to you. And again, that is an edge. And the deals, like we talked about, they're too small for the big funds.
So you're not even competing with the general catalyst for them. I've run a holdco now for, wow, six years almost.
Actually, no, six years. I've run a hold code for six years. And I just got to say that who the GM is is going to be such an important part of making this thing work.
And I've seen not good GMs just... not be able to take the business where it should get to. And incredible GMs just really fly.
So finding that person is super, super important. And this is pretty, you know, it's pretty much how I think about my own holding company. You know, we have a few businesses with amazing people running each one.
And they've got like a shared layer underneath that keeps getting better. And we basically support those businesses. Now, let me show you what that shared layer could look like.
This is something that you might want to screenshot. Or like I said, if there's some likes and comments, I could put in the pinned comment. So this folder structure is kind of like the org chart.
I'll walk you through it so you can clearly understand how I think about this. The first folder is the thesis. This is where you're going to write down what you believe, which industries you're in and why, what a good business looks like to you.
That's going to keep you from buying something just because you got excited about it. The shared folder is that bar at the bottom of the drawing, meaning everything every business uses. So the agents and the global rules live here.
There's also a folder of real accepted work that you use to test the agents every time you change something. And there are runbooks, which are just step -by -step instructions for things that happen more than once. Like, okay, it's day one after taking over a business, or what you do when an agent sends something wrong.
then you're going to want to make sure each business has its own folder. So the client's file says who's been with the firm for 20 years, who's sensitive, who's loyal to a specific person. Those details are going to matter.
The people file says who knows what. If the senior bookkeeper is the only one who knows how a certain client likes things done, that gets written down here. The rules file holds things that only apply to this business.
So maybe this firm always sends reports on the third instead of the first because one big client asked for that in like 2011. And the corrections log is where every fix goes. Anytime a person changes something, an agent did it, it gets logged here.
If you set up only three files, I'd say set up the global rules. each business rules, and the corrections log. Because that's the stuff that turns agents into something you can actually trust with a client, right?
And honestly, even if you don't do this, like don't do an AI roll -up, thinking about your business in this way is pretty darn helpful from a structure perspective, from getting more out of your AI agents, and for just being more AI native. So those are the folders. Let's talk about the agents living inside of them because there's one rule here that I think prevents most of the disasters people worry about.
This is how work actually moves through one of these businesses. So when work comes in, the intake agent is going to collect the documents and chases anything that's missing. That's going to save a ton of time.
You have the preparer agent who does the first draft of the actual work. In a bookkeeping firm, what could that be? Categorizing transactions or drafting the months and close, things like that.
You have the reviewer agent which checks that draft against the rules. Something's off or something, it sends it back. Then, yes, there's human beings.
You still want human beings a part of this. You're going to need a person to approve it. Usually the same person who used to do this work by hand and only after that does it go to the client.
I think the most important rule of this whole system is the reviewer can block, but it can never send anything. And the preparer can't send anything either, by the way. A person always approves before anything reaches the client.
Yes, can we get to a point where it's completely AI, you know? agents doing everything, maybe one day. But I think at this point, you do need to have humans do certain things.
And also human beings are the accountability layer for you. I think I'm going to do a whole episode of the podcast about that. But yeah, the person is the accountability layer.
I thought it might be interesting for people to see what do these agent files actually look like. Here's one. It's the one for the reviewer.
So you can check it out here. Your job. Check every draft from the preparer before a person sees it.
You score it or pass it to the person or send it back. Here's what you can do. Here's what you can't do.
Check every draft for these things. Send it back if this. When you pass it on.
What is this? Basically, I won't bore you with the details. It's basically like a plain English job description.
Every agent in the system gets one. And it covers what its job is, what it's allowed to do, more importantly, what it can't do. And when it has to stop, ask a question and get human intervention.
Most of the agent problems I've seen... come from nobody telling the agent where its job ends and writing down what it can never do fixes a lot of that. So I think this will be helpful for you.
So imagine you've rolled up a few of these businesses. What does an average week look like for you? People ask me about this.
the common feedback I get is, it must be so hectic. It's actually a lot less hectic than you think because your GMs and your agents handle the day -to -day work. So here's how I would spend my week if I own a few of these businesses.
On Monday, you want to look at the numbers. For every business, check five core... core metrics.
Number one, profit margin. Two, minutes of human time per job. Number three, how often agent drafts need fixing.
Number four, client retention. And number five, whether the key people are happy and staying. A lot of people don't have that, but it's going to be key to the whole system, right?
If profit is going up while clients are leaving, well, something's wrong. You're going to want to catch that early. On Tuesday, I do one call with each GM to hear what's working, what's annoying them, which clients need attention, how you can help.
Wednesday, maybe it's your corrections log. I think it's the most important hour of the week because I'd go through every fix a person made to an agent's work and I'd turn the ones that keep happening into rules.
I'll just put up a prompt here, show you a prompt. Compare each agent draft with the version a person approved this week. Sort every change into factual error, client preference, missing information or style.
For any correction that happened more than once, propose a rule. Add each approved rule as a test using the original input and the accepted output. And that's how agents get better every week and it just compounds.
After a few hundred jobs, that list of rules becomes the most valuable thing you own because anyone can use the same AI models, but nobody else has your list of every way they go wrong in your kind of business. And then Thursday and Friday. That's for finding your next business.
Coffee with owners, learning a new industry, figuring out what comes next. People ask me about the marketplaces to buy some of these businesses. I think it's BizBuySell and some other ones.
My experience is once they hit the marketplaces, unless it's super curated, you're not going to get a good deal. The best way to find these deals is to reach out and things like that. The last thing, by the way, people ask me is how do you get the first one?
How does the first one start? The path I like to pick, one industry, say bookkeeping firms, and I start by selling them a service. Take one annoying job and do it with agents.
like cleaning up month end for their messiest clients or chasing missing documents. And then doing that, you learn how these firms work from the inside. So you're building your agents on real work and you get to meet a lot of these owners.
After six months, nine months, 12 months of doing good work for a bunch of firms, one of those owners is going to be ready to step back and you'll be the obvious person to take it over. So you start by serving them, you earn their trust, and eventually you own one. That's how usually these things happen.
Okay, so when I posted this to X, like I said, I had actually in my DMs several billionaires, several of the biggest firms just wanted to talk more about this with me. And then I had a bunch of people being like, keep this quiet. But then I had a ton of quote retweets, plenty of smart people telling me why this won't work.
Some of them have a point or there's a reason why I understand where they're coming from. So I wanted to do a lightning round of some of the common complaints about doing AI roll -up. So the first thing that I saw a lot of is roll -ups always fail.
This is private equity with an AI sticker on it. A lot of people saying, this is just private equity, this is private equity. And a lot of private equity fails.
So yes, a lot of roll -ups do fail, but they almost always fail for the same reasons. Like they overpay or they buy faster than they can integrate or the culture falls apart. And yes, AI doesn't fix any of that on its own.
That's actually why I like the one -person version or the small team version. You buy slowly, you're the first one doing the integration, and you price the business on what it makes today instead of paying the seller for the AI upside you're planning to create. I understand where this is coming from, but it's like saying, don't do startups because most startups fail.
The second thing that I heard a lot of is once everyone uses AI, the margins are going to get competed away. Eventually, probably yes, but there's a window between when your costs drop and when prices in your industry catch up. And in a lot of these industries, that window is actually years long.
It's not like a month. And the stuff that keeps clients around, like 20 years of relationships and your own list of rules, for how the work gets done.
Like it's actually harder than people think to copy. The third thing I heard a lot of was these are regulated and high stakes businesses. You need human beings checking everything.
There's no savings. Yeah, you need human beings check. You know, in some industries, you need human beings checking everything.
But checking a draft takes a lot less time than making it from scratch. And remember we were talking about Larson Gross' 31 % margin?
Well, that's where that came from. The people are still there, but what they spend their day on is different, and now they have just a whole lot more leverage. It's basically like you're adding leverage, not in the financial leverage sense.
In the productivity sense of the word, they're just way more leveraged to do stuff. Fourth thing that people were saying was, people hate change, the staff and the clients are going to leave. I take this the most seriously.
Some will. And that's why the first 30 days changed nothing the clients can see. why the agents run in the background before they touch anything real, and why the person who knows every client becomes a GM with a real piece of upside.
So you want to really incentivize your senior leadership. I think that's really important.
I'll do one last one. A bunch of replies saying, this is a fancy way of saying layoffs. And I get why people feel that way.
It's a fair question to ask. But what the early examples show is people are shifting from doing the work to checking it, and each person is handling more clients than before. So we talked about that with the interns at Larson Gross.
I'd be lying if I said no jobs are going to change. A lot of them will. And then the question becomes, if you've now evolved your role, some people are going to be...
good in that new role. Some people are going to be just satisfactory and some people are just not going to be meeting expectations. Those people are going to probably get laid off.
We haven't seen a lot of that yet. But I expect it will happen. Yes.
So overall, like the TLDR on some of the replies I got, there are real risks with this type of business and business model. But none of them changed my mind about the size of the opportunity.
And none of it changes my mind about how you approach the opportunity. So that's kind of how I see this $5 trillion AI roll -up opportunity. I think it's a huge, huge deal.
It's happening. The early stories are remarkable.
And I do think that, like I said, there's tons of businesses that are too small for these big funds. I think this idea of a one -person holdco or small holdco is really interesting. A few good businesses, a great GM running each, a shared layer of agents and rules underneath that gets better every week.
If people want to hear more about these types of businesses, holdcos, I call it multi -preneurship. This idea of... You have entrepreneurship is obviously you're starting businesses.
And multi -preneurship is when you own multiple businesses. I think it's a huge opportunity. I think you can cash flow it.
And I hope this episode got the creative juices flowing. Like I said in the intro, I can expand on this and I can put it in a pinned comment. I can put up the folder structure, the agent files, any prompts in there.
If that hits 5 ,000 likes, comments, and subscribes, I'll put in the pinned comment. And if you start one of these businesses, tell me because I generally love to hear about it. I love doing these videos and I love putting out the alpha.
And although I might get some haters for it, for me it's worth it because I think if it changes the lives of a few, it's absolutely worth it. This has been really fun and I'll see you in the next one. Take care.
The Hook
The bait, then the rug-pull.
Greg Isenberg opens with a blunt number: $5 trillion in small businesses are about to change hands as their owners retire, and he says the AI agents now good enough to run the back office are the real unlock nobody's pricing in yet.
Frameworks
Named ideas worth stealing.
08:48list
The Fund Playbook in 5 Steps
Build the platform
Buy a trusted firm
Run agents in the background
Move the back office over
Buy the next firm
How Thrive and General Catalyst turn a service firm from 5-10% profit into 30-40% profit without changing revenue.
Steal forany small-business acquisition or agency-to-holdco transition plan
15:03list
The Holdco Folder Structure
/thesis (industries.md, what-good-looks-like.md)
/shared (agents, global rules, test golden-files, runbooks)
The three-folder system that keeps a shared agent layer usable across every business in the holdco, with global rules, business rules, and the corrections log as the minimum viable version.
Steal forstructuring any multi-entity AI-agent operation
17:20model
Agents Draft, A Person Sends
Intake
Preparer
Reviewer
Person approves
Client gets the work
A five-stage pipeline where AI agents do all the drafting and checking, but only a human can send anything to a client, and the reviewer agent can block but never send.
Steal forany AI-assisted client-facing workflow that needs a trust safety rail
20:30list
The Corrections Log Loop
Compare each agent draft to the human-approved version
Sort every change into factual error, client preference, missing information, or style
Write a rule for any correction that happened more than once
Add each approved rule as a test using the original input and accepted output
The weekly process that turns human corrections into permanent rules and tests, compounding an agent's accuracy over time.
Steal forany recurring AI-agent QA process
CTA Breakdown
How they asked for the click.
VERBAL ASK
01:40subscribe
“If this episode gets 5,000 likes, comments, subscribes, I'll give away more of my notes on this whole topic. It'll be in the pinned comment.”
Engagement-gated content giveaway, repeated a second time mid-video before the folder-structure section, tying the ask directly to the payoff viewers are already watching for.
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 builds a one-person creative agency where Astra plans and repairs the work, 50+ Higgsfield models produce it, and he just approves the output and gets paid.
Greg Isenberg and Remy put the invite-only Instinct agent through real errands, a haircut, a restaurant table, and a Bali visa, and find out it keeps a copy of your email even after you disconnect it.
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 names the role he thinks AI agents are about to make the most valuable job in tech, and hands over the folder structure, tool stack, and 30-day plan to become one.
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.
00:34
00:46
01:07
01:29
02:25
02:48
03:10
03:33
03:55
04:18
04:40
05:03
05:48
06:21
06:34
06:45
07:18
07:41
08:06
08:26
09:11
09:33
09:55
10:29
10:42
10:56
11:26
12:13
12:37
13:00
13:24
13:47
14:11
14:23
14:49
15:29
16:12
16:35
16:47
17:42
18:04
18:26
19:09
19:30
19:51
20:34
20:55
21:13
21:30
21:54
22:18
22:24
23:03
23:30
24:23
24:50
25:16
25:43
26:09
26:36
27:03
27:29
27:43
28:14
28:35
28:56
29:17
48:05
27:30
22:52
24:43
35:18
28:58