There is $5 trillion worth of boring small businesses about to change hands.
Accounting firms. Insurance agencies. Property management companies. Their owners started them in the 80s and 90s, they want to retire, and their kids went off to do something else.
Greg Isenberg posted about AI rollups on X and got over a million impressions. Then his DMs filled up with billionaires wanting to talk and founders begging him to stop sharing it publicly.
So he did the opposite. He made a full podcast episode about it.
I watched the whole thing and pulled out the playbook, the numbers, and the parts I would actually copy as a solo founder.
Why $5 trillion is up for grabs right now
The $5 trillion number comes from McKinsey. They expect about a million small businesses to actually sell by 2035 because their owners are retiring.
Greg's argument is that three things are happening at the same time, and when that happens you usually get a new kind of business.
→ Owners are retiring. Thirty years of loyal clients, no successor lined up, employees who don't want to run the whole thing.
→ AI can do the work now. Eight months ago you pointed AI at this stuff and got a rough draft someone had to redo. Today, set up correctly, you get a draft somebody checks. And look at what fills the day inside these firms: typing numbers from a PDF into a spreadsheet, chasing a client for the same missing document a third time, writing the same status update 50 times a week. Agents eat that.
→ Service businesses are priced like their margins are stuck. A normal service firm runs at 5 to 10% profit, and buyers value it as if that can never change.
Put those together: a wave of businesses changing owners, full of exactly the work agents are good at, priced as if nothing about them can improve.
People have talked about the boomer retirement wave for years. The new part is that the agents actually work now. Greg says he couldn't have made this episode eight months ago.
What Thrive and General Catalyst are actually doing
The big funds figured this out first, and the numbers are wild.
Thrive Capital (Josh Kushner's firm, early in Instagram and OpenAI) spun out Thrive Holdings. It bought close to 50 local accounting firms in 24 months and just committed another billion dollars to buy more.
One of them is Larson Gross, a firm in Bellingham, Washington that's been around since 1949. Five offices, 200 people. After Thrive got involved, the firm started running AI built on OpenAI's Codex.
The results from one tax season:
- 7,000 returns processed
- Accountants saved 31% of their time on average
- One accountant took a job that used to eat 180 hours a year down to 15
- The interns who used to prepare returns now train to review them instead
That last one is the detail Greg can't stop thinking about. The people who used to do the work are becoming the people who check it.
General Catalyst set aside $1.5 billion for the same strategy and has put more than $750 million into at least 10 companies. Long Lake buys property management companies: 18 businesses, $100 million in EBITDA in under two years, margins doubling. Crescendo runs customer support centers where AI now handles about 90% of frontline tickets.
The deal structure is smart too. General Catalyst companies typically pay 60 to 70% in cash, and the founder rolls about 30% equity into the new company. The person who built the client relationships for 30 years has a real reason to stay and make the handoff work.
One honest caveat Greg flags himself: a lot of these numbers are self-reported by young companies that are raising money, and none of them have been through a recession. Strong signal of where this is going, not proof.
The playbook: same revenue, 3x the profit
Strip the funds' approach down and it's four moves.
Build the platform first. The agents, the software, the rule system. Before buying anything.
Buy a firm people already trust. Good brand, clients who've been around for years.
Run the agents in the background. They work alongside the people and nobody outside sees anything change, except clients hear back faster.
Once the agents are reliable, move the back office over. Data entry, document collection, first drafts. Then buy the next firm and plug it into the same platform.
Watch what that does to the business. Revenue stays the same because the clients are the same and pay the same invoices. Costs go down because agents do a big share of the work. Profit goes from 5-10% to, if the thesis holds, 30-40%.
Same business. Three to four times the profit.
The one-person holdco version
Here's the gap: billion-dollar funds need big deals. A bookkeeping firm doing $2 million a year can't move their needle, so they won't even call. And most of McKinsey's million retiring businesses are exactly that size.
That's the opening for you.
Greg has run his own holding company for six years, and his one-person version looks like this: you at the top, a holding company that owns a few small businesses, and a GM running each one. The GM is usually someone already there. The senior bookkeeper who's been with the firm 15 years and knows every client. Make her the GM and give her a real piece of the upside.
Underneath everything sits a shared layer: the same agents, the same rules, the same dashboards for every business. You build it once, so the second business is easier than the first, and the third is easier again.
Why a solo founder can actually win here:
→ You use the same models the funds use. Thrive runs on Codex. You have Codex, Claude Code, and Gemini too.
→ You are the integration. The hardest part of any rollup is getting a new business onto the system without breaking anything. The funds hire managers who learn each business from scratch. You're the one in the room reading the agent drafts yourself.
→ Sellers pick people, not funds. If you spent 40 years building something, you care who takes it over. Plenty of owners will sell to a person before they sell to a venture firm.
Greg even shares his folder structure: a thesis folder (what you believe, so you don't buy something just because you got excited), a shared folder (agents, global rules, runbooks, a library of accepted work you test the agents against), and one folder per business with a clients file, a people file, a rules file, and a corrections log.
If you only set up three files, he says, make it the global rules, each business's rules, and the corrections log. That's what turns agents into something you can trust with a client.
The agent pipeline (and the one rule that prevents disasters)
Inside each business, work moves through four hands.
An intake agent collects documents and chases anything missing. A preparer agent does the first draft: categorizing transactions, drafting the month-end close. A reviewer agent checks the draft against the rules and sends it back if something's off. Then a human approves it, usually the same person who used to do the work by hand.
The rule that matters: the reviewer can block, but it can never send. Nothing reaches a client without a person approving it.
Every agent gets a plain-English job description. What its job is, what it's allowed to do, what it can never do, and when it has to stop and ask a human. Greg's take is that most agent problems come from nobody telling the agent where its job ends.
This is the same shift I keep seeing everywhere: founders stop hiring for the doing and start designing systems that do the doing. I broke down how that looks in practice in my piece on the AI agent workforce, and TwiLead has a solid practical guide to business automations if you want the nuts and bolts of wiring this into a normal service business.
What your week actually looks like
People assume owning several businesses is hectic. Greg says it's the opposite, because the GMs and agents handle the day-to-day. Here's his week:
Monday: numbers. Five metrics per business: profit margin, minutes of human time per job, how often agent drafts need fixing, client retention, and whether the key people are happy and staying. If profit is up while clients are leaving, something's wrong and you want to catch it early.
Tuesday: one call with each GM. What's working, what's annoying them, which clients need attention.
Wednesday: the corrections log. Greg calls it the most important hour of the week. Go through every fix a human made to an agent's work and turn the repeating ones into rules, each rule backed by a test. After a few hundred jobs, that list of rules becomes the most valuable thing you own. Anyone can use the same models. Nobody else has your list of every way they go wrong in your niche.
Thursday and Friday: finding the next business. Coffee with owners, learning an industry. Greg's warning: once deals hit marketplaces like BizBuySell, they're mostly picked over. Direct outreach wins.
And the first deal? Don't start by buying. Start by selling agents as a service to one industry. Take one annoying job (cleaning up month-end for their messiest clients, chasing missing documents) and do it well. You learn the industry from the inside, you build your agents on real work, and after 6 to 12 months one of those owners is ready to step back. You're the obvious person to take over.
That service-first path also works if you never buy anything. It's basically the AI employee model applied to back offices instead of marketing.
The objections worth taking seriously
Greg's post got quote-tweeted by plenty of smart people telling him why this won't work. He answered the big ones.
"Rollups always fail, this is private equity with an AI sticker." A lot of rollups do fail. They overpay, buy faster than they can integrate, or the culture collapses. AI fixes none of that. The one-person version answers it differently: buy slowly, do the integration yourself, and price the business on what it makes today instead of paying the seller for the AI upside you plan to create.
"Once everyone uses AI, the margins get competed away." Eventually, probably yes. But there's a window of years between when your costs drop and when prices in the industry catch up. And 20-year client relationships plus your corrections log are harder to copy than people think.
"These are regulated businesses, humans must check everything, so there are no savings." Humans do check everything. Checking a draft takes far less time than making it from scratch. That's exactly where Larson Gross's 31% came from.
"Staff and clients will leave." This is the one Greg takes most seriously. Some will. That's why the first 30 days change nothing clients can see, why agents run in the background before touching anything real, and why the person who knows every client becomes a GM with real upside.
"This is a fancy way of saying layoffs." His honest answer: roles are shifting from doing to reviewing, and each person handles more clients. Some jobs will change, and some people won't make the jump. He expects layoffs will come. He's not pretending otherwise.
FAQ
What is an AI rollup?
An AI rollup is buying established service businesses (accounting, property management, support centers) and running AI agents through their back office to raise profit margins. Revenue stays the same because the clients stay. Costs drop because agents handle intake, first drafts, and document chasing. The thesis: margins go from 5-10% to 30-40%.
Who is doing AI rollups right now?
Thrive Holdings (spun out of Josh Kushner's Thrive Capital) has bought close to 50 accounting firms in 24 months and committed another $1 billion. General Catalyst set aside $1.5 billion and funded companies like Long Lake (18 property management businesses, $100M EBITDA in under two years) and Crescendo (AI handling ~90% of frontline support tickets).
How do you start an AI rollup with no capital?
Greg's path: don't buy first, serve first. Pick one industry and sell an agent-powered service that fixes one annoying job, like month-end cleanup or document collection. After 6 to 12 months of good work, you know the industry, your agents are trained on real work, and retiring owners trust you enough to sell to you, often rolling equity so you need less cash upfront.
Do AI rollups replace all the employees?
No. The pipeline keeps a human approval step on everything that reaches a client, and the reviewer agent can block work but never send it. At Larson Gross, interns who prepared tax returns now review them instead. Roles shift from doing to checking. Greg is upfront that some jobs will change and some layoffs will eventually happen.
Steal playbooks like this every week
Every week on the Profitable Founder Podcast I sit down with bootstrapped founders who spotted opportunities like this one and turned them into real revenue. You hear the actual numbers, the deals, and the mistakes.
