Build it, prove it, roll it up.

AI3 Labs is starting a new kind of roll-up. An AI-first firm in one trade, built by us or by a founder already on AI3, runs on its own books until they show it is plainly cheaper and better than the incumbents. Then AI3 Labs takes a stake in it and brings the incumbents onto the same floor, by share option or by purchase, with their owners still in the chair. Here is how it works, why we are doing it in this order, and what we are offering the owner of a firm today.

Joe Charlesworth, founder, AI3 · 2 October 2026 · 10 min read

The bet we share

In June, Thrive Holdings, a New York holding company, published an essay called Long Humans. In markets, to go long is to bet that something gets more valuable over time. The world, they say, is going short on humans, wagering that AI makes people redundant. They bet the opposite: as routine work gets cheap and abundant, the things only a person can hold, the relationship, the judgement behind it and the trust earned over years, get scarcer and dearer. They act on that bet by buying advisory firms, accountants and IT shops, keeping the founders as owners alongside them, holding forever, and rebuilding each firm from the inside.

We think they are right about humans. Our last post said the same thing about accountants: the client was never paying for arithmetic. Thrive's example is a partner accountant who, with one of their tools, cut her tax preparation from 180 hours to 15 last season and spent the time walking every client through their return. The figure is theirs. The lesson is the one this whole site is built on.

Where we differ is the order. Thrive buys the firm first and then works out how to rebuild it, which is why they need permanent capital, engineers in the building and years of patience per firm. We are going to do it the other way round. Start from a firm that is already AI-first, whether we built it or a founder did. Prove it on its books. Then roll up the incumbents onto what has already been proven. The firm at the centre does not have to be ours; it has to be on AI3, and it has to be better. This post is the announcement of that programme, and of what it offers an owner, whichever side of it they are on.

Build, prove, roll up, pave

The programme has four steps, and the order is the whole point.

1. Build, or back. The programme starts from one AI-first firm in a trade, an organisation on ai3.co staffed with workers from the day it opens: the receptionist, the bookkeeper, the one who chases, the one who files. It sells one defined unit of finished work at one price, never hours. A small number of people do what still needs a person. Sometimes AI3 Labs opens that firm itself, in the incubator. More often, we expect, a founder opens it, because that is what AI3 is for, and Labs backs it: a minority share option in return for the plan fee waived, credits, the playbooks, distribution through the directory, and the roll-up that follows. Either way it is an AI-first firm, not an old firm with a tool bolted on, and it runs on the same floor every organisation on AI3 gets: tasks, approvals, a rulebook, spending limits, its own books, a customer record and a calendar.

2. Prove. The firm runs on its books, and the books are the proof. Every unit it sells is a line in its ledger, with what it cost to produce beside what it was sold for, and whether the customer accepted it. When those books show a cost per unit, an acceptance rate and a turnaround that an incumbent in the same trade cannot match, the proof exists, and it is a ledger, not a slide. We do not publish it. A firm that has found a way to do the work at a margin the trade has never seen has no reason to show the trade how, and the books stay closed, as every organisation's books on AI3 are. What we do is open them across the table, under agreement, to the one owner we are talking to, beside her own. That is the moment Labs' stake in the firm earns its keep: the proving firm is the one the rest of the trade is rolled up around, and its books are the only argument we need to make.

3. Roll up. Then we go to the incumbents in that trade and make one of two offers. Either the owner keeps the firm and grants AI3 Labs a share option in return for being moved onto the proven floor, the plan fee waived for a period, and a forward-deployed worker beside them for the first ninety days. Or, where an owner wants out, Labs buys the firm outright, and the people who want to stay keep running the client side. In both cases the client keeps calling the same person. Behind that call, the work moves to the workers, under the rulebook the proving firm already wrote.

4. Pave. Every firm that joins catches mistakes the proving firm never saw, and each one becomes a rule. The rules that hold across firms become the trade's playbook, and the next firm to join starts from it. The trade gets better with every firm rolled in, and the proving firm, the one that was AI-first from the start, stays the place where the next improvement is tried before it reaches anyone else.

Thrive calls the result local trust, national capability. The brand stays on the door and the people stay close to their clients, while the technology, the operating knowledge and the capital compound across every firm that joins. We would describe the roll-up the same way. The owner's name on the door; the same operating system, and the same proving firm, behind every door.

The forward-deployed worker

There is a name for the engineer who goes and sits inside a customer's business, builds against the real data and owns whether it works: the forward-deployed engineer. Palantir invented the role when its intelligence customers could not say what they needed, and ran two kinds, the domain expert who found the highest-value problem and the builder who solved it fast for that one customer. The loop that made it a product business rather than a consultancy was what they called the gravel road and the paved highway: the rough thing built for one customer, studied by the core team and turned into a feature for all. By 2016 Palantir had more forward-deployed engineers than product engineers. OpenAI and Anthropic rebuilt the role in 2023, and a survey of 1,500 of them this spring, by Perspective, found the median senior one at a growth-stage company paid $355,000 a year and spending nearly half the week in customer meetings and travel. The same survey is blunt that the model does not reach small firms: a $350,000 engineer only pays against a six-figure process burden. Only the logic reaches them, a narrow slice of work, real data, a clear acceptance test.

On AI3 the worker is the forward-deployed engineer, and it reaches the firm of one. It sits inside the business, on the real books, record and calendar. It is judged on the owner's verdict, not on a deliverable. It costs what the model provider charges, with nothing on top. Its first job is the day-one task, a narrow slice with a written acceptance check. And when the owner returns a piece of work with a reason, that reason becomes a candidate rule, she promotes it with one tap, and the worker reads it before every later job. The gravel road is the organisation's rulebook. The paved highway is the trade's playbook. The step between them, reading what held across firms and proposing it to all, is the job of our product manager worker, and in the roll-up it is the job we care about most.

Thrive-sized firms, fifty people and thirty years of habit, still need a person in the building for a while. In the roll-up that person is a Labs engineer or one of our regional partners, there for the first ninety days to find the bottleneck and hand the build to the workers. After that the floor runs itself, and the owner runs the clients.

What the owner keeps

Thrive's three principles are the right ones for a roll-up that is long on humans, and we hold to them with one change each.

• Hold forever. Labs has no fund to wind down and no date to sell by. An option it takes vests against what the firm's own books show, never against a cash-out, and a seller who wants out this quarter is the wrong partner for the option and the right one for the purchase.
• Own the work. Thrive says only the owner can redesign the factory floor, and buys that permission with equity. We get it with a floor plan. The owner already has the decision rights; what she lacked was a floor a worker could be bolted to. Where Labs holds a stake, it never holds control, and the owner keeps leading.
• Build from the ground up. Thrive pairs an engineer with one advisor. Our owner is the advisor and teaches the worker herself, in the course of doing the work, and what she catches becomes the trade's playbook. Her name stays on the door.

And one thing Thrive does not have to spell out, which we do. Everything a firm accumulates on AI3, its books, its customers, its calendar and its rulebook, belongs to the firm, is readable over an API and the standard agent protocol by anything the owner chooses, and leaves with her if she goes. A roll-up that is long on humans has to be one the humans can leave.

The first trade

Accounting. It is the trade Thrive's essay turns on, it is the one Greg Isenberg's guide turns on, and it is the one AI3 has already built the hardest part of: the books. Every organisation on AI3 keeps double-entry books with invoices, bills, bank and wallet feeds, and spend attributed to the worker that spent it. A bookkeeping firm built on that floor starts with the ledger, the customer record, the calendar and the rulebook already in place, and sells the thing a small business actually wants, the closed month, at a price per month. The incumbents are small, numerous and, by every account including Thrive's, short of successors. If a founder is already building that firm on AI3, we would rather back it than compete with it. If nobody is, Labs opens it next.

What we are offering

Three things, from today.

• To the owner of a firm in the trade: join the roll-up. Keep the firm, keep leading it, keep your name on the door. Move onto the floor the proving firm has proven, with a forward-deployed worker beside you for ninety days and the plan fee waived while it beds in, in return for a minority share option to AI3 Labs that vests against what your own books show. If you would rather sell, say so, and we will talk about a purchase.
• To the founder of an AI-first firm, in any trade: be the proving firm. If you are already running one on AI3, or want to, Labs will back it with a minority share option in return for the fee waived, credits, the playbooks and the directory, and when your books prove it, we roll the trade up around you, with you at the centre of it. Accounting is the first trade we are looking at; tell us if yours is already further along.
• To everyone else: nothing changes. The template, the workers, the books and the directory are there for any business, and Labs takes no stake in a business it has not been invited into.

What to do about it

• If you own an accounting or bookkeeping firm: write to us with the firm's name and roughly how many clients. We will open the proving firm's books beside your own, under agreement, and you can judge.
• If you want to found the AI-first firm with Labs: write to us and name the trade.
• If you run any other business: connect your business, free, and hire a worker. Nothing about the roll-up touches you unless you ask it to.

More from the blog · RSS · Questions and corrections: contact. A correction that changes a fact is noted above with its date.