Not assisted.Operated.
Over the next few years, a large number of companies will be run end to end by AI agents. This is what we believe about that shift, what we are building for it, and how we expect to be judged.
Our core thesis
The agents will run the company. The humans will own it.
They will answer the customers, ship the fixes, run the ads, file the taxes and write the weekly report. Most of these companies will be small. Many will be boring.
Together they will form a new kind of asset: revenue that behaves like software, with a cost base that behaves like infrastructure.
We think buying one will feel less like acquiring a business and more like acquiring an operating system, with customers, revenue and a measurable cost base.
When a company has no employees, its knowledge can be inspected.
Traditional small-business M&A is haunted by key-person risk. The founder leaves, the customers leave, the tacit knowledge leaves.
An agent-run company inverts this. Its knowledge lives in the prompts, the runbooks, the evals and the tool integrations, where a buyer can read it.
A successful transfer still needs a plan for accounts, permissions, contracts, and the operating knowledge a new owner needs.
Diligence should read the system, not a teaser deck.
Instead of interviewing staff, a buyer should be able to:
- Read the agent architecture.
- Replay a week of decisions.
- Check that failover actually fails over.
The marketplace should make that easy, not hide it.
Autonomy should be priced like a yield, not like a job.
A business that needs three hours of human attention a week is not a job. As the supply of truly autonomous companies grows and buyers learn to trust the evidence, we expect valuation multiples to decouple from traditional small-business patterns.
This is our thesis, not a forecast or a promise of returns. Revenue quality, durability, dependency risk and transferability still matter.
We want to be the venue where that repricing happens, and the name buyers already trust when it does.
Being early is the whole point.
“Fully automated” is a claim. The Autonomy Score is a measurement.
The score is shown on every listing. Sellers cannot type a number, and buyers can read exactly how it was produced.
| A claim | The Autonomy Score |
|---|---|
| Typed in by the seller | Computed from a structured questionnaire |
| No published method | Produced by a published formula |
| Written once | Recomputed by the database on every change |
| Unchecked | Inputs reviewed against evidence on verified listings |
Look beyond revenue.
Understand the autonomy.
Two businesses can earn the same revenue and demand very different things from their owners. Our Autonomy Score makes that difference easier to inspect.
A published formula. Calculated from seller inputs.
What the agents do
How much of sales, marketing, support, operations, product & engineering, and finance is handled by agents. The average coverage across those six functions contributes up to 50 points.
How we expect to be judged.
Transparency over polish.
Metrics are labelled self-reported until we have checked them. Sample listings are marked as such.
Diligence in the open.
Agent stacks, model dependencies and resilience artefacts are part of the listing, not an afterthought.
The seller controls disclosure.
Legal names and URLs are released to a specific buyer, under an NDA, by the seller.
No dark patterns.
No fake urgency, no hidden fees, no pay-to-rank.