If the AI becomes the business, what is the business?

This is the question underneath every value-creation conversation I have, and almost nobody asks it out loud. Strip the labour out of a people business and what remains had better be worth owning. It is entirely possible to disrupt yourself enthusiastically, remove the work that made your judgement valuable, and end up cheaper and worth less — a smaller version of yourself with better margins for about two years, until the same tools reach your competitors and price follows cost down. Automation savings are competed away. Assets are not.

So run the scarcity question honestly. If intelligence can be rented by the token, execution is near-free, and everyone's strategy deck comes from the same models — what does your organisation still hold that a competitor with the same subscriptions does not? The honest list is shorter than people expect, and everything on it maps to something in this architecture:

What is actually scarce	Where it lives in the architecture
Proprietary data and context	The graph. Resolved entities, obligations, documents linked to the decisions they supported — context no vendor arrives with and no competitor can prompt their way into.
Judgement, and the tests that encode it	The harness. A versioned set of your own files with known correct outcomes is your professionals' judgement, bottled. It is what makes you model-agnostic in practice rather than on paper — and no vendor can sell it to you, because it is made of your business.
The decision record	Decisions as first-class citizens in the graph: who signed, on what evidence, with what reasoning. The models forget. The system remembers. Every human override is training data your competitor does not have.
Evidence of control	The register and the gates. What is running, on whose authority, how well, at what cost — the thing a regulator, an insurer and a buyer's diligence team all read.
Distribution and trust	Not built here, but protected here. In the enterprise, distribution, integration and permissioning matter as much as intelligence — value migrates to whoever owns the trust layer. For an incumbent that is already you, unless you rent it away.

The compounding test from Part six is the same question asked quarterly: is each deployment generating data and context that makes the next one better — accumulating an advantage others cannot enter — or does every initiative start from zero? An estate that compounds is an asset. A collection of vendor pilots is an expense with a logo.

For a private equity owner the framing is even simpler, because diligence will do it for you. A buyer discounts claimed savings and capitalises banked ones; that alone can be worth a large multiple of the work it takes to measure properly. But the deeper question a buyer's technical team now asks is portability: if this company's AI capability is a set of prompts inside one vendor's product, it is not an asset, it is a dependency with the company's logo on it. The register, the harness and the graph are what make the estate survive a model change in a day — and an estate that can change its mind about a vendor is the only kind worth paying for. In a corporate the same logic holds with different words: engine one funds engine two. Automate the core to fund the edge; transform at the edge to replace the core — and never confuse running the first for having done the second.

Turning what technology can do into what a business is worth — that is the entire job. Both halves are measurable, and the four owned artefacts are the bridge between them.