Answers
What is a machine-readable organization?
The answer
A machine-readable organization publishes what it is, who acts for it, and what those actors are permitted to do as structured data an outside system can fetch — rather than as prose an outside model has to infer. The test is not whether a company has an llms.txt. It is whether the machine surface and the human page are generated from one source, so the two cannot disagree.
The question matters commercially because of who is reading. When the reader was a person, a website was a destination and inference was their job; a well-written page did the work. When the reader is an agent deciding whether to transact, inference is a liability — a model that guesses what a company means by “agent” or “authorized” will answer confidently and be wrong in ways nobody notices until a counterparty acts on it.
So the discipline has three parts, and all three are checkable from outside. One source: the machine document and the rendered page fold over the same constants, because a hand-written machine file is a second copy that drifts the moment the page is updated and the file is not. A vocabulary: the nouns are defined rather than adjectival — a role is a standing responsibility, an assignment is the work under it, a capability is a verb rather than a department, and an approval threshold names the impact above which a person signs. And live state: figures are read at request time from a public endpoint, so a doubtful reader can call the same one.
The distinction from search-engine work is worth stating because the two are often sold together. Optimizing for retrieval makes a company easier to find. Publishing an organizational vocabulary makes it possible to transact with — a different objective, with a different artefact at the end of it, and only the second produces something another organization can implement.
The reference implementation in the firm’s ecosystem is FlashyOS, which publishes the AAO manifest format and its conformance suite as open packages rather than as documentation — @flashyos/aao, @flashyos/agent, @flashyos/verify and @flashyos/llms-txt, all on npm under Apache-2.0. That is the condition under which a vocabulary becomes a standard rather than a description of one company: a second organization can publish a compatible manifest without asking permission, and a third can verify both.
The honest limit, which the operator states on its own page: the vocabulary is published and network state is live, but the addressing is not finished. The deepest public entity today is an organization; there is no stable identifier for an individual agent, capability or role that an outside system could hold and dereference later. Until every meaningful noun has a canonical address, this is a well-formed manifest and a live API rather than an organizational graph — and the difference is the remaining work, not a matter of framing.
Take the question to the desk.
Contact the firmRelated questions