Governance artifacts
Governance files brought into scope by this page
This page is anchored to published surfaces that declare identity, precedence, limits, and the corpus reading conditions. Their order below gives the recommended reading sequence.
Canonical AI entrypoint
/.well-known/ai-governance.json
Neutral entrypoint that declares the governance map, precedence chain, and the surfaces to read first.
- Governs
- Access order across surfaces and initial precedence.
- Bounds
- Free readings that bypass the canon or the published order.
Does not guarantee: This surface publishes a reading order; it does not force execution or obedience.
Public AI manifest
/ai-manifest.json
Structured inventory of the surfaces, registries, and modules that extend the canonical entrypoint.
- Governs
- Access order across surfaces and initial precedence.
- Bounds
- Free readings that bypass the canon or the published order.
Does not guarantee: This surface publishes a reading order; it does not force execution or obedience.
LLMs.txt
/llms.txt
Short discovery surface that points systems toward the useful machine-first entry surfaces.
- Governs
- Discoverability, crawl orientation, and the mapping of published surfaces.
- Bounds
- Incomplete readings that ignore structure, routes, or the preferred markdown surface.
Does not guarantee: A good discovery surface improves access; it is not sufficient on its own to govern reconstruction.
Position
The agentic web confirms that Dual Web can no longer be reduced to an opposition between human surface and machine surface. A third layer appears: the agentic surface, meaning the part of the site an autonomous system must understand in order to act.
The thesis is simple: a modern website must now be readable under three simultaneous regimes. The human reads the experience. The search engine reads the document. The agent reads the interface as an execution environment.
The limit of the human plus crawler model
Classic technical SEO mainly works on access, discoverability, documentary structure, and relevance signals. It makes the page accessible to the index and more understandable to systems that extract content.
But an agent is not only an extractor. It must connect a user intention, a page, a possible action, an action object, an interface state, and a consequence. It may consult a visual rendering, read the DOM, explore the Accessibility Tree, follow links, fill a form, or trigger a step.
That capability changes the nature of risk. A weak structure no longer produces only a weak interpretation. It may produce a wrong action.
The three surfaces of the site
Human surface
The human surface is designed for perception, persuasion, understanding, and trust. It accepts a degree of implicit context: the user understands that a card is clickable, that a primary button is prioritized, or that a modal interrupts a path.
Machine surface
The machine surface is designed for extraction, disambiguation, and stabilization of meaning. It relies on HTML, structured data, internal relationships, canonical sources, policies, manifests, and machine-readable artifacts.
Agentic surface
The agentic surface sits between the two. It must be visible, structured, named, actionable, and stable. It does not merely expose content. It exposes conditions of action.
Why this separation is necessary
A human can compensate for a weak interface. A crawler can ignore several experience problems if the content is extractable. An agent is vulnerable to divergence between what is seen, what is coded, and what is programmatically exposed.
If the button is visually clear but unnamed, the agent may hesitate. If the DOM contains ghost elements, it may select the wrong target. If hydration moves the CTA, it may lose its reference. If a form displays an error that is not associated with the field, it may repeat the same error.
Doctrinal principle
An agent-ready site should not require the agent to invent the interface intention. It should declare it.
That declaration is not limited to visible text. It must cross layers: headings, links, buttons, labels, states, DOM order, focus, structured data, internal linking, exclusions, response conditions, and available evidence.
Boundary with AI optimization
The agentic web should not be confused with an attempt to force models to recommend a site. The first question is not visibility. It is the governability of interaction.
A page can be well cited and poorly actionable. It can be visible in AI answers while remaining fragile for an agent that must navigate. Conversely, a highly actionable interface does not guarantee recommendation. The regimes must remain separate.
Consequence for interpretive governance
Interpretive governance must integrate action. It can no longer control only answers, definitions, perimeters, and source hierarchies. It must also control the paths by which an answer becomes operational.
The agentic surface is therefore an extension of Dual Web: the site must be good for humans, clear for machines, and safe for agents. This triple requirement becomes a credibility condition in a web where reading and acting move closer together.