Skip to content
TREATISE / 09.0500PART 05 OF 10

Intelligence

Agents are not assistants bolted onto a product. They are citizens of the world with stated authority, auditable actions and owners who answer for them.

In this civilization an agent is an object like any other: it has an owner, a mandate, a scope of authority, a memory, and a public history of what it did. It can hold a role in an organisation, be assigned work, be paid, be suspended, and be retired. Because it is an object, it is governable.

Every action an agent takes is attributable. Every artefact it produces is marked as machine-made at the point of creation. The question 'was this written by a person?' is answered in the record, permanently, rather than guessed at by a detector.

OWNED MODELS, OWNED COMPUTE

A world that runs on borrowed intelligence is a tenant of whoever owns the weights. So the intelligence layer descends into the stack: owned models trained for the world's own objects, governed open source where openness is the safer choice, private cloud for anything touching the vault, and eventually compute the civilization holds rather than rents.

This is expensive and slow, and it is done anyway, because sovereignty that stops at the API boundary is decorative.

LIMITS WE BIND OURSELVES TO

Intelligence here does not optimise for time-on-surface, does not simulate affection to increase retention, and does not act on a person's behalf beyond the authority they granted in writing. Where a decision affects someone's access to work, money, learning or standing, a human is accountable and the reasoning is disclosable to the person affected.

These are not aspirations printed on a wall. They are constraints the constitution states, the architecture enforces, and the public record makes checkable.

AGENT ACTIONS
ATTRIBUTABLE
MACHINE-MADE
MARKED AT CREATION
AUTHORITY
EXPLICIT AND BOUNDED