Agent
The framework already speaks of you, your team, your company, your VP — each is an Agent at a different scale. The word travels into plain language without translation.
Definition
An Agent is any entity that takes in information from its surroundings, models that information into some understanding, and acts on its surroundings as a result. Three properties define the category jointly: an Agent has a Computational Boundary (it actually receives information from outside itself), a Cognitive Boundary (it does some modeling on what it receives, however minimal), and a Causal Boundary (it can affect its surroundings). The framework treats the category as scale-free and substrate-free — a person, an organization, a biological cell, and an AI system all qualify if and only if they have all three. Agent is not a label for what is “smart” or “purposive” in any colloquial sense; those are downstream descriptions of what kind of Agent it is, not threshold criteria for the category. An entity with no intake at all is not an Agent. An entity with no causal effect at all is not an Agent. An entity with intake and effect but no modeling is also not an Agent — it is a Tool.
The framework takes a physicalist stance on Agent identity: an Agent is the persistent thing that bears the three Boundaries continuously across situations. A person, a company, an AI system, an ant colony — each is one Agent across time, and what shifts with role, context, or environment is the shape of its Boundaries, not its Agent-identity. Substrate is defined functionally (whatever persists across situations and bears the three Boundaries continuously), not physically (the substrate does not have to be a single physical body). This is a model-clarity choice — it makes the framework legible to both researcher and practitioner audiences without forcing ontological commitments the framework does not need.
Relations
Every Agent has all three Boundaries — Computational, Cognitive, and Causal. Tool is the contrastive category: a non-Agent that an Agent uses to extend its reach, on either the Computational side (a telescope, which extends what the Agent can see) or the Causal side (a gun, which extends what the Agent can do). Composition is what happens when Agents combine. Every Agent’s Boundaries can be viewed from the perspective of Capacity (what is theoretically possible for the Agent) or Realized (what the Agent actually does given its information and environment). The framework’s Turtles All The Way Down principle holds that the same Agent test applies at every scale.
Example — CEO
A sales VP is an Agent. Market data and account reports cross his Computational Boundary; he forms a model of account risk and revenue trajectory at his Cognitive Boundary; he reassigns reps and escalates deals through his Causal Boundary. The company he works inside is also an Agent at a larger scale, with its own three Boundaries. Those Boundaries are not found by adding up the Boundaries of its member Agents — the company can both see and reach things no member can (because of how members are arranged and how their Boundaries connect), and at the same time fail to see or reach things any individual member could (because the arrangement loses signal in transit). The arithmetic is structural, not summative; Composition is where the framework names the principles that govern it. The same person in a board meeting receives different Actual Input, engages different regions of his modeling, and exercises different slices of his reach than the same person at a family dinner — but the underlying Boundaries are the same throughout. What shifts with context is which region the Agent engages, not the Boundary itself.
Example — Research
(Forthcoming.)