Agent

III node keystone published
Published 2026-07-11

Type: Node

Working Definition: A bearer of action, perspective, intention, responsibility, or recognition within social-process reality.

Function in the Grammar: Agent provides the center from which actions, claims, commitments, and interpretations originate. Without agents, events happen but are not performed; claims are made but are not owned; commitments exist but are not borne. Agent is the grammar's answer to the question: who?


Formal Grammar Representation

Agent(a) iff
  a is capable of performing Action(s) [X₁, X₂, ...]
  a may occupy Role(s) [R₁, R₂, ...] within Institution(s)
  a may make Claim(s) [C₁, ...] that carry a's own standing
  a may bear Commitment(s) [M₁, ...] — obligations attributable specifically to a
  a holds Perspective(P): interpretation and evidence-weighting are agent-relative
  a persists identifiably across Events, Acts, and Contexts
  a may be recognized by Institution(I) as having standing to act within it

Agent(a) is a grammatical primitive: it is not derived from Action, Claim, or
Commitment — rather, Action, Claim, and Commitment each presuppose some Agent(a)
as their bearer. There is no well-formed Action, Claim, or Commitment without an
identified (even if provisional or contested) Agent to attribute it to.

Semantic Constraints

The formal expression Agent(a) is well-formed whenever something is treated as the bearer of an action, claim, or commitment — the syntax does not require that a actually have the capacity, standing, or moral status the attribution implies. Whether an entity genuinely counts as an agent in a sense that can bear responsibility — as opposed to being merely a locus through which actions pass — is a question the grammar cannot settle by form alone. Corporations, states, and AI systems can all satisfy Agent(a) syntactically; whether each is a genuine bearer of responsibility, a legal fiction, a distributed collective, or a tool whose outputs must be attributed to the human agent standing behind it, is an institutional and often contested judgment. The most consequential case in this project's own domain: an AI system may satisfy the formal shape of Agent(a) — it performs actions, its outputs can be logged and traced — without thereby holding the genuine agency that grounds responsibility. Institutional participation by such a system must be explicitly granted and scoped; the formal pattern does not do this work on its own.


Core Relations

Relation Notes
performs Action Action is attributed to an agent; this attribution grounds responsibility
occupies Role Roles are situational; an agent may occupy different roles in different contexts
makes Claim Claims are owned; authorship creates discursive commitment
bears Commitment Obligations, duties, and promises are borne by agents, not by events
interprets Evidence Perspective is agent-relative; the same evidence may be interpreted differently
is recognized by Institution Institutional standing is conferred by recognition, not merely asserted

Typical Questions


Examples

Agents are not limited to individual human beings:

Agent Notes
A person The paradigm case
A committee Collective agent with defined procedures
A court Institutional agent authorized to produce binding decisions
A corporation Legal person with standing, rights, and obligations
A state Sovereign agent in international contexts
A scientific community Distributed epistemic agent
A narrator Discursive agent within a text or record
A model assistant or AI system A new category requiring careful specification of capacity, authorization, and accountability — not a conventional agent; any institutional participation must be explicitly granted and scoped

Distinctions

Agent ≠ Role. A Role is the situational function an agent occupies. An agent can persist across roles and contexts; a role does not persist without an occupant. The judge and the parent may be the same person occupying different roles.

Agent ≠ Person. Person is a legal or moral status conferred by an institution. Agent is a functional category — the grammar uses it wherever something acts, claims, or bears responsibility, regardless of whether it has been granted personhood.

Agent ≠ Actor. Actor emphasizes performance; Agent emphasizes capacity, responsibility, and persistence. The same distinction appears between playing a role in a play and bearing responsibility for what you do.

Agent ≠ Subject. Subject is a grammatical or phenomenological category. Agent is a social-process category: it tracks who acted, who claimed, and who is accountable.


Common Failure Modes

Mode Description
Unattributed action An action occurs but no agent is identified as responsible for it — the record shows what happened without showing who did it. Where to look: consequential acts logged with no actor field, or with an actor field naming a process or system rather than an accountable agent.
False attribution Responsibility is assigned to the wrong agent — a delegate's act attributed to their principal, or vice versa, without the delegation itself being traceable. Where to look: a recorded actor that does not match who actually performed the act, discovered only when the record is checked against independent evidence.
Insufficient authorization The agent performed the action, but lacked the institutional standing required for it to have force — the action happened, but does not bind the institution. Where to look: acts performed by an agent whose role does not include authority for that act type.
Capacity gap The agent lacked the competence, information, or resources the attributed act required — the agent is real and identified, but was not actually equipped to do what is attributed to them. Where to look: acts attributed to an agent operating outside their documented expertise or authorized scope, especially under time pressure.
Diffuse agency Responsibility is distributed across so many agents, or so many intermediate steps, that no single agent can be held accountable for the outcome. Where to look: consequential outcomes where every contributing act was individually authorized, but no agent bears responsibility for the outcome as a whole.
Institutional non-recognition An agent's standing is contested or denied by the relevant institution, even though the agent performed the act in question. Where to look: acts performed in apparent good faith whose actor's standing to perform them the institution later disputes.
Capacity mistaken for agency An entity that can produce outputs — a model, a script, an automated process — is treated as an Agent in the full sense, bearing claims and commitments, when its participation was never explicitly granted or scoped by the institution. Where to look: institutional records naming a system or process as the bearer of a claim or commitment, rather than the human agent accountable for deploying it.

Minimum Viable Test Case

Agent(a₁) = human_author
  performs: Action(reviews draft)
  occupies: Role(editorial authority, AlwaysBecoming)
  bears: Commitment("publish only reviewed content")
  recognized_by: Institution(AlwaysBecoming) as holding publication authority

Agent(a₂) = model_assistant
  performs: Action(drafts content)
  occupies: Role(drafting authority, AlwaysBecoming)
  bears: Commitment("mark all drafts human_reviewed: false")
  recognized_by: Institution(AlwaysBecoming) as holding drafting authority only

Claim(c₁, claimant: a₂, proposition: "this entry is ready to publish")
  a₂ is a well-formed Agent — the claim has an identified bearer
  but a₂'s recognized standing does not extend to publication judgments
  → c₁ is a pending claim; asserting it does not make it true, authorized, or operative

Without an identified Agent, "this entry is ready to publish" would be an orphaned
assertion with no bearer to hold accountable for it and no standing to evaluate.
Agent is what makes a claim answerable to someone in the first place — prior to,
and independent of, whether that someone actually had the standing the claim assumes.

Cross-References

Required

Consequential

Evidential

Cross-References

This entry is AI-assisted. Reviewed by the human author before publication.