Cooperative Agents: a glossary
The Term, Defined Precisely
Source material for this terminology: applied AI-engineering explainer content on cooperative agents, narrative intelligence, and governance architecture (LangGraph/AutoGen-adjacent developer education, reviewed June 2026). These are working engineering definitions, not peer-reviewed research — cite them as such if asked. IS's contribution is applying this vocabulary to institutional governance, where it does not yet exist.
Cooperative agent — an autonomous or semi-autonomous AI participant that must not only complete a task, but do so within the bounds of a social or organizational contract. Distinct from a single-turn tool: it reasons, plans, and self-corrects across a loop, not a single call.
Three required capabilities of a cooperative agent:
State Awareness — understanding current context, conversation history, and the status of governance constraints.
Intent Alignment — ensuring actions serve the user's or institution's actual goals, not just a proxy objective.
Reflexive Communication — the ability to explain its process or ask for clarification when a constraint blocks a task.
The Reasoning-Action Loop (ReAct) — the engine of a cooperative agent: Observation → Thought → Action → Governance Check → Correction. The "Governance Check" step is where accountability gets enforced in real time, not after the fact.
The Multi-Agent Feedback Loop — a Narrative Agent (proposes a response, optimized for context and clarity) working in tandem with a Guardrail Agent (audits the response against policy). The Narrative Agent iterates until the Guardrail Agent clears the output. Two specialized agents outperform one model trying to balance persuasion and safety at once.
Human-in-the-loop (HITL) gate — the point at which governance architecture requires the system to pause and route a decision to a person, rather than resolve it autonomously. Well-designed cooperative systems make this pause seamless, not an interruption.
Why this maps onto IS exactly: Narrative Intelligence is the Narrative Agent function — creating context, meaning, and flow, understanding how institutional narratives establish credibility. Governance Architecture is the Guardrail Agent function — auditing for safety, compliance, and control. IS has been operating this two-role loop as an institutional practice (Verity's schema gate, the ODNI narrative-mechanism assessment, the Housing Authority governance analysis) since before "cooperative agents" was a phrase anyone used outside developer tooling.