{
  "name": "Agentic AI Lexicon",
  "description": "A lexicon of agentic AI concepts — agents, orchestration, identity, policy and execution layers.",
  "url": "https://gianlucabusato.com/datasets/agentic-ai-lexicon/",
  "creator": {
    "name": "Gianluca Busato",
    "id": "https://gianlucabusato.com/#person"
  },
  "dateModified": "2026-07-15",
  "inLanguage": "en",
  "columns": [
    "term",
    "definition"
  ],
  "records": [
    {
      "term": "Plan-Act-Observe Loop",
      "definition": "The core agent cycle: form a plan, execute an action, observe the result, and iterate."
    },
    {
      "term": "Multi-Agent System",
      "definition": "A system where several agents coordinate, negotiate or compete to achieve goals."
    },
    {
      "term": "Agent Identity",
      "definition": "A cryptographically verifiable identity assigned to an autonomous agent."
    },
    {
      "term": "Action Mediation",
      "definition": "Intercepting an agent's intended actions to check them against policy before execution."
    },
    {
      "term": "Execution Layer",
      "definition": "The controlled runtime in which an agent's actions are mediated and logged."
    },
    {
      "term": "Policy Layer",
      "definition": "The layer that defines and evaluates the deterministic rules governing agent behavior."
    },
    {
      "term": "Identity Layer",
      "definition": "The layer that establishes and verifies who an agent is acting as."
    },
    {
      "term": "Audit Layer",
      "definition": "The layer that records an immutable trail of agent decisions and actions."
    },
    {
      "term": "Capability Token",
      "definition": "A scoped credential granting an agent a specific, time-bound permission."
    },
    {
      "term": "Delegation",
      "definition": "Granting one agent the authority to act on behalf of a user or another agent."
    },
    {
      "term": "Escalation Policy",
      "definition": "Rules defining when an action must be routed to a human or higher authority."
    },
    {
      "term": "Blast Radius",
      "definition": "The maximum potential impact of an agent action if it goes wrong."
    },
    {
      "term": "Containment",
      "definition": "Limiting an agent's reach so failures cannot propagate across systems."
    },
    {
      "term": "Deterministic Checkpoint",
      "definition": "A fixed control point where the same conditions always yield the same governance decision."
    },
    {
      "term": "Agent Registry",
      "definition": "A catalog of known agents, their identities, roles and permitted capabilities."
    },
    {
      "term": "Context Window",
      "definition": "The span of information an LLM can consider at once when reasoning."
    },
    {
      "term": "Grounding",
      "definition": "Anchoring an agent's outputs to verified data or sources to reduce hallucination."
    },
    {
      "term": "Hallucination",
      "definition": "A confident but unsupported or fabricated output from a model."
    },
    {
      "term": "Guard Agent",
      "definition": "A supervisory agent that monitors and constrains other agents."
    },
    {
      "term": "Orchestrator",
      "definition": "The component that assigns tasks, sequences agents and enforces shared policy."
    }
  ]
}