# Agentic Glossary > The vocabulary of AI agents, quoted at the source. Every term defined directly from a canonical primary source — Anthropic, Google, NIST, Gartner, McKinsey, or the foundational paper. Most glossaries paraphrase. This one quotes. Maintained by AgentsBooks (https://agentsbooks.com). Last refresh: 2026-05-07. ## TL;DR The terms that matter for agents in 2026 — *agent*, *MCP*, *A2A*, *RAG*, *Constitutional AI*, *LangGraph / CrewAI / AutoGen*, and the *8 primitives* on which agentic firms are built — defined with direct quotes from Anthropic, Google, NIST, the EU Commission, and Gartner. Foundational entries are flagged Foundational; live regulation is flagged In force; brand-new vocabulary is flagged Emerging 2026; contested entries carry a counter-claim. ## Categories (31 terms total in v0) - **8 Primitives** (8) — Identity, Brain, Heart, Memory, Control, Knowledge, Friends, Shares - **Core Concepts** (6) — Agent, Agentic AI, Agentic firm, Agent fleet, Workflow, Tool use - **Protocols & Standards** (5) — MCP, A2A, Agent Card, Constitutional AI, RAG - **Frameworks** (4) — LangGraph, CrewAI, AutoGen, Multi-agent system - **Operations** (4) — ADLC, AX, Context graph, Agent management platform - **Compliance & Risk** (4) — AI RMF (NIST), High-risk AI system (EU AI Act), AI Agent Interoperability Profile, Reliability gap ## Selected entries with canonical quotes ### Agent *An AI system that decides for itself how to accomplish a task — choosing tools, sequencing steps, observing results, and adjusting — rather than executing a fixed script.* > "An agent is an AI model that directs its own processes and tool use when accomplishing a task — that is, deciding for itself how to achieve what users want, rather than following a fixed script." — Anthropic, *Building Effective Agents* (https://www.anthropic.com/research/building-effective-agents) ### Agentic AI *The category of AI systems that perceive, decide, and act toward goals over multiple steps — distinct from chatbots that only respond.* As of 2026, Gartner places agentic AI at the Peak of Inflated Expectations: 17% deployed, 60% intend within two years. > "Agentic AI represents the shift from 'tools that talk' (chatbots) to 'tools that do' (autonomous agents)." — Gartner, *Hype Cycle for Agentic AI 2026* (https://www.gartner.com/en/articles/hype-cycle-for-agentic-ai) **Contested:** Gary Marcus argues agents have not yet proven reliable, citing unsolved problems in planning, reasoning, and factuality (https://garymarcus.substack.com/p/six-or-seven-predictions-for-ai-2026). ### MCP / Model Context Protocol *An open standard that defines how AI systems integrate with external data sources and tools.* Introduced by Anthropic November 2024, donated to the Linux Foundation's Agentic AI Foundation December 2025. As of December 2025: 97 million monthly SDK downloads, 10,000+ active servers. > "An open protocol that enables seamless integration between LLM applications and external data sources and tools." — MCP Specification (https://modelcontextprotocol.io/specification/2025-11-25) ### A2A / Agent2Agent Protocol *An open standard for communication and interoperability between AI agents built by different vendors.* Originally Google, donated to Linux Foundation June 2025. As of April 2026: 150+ supporting organisations including Google, Microsoft, AWS, Salesforce, SAP, ServiceNow, Workday, IBM. > "A2A provides the definitive common language for agent interoperability in a world where agents are built using diverse frameworks and by different vendors." — A2A Protocol Specification (https://a2a-protocol.org/latest/specification/) ### Constitutional AI [Foundational] *A training method that gives an AI system a written set of principles and uses the model itself to critique and revise its outputs against those principles.* Introduced by Anthropic in 2022. > "Constitutional AI is a method for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs." — Bai et al. (Anthropic), *Constitutional AI: Harmlessness from AI Feedback*, arXiv:2212.08073 ### RAG / Retrieval-Augmented Generation [Foundational] *An architecture combining a generative model with an external retrieval step.* Coined by Lewis et al. (Meta/UCL) in 2020. > "We introduce RAG models which combine pre-trained parametric and non-parametric memory for language generation." — Lewis et al., *Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks*, arXiv:2005.11401 ### High-risk AI system (EU AI Act) [In force 2026-08-02] *Under the EU AI Act, AI systems falling into the categories listed in Annex III.* Subject to conformity assessment, transparency, human oversight, robustness, and cybersecurity obligations. Currently scheduled for 2 August 2026 enforcement; the Digital Omnibus on AI (November 2025) proposes deferral to 2 December 2027 — pending trilogue resolution. > "The majority of rules of the AI Act come into force and enforcement starts on 2 August 2026, with rules for high-risk AI systems in Annex III entering into application." — European Commission (https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act) ### AI RMF (NIST) [In force] *The U.S. NIST AI Risk Management Framework — voluntary, four core functions: Govern, Map, Measure, Manage.* NIST announced an AI Agent Standards Initiative through CAISI in February 2026, with an AI Agent Interoperability Profile planned for Q4 2026. Source: NIST AI 100-1 (https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf) ### LangGraph / CrewAI / AutoGen The three production-mature multi-agent frameworks of 2026. LangGraph (LangChain): nodes-and-edges graph, most production-mature. CrewAI: role-based teams, easiest learning curve. AutoGen (Microsoft): conversational group-chat metaphor. ## See also - AgentsBooks 8-Primitives pillar: https://agentsbooks.com/blog/eight-primitives-agentic-firm - AgentsBooks Anatomy of a Firm: https://agentsbooks.com/anatomy - Methodology & full bibliography: /pages/methodology.html - Try AgentsBooks Free: https://agentsbooks.com/login?returnTo=/onboarding --- *Last research refresh: 2026-05-07. Built by AgentsBooks.*