AI Agents Glossary

Plain-English definitions of the terms behind AI agents and agentic AI.

AI agent: Software that pursues a goal on its own: it plans steps, uses tools to act, observes results, and iterates until done. The LLM is its reasoning core; tools and a control loop let it act, not just answer. Learn more.
Agentic AI: AI systems that pursue goals autonomously by planning, using tools, and iterating, rather than only generating content on request. Agentic AI is generative AI plus tools, memory, and a control loop. Learn more.
Assistant (autonomy level): An AI that produces output when asked and takes no independent action. The lowest rung of the autonomy ladder.
Copilot (autonomy level): An AI that suggests an action inline (a completion, a drafted reply) which a human accepts and executes.
Supervised agent (autonomy level): An AI that performs multi-step work on its own but a human approves the consequential actions. The most common production pattern.
Autonomous agent (autonomy level): An AI that acts end to end within guardrails, with no per-action human approval. Reserved for narrow, verifiable tasks.
MCP (Model Context Protocol): An open standard for connecting AI agents to external tools and data sources, now stewarded by the Linux Foundation. The vertical agent-to-tool layer. Learn more.
A2A (Agent-to-Agent): An open protocol for agents to discover and coordinate with each other. The horizontal agent-to-agent layer, complementary to MCP.
Multi-agent system: An architecture where several specialized agents collaborate on a task, each handling a part, coordinated by an orchestrator or shared protocol.
Agent platform: A product for building, running, and orchestrating agents (often no-code), with prebuilt agents, integrations, and guardrails.
Agent framework: A developer library for building agents in code (e.g. LangChain, CrewAI). Autonomy is defined by the developer, not the framework.
Human-in-the-loop: A design where a human reviews or approves an agent's consequential actions, the safeguard that makes supervised agents shippable.
AEO / GEO: Answer-Engine / Generative-Engine Optimization: optimizing content to be discovered and cited by AI search and LLMs, alongside classic SEO.