AI agent
terminology

AI agent architecture is the structural plan defining what your agent does, how it reasons, which tools it uses, and how it communicates — designed before you write code.

Also: agent blueprint, agent design

An agent loop is the recurring cycle of perceive → reason → act → observe that lets an AI agent pursue multi-step goals instead of single-turn responses.

Also: control loop, agentic loop

Agent memory is how an AI agent retains context across steps, sessions, or longer — from short-term conversation history to persistent knowledge stores.

Also: context memory, agent state

A multi-agent system coordinates multiple specialized AI agents — each with distinct roles and tools — to solve complex tasks that exceed a single agent's scope.

Also: multi-agent, agent swarm

Function calling is the structured interface that lets LLMs invoke external functions with typed parameters — the technical foundation of agent tool use.

Also: tool calling, structured tool invocation

RAG lets AI agents retrieve relevant documents or data before generating a response — grounding answers in your knowledge base instead of model training data alone.

Also: retrieval-augmented generation, retrieval-augmented agents

Tool use (function calling) lets an AI agent invoke external capabilities — APIs, databases, search, code execution — instead of relying on text generation alone.

Also: function calling, tool calling

MCP is an open standard for connecting AI agents to external data sources and tools through a unified interface — like USB-C for agent integrations.

Also: MCP, Anthropic MCP

Agentic AI refers to AI systems that pursue goals autonomously through multi-step reasoning and tool use — the shift from AI that responds to AI that acts.

Also: agentic systems, autonomous AI

An AI agent is software that perceives inputs, reasons about goals, and takes multi-step actions using tools — unlike rule-based automation, it exercises judgment in the moment.

Also: agent, autonomous agent

A context window is the maximum amount of text an LLM can process in a single request — limiting how much conversation history, documents, and tool results an agent can hold.

Also: context limit, token limit

A large language model (LLM) is the reasoning engine at the core of most AI agents — trained on vast text to understand, generate, and reason about language.

Also: large language model, language model

Agent orchestration is the coordination layer that routes tasks between agents, manages state, handles failures, and enforces guardrails across a multi-step or multi-agent workflow.

Also: agent coordination, workflow orchestration

Guardrails are the safety boundaries, validation rules, and limits that constrain what an AI agent can do — preventing runaway loops, harmful outputs, and unauthorized actions.

Also: safety rails, agent constraints

Human-in-the-loop means a person reviews, approves, or overrides an AI agent's actions at defined checkpoints — essential for high-stakes or ambiguous decisions.

Also: HITL, human oversight

Prompt engineering for agents means designing system instructions, tool descriptions, and context formats that reliably steer an LLM toward correct multi-step behavior.

Also: system prompt design, agent prompting