AI Agents and Agentic Systems: A Glossary

Part of the Five Stones knowledge base - 24 terms on AI agents and agentic systems, in plain English.

Part of the Five Stones knowledge base - 24 terms on ai agents and agentic systems, in plain English.

AI agent - Software that can take actions on its own to reach a goal, not just answer a question, e.g. monitoring enquiries and actually booking the appointment.
Agentic AI - AI designed to work autonomously toward a goal: it breaks the goal into steps, carries them out, checks its own progress, and adjusts as needed.
Autonomous AI - AI acting without needing a human to approve every individual step, within boundaries a business has set for it.
Multi-agent system - A setup where several AI agents, each handling a different task, work together and hand work between each other.
Tool use (in AI) - An AI agent's ability to use external software (like a calendar, a database or a search engine) to complete part of its task.
Function calling - The technical mechanism that lets an AI model trigger a specific action in another piece of software, such as "create_booking()."
Orchestration (AI orchestration) - Managing how multiple AI agents or automated steps work together in the right order to complete a larger task.
Goal-directed behaviour - An AI system's ability to keep working toward an outcome across several steps, rather than only responding to one message at a time.
Reasoning (AI reasoning) - An AI model's ability to work through a problem in logical steps before producing an answer, rather than answering instantly.
Chain-of-thought - A technique where an AI model works through a problem step by step "out loud" before giving its final answer, often improving accuracy.
Retrieval-Augmented Generation (RAG) - A technique where an AI looks up relevant information from a business's own documents before answering, so it responds using real facts, not just its general training.
Vector database - A specialised database that stores information in a way that lets an AI quickly find the most relevant matches to a question, the technology behind most RAG systems.
Embedding - A numerical representation of text (or an image) that lets an AI system compare how similar two pieces of content are in meaning.
Knowledge base (AI-connected) - A business's documents, FAQs and policies made searchable by an AI system so it can answer accurately from real, current information.
Guardrails - Rules and limits deliberately built into an AI system to stop it taking actions or giving answers outside what is safe or approved.
Autonomy level - How much an AI agent is allowed to do without a human checking in first, ranging from "suggest only" to "act and report afterwards."
Task decomposition - An AI agent breaking a large goal down into smaller, ordered steps it can actually execute.
AI copilot - An AI assistant that works alongside a person, suggesting or drafting actions for a human to approve, rather than acting fully independently.
AI operator - An AI agent given permission to directly operate software (like a browser or an app) on a person's behalf to complete tasks.
Memory (AI agent memory) - An AI agent's ability to retain relevant information across separate sessions, rather than starting fresh every conversation.
Context engineering - Deliberately structuring the information an AI agent is given so it can complete a task accurately, the evolving discipline beyond simple prompt writing.
Agent framework - A software toolkit developers use to build AI agents, defining how they reason, use tools and hand off tasks.
Human oversight - A deliberately kept checkpoint where a person reviews or can override what an autonomous AI agent is doing.
Failure mode (AI agent) - A predictable way an AI agent can go wrong, which a well-built system is designed to detect and stop before it causes damage.

Looking for a different topic? Browse the full knowledge base, or tell us what's missing.