Plain-language definitions of the terms you will meet when comparing AI agent tools. Looking for how to choose one? Start with our 7-point checklist.
- AI agent
- A system that pursues a goal by deciding its own steps, using tools and checking results along the way, rather than only replying to a message.
- Agentic workflow
- A multi-step process in which a model plans, acts, observes the result and adjusts, often looping until a task is done.
- Autonomy level
- How much an agent can do without approval, from suggesting actions only to executing them independently.
- Context window
- The amount of text a model can consider at one time, including your instructions, documents and the conversation so far.
- Function calling
- A model feature that lets it request a specific tool or API call with structured inputs, which the surrounding software then runs.
- Guardrails
- Rules, filters and permission limits that restrict what an AI system can say or do.
- Hallucination
- A confident but false or unsupported output from a language model.
- Human in the loop
- A design where a person reviews or approves an agent's actions before they take effect.
- Large language model (LLM)
- A model trained on large amounts of text that predicts and generates language. Most AI agents use one as their reasoning core.
- Model Context Protocol (MCP)
- An open standard for connecting AI applications to external tools and data sources in a consistent way.
- Orchestration
- The coordination layer that decides which model, tool or agent runs next and passes information between them.
- Prompt injection
- An attack where hidden instructions inside content an agent reads trick it into doing something its user did not intend.
- Retrieval-augmented generation (RAG)
- A technique where a model first looks up relevant documents and then uses them to write its answer.
- Tool use
- The ability of a model to call external tools such as search, code execution or business apps, instead of only generating text.