AI terms, in plain English
The words you will meet when you deploy AI in a business, explained without the jargon.
- Agentic AI
- AI that carries out tasks rather than only answering questions: it plans steps, uses tools and systems, and checks its own work, within limits people set.
- AI agent
- A piece of software, driven by an AI model, that is given a role, tools and permissions so it can do a defined job, such as drafting invoices or resolving tickets.
- AI team
- A group of agents, each in its own role and department, working together and handing work to each other, with people approving what matters.
- Agent harness
- The software that runs agents: it gives them their instructions, tools, memory and schedule. OpenClaw is an example of an open-source agent harness.
- Autonomy level
- How much an agent may do on its own: observe only, draft for approval, follow a fixed template, or act alone. Autonomy is raised one task at a time.
- Context window
- How much text an AI model can take into account at once. Larger windows let a model work with longer documents and histories.
- Data classification
- Labelling data by sensitivity (for example public, internal, confidential, restricted) so rules can decide which models and people may see it.
- Embedding
- A list of numbers that represents the meaning of a piece of text, so similar ideas can be found even when the words are different.
- Fine-tuning
- Further training a model on your own examples so it behaves in a particular way. Often unnecessary when a model can use your company knowledge directly.
- Frontier model
- The most capable AI models available at a given time, such as the latest models from Anthropic (Claude) and OpenAI, usually used through a cloud API.
- Guardrails
- The limits placed around AI: what it may access, what it may do, what needs approval, and what is logged.
- Hallucination
- When an AI model states something that sounds right but is not true. Grounding answers in your own data and checking sources reduces it.
- Human in the loop
- A design where a person reviews or approves an AI’s work before it takes effect, used for anything high risk.
- Integration broker
- A gateway between AI agents and business systems that holds the real credentials, gives agents short-lived permissions and logs every call.
- Knowledge graph
- A store of facts as connected things and relationships (who works with whom, what depends on what), useful for questions a search alone cannot answer.
- Large language model (LLM)
- The kind of AI model behind modern chat assistants and agents, trained on large amounts of text to understand and generate language.
- Least privilege
- Giving each person or agent only the access its role needs, and nothing more.
- Local model
- An AI model that runs on your own hardware, so the data it works on never has to leave your building.
- MCP (Model Context Protocol)
- An open standard for connecting AI models and agents to tools and data sources in a consistent way.
- Private AI
- AI that runs on infrastructure you control, with your data, knowledge and agents staying in your own environment.
- Prompt injection
- An attack where hidden instructions in a document, email or web page try to make an AI do something it should not. Limited permissions and approvals contain it.
- RAG (retrieval-augmented generation)
- A method where the AI first looks up relevant information from your own knowledge, then answers using it, so answers are grounded in your business.
- 2nd brain
- A company memory that brings together your documents, records and relationships so people and agents can use the whole of your business knowledge.
- Token
- The unit AI models read and write in, roughly three quarters of a word. Cloud AI is usually priced per token.
- Vector database
- A database that stores embeddings so meaning-based search is fast, the backbone of searching company knowledge with AI.
Questions about what any of this means for you?
Book a short conversation and we will explain what applies to your business, in plain English.