AI terms in plain words
- Free
- Under an hour
- No pitch
By Karim Nassar.
These are general definitions, written for owners who run a business and do not want a lecture. Vendors name the same ideas differently, so where it helps I list the other names.
AI agent
An AI that does more than answer. You give it a goal, and it picks steps, uses tools such as your calendar or inbox, checks its own results and keeps going until the job is done or it needs you. A chatbot waits for each message; an agent works through a task.
Chatbot
A program you talk to in a chat window. It answers the message in front of it. A chatbot can be simple (fixed replies) or powered by a language model, but it does not go off and do multi-step work on its own. That is the difference from an agent.
Custom GPT (or custom assistant)
A saved set of instructions, reference files and settings that turns a general AI assistant into one aimed at a single job, such as drafting your quotes in your voice. Several vendors offer the same idea under different names. It changes how the assistant behaves; it does not retrain the underlying model.
Hallucination
When an AI states something false with total confidence: an invented fact, a made-up quote, a link that does not exist. It happens because the model predicts likely words, not verified facts. Check anything with a number, a name, a date or a legal or medical claim before it leaves your business.
MCP (Model Context Protocol)
An open standard for connecting an AI assistant to other software, such as your files, calendar or accounting tool, in a consistent way. Think of it as a shared plug shape: a tool that speaks MCP can be reached by any assistant that does. It matters to you mainly because it lets an assistant read from and act in your real systems.
Context window
The amount of text an AI can hold in view at once: your messages, any files you gave it and its own replies. When a conversation outgrows the window, the oldest parts fall out and the assistant stops being able to use them. Long documents and long chats are where you notice it.
RAG (retrieval-augmented generation)
A setup where the AI first looks up relevant passages in your own documents, then writes its answer using what it found. It lets an assistant answer from your price list or handbook without retraining a model, and it can point to the source it used.
Automation vs AI
Automation follows fixed rules you set: when a form is submitted, add a row and send an email. It is predictable and cheap to run. AI handles work that needs judgment about messy input, such as reading an email and deciding what it is asking. Many good setups use both: rules for the steps that never change, AI for the step that needs reading.
Large language model (LLM)
The engine behind assistants such as ChatGPT, Claude, Gemini and Copilot. It is trained on large amounts of text to predict what comes next, which is why it can write, summarize and answer questions. It is strong with language and weak at guaranteed accuracy.
Prompt
What you type or paste into an AI to tell it what you want. A good prompt says the job, who it is for, what to use and what the result should look like. Saving prompts that work is the cheapest way to get repeatable results.
Training data
The text, images and other material a model learned from before you ever used it. A separate question for your business is whether a vendor uses what you type in to train future models. That depends on the plan and its settings, so check the vendor's current terms before pasting in client information.
Fine-tuning
Further training of an existing model on your own examples so it behaves a certain way. It is rarely the first thing a small business needs. Clear instructions, good reference documents and a custom assistant usually get you most of the way for far less work.
Token
The small chunk of text, roughly a short word or part of one, that models read and write. Vendors measure usage and the context window in tokens. A page of text is a few hundred tokens.
Workflow
The series of steps a piece of work goes through from start to finish, such as a new inquiry becoming a booked job. Mapping your workflows is how I find which steps are worth handing to automation or AI and which are not.
Human in the loop
A setup where a person reviews or approves the AI's output before it goes out or takes effect. For anything customer-facing, money-related or hard to undo, this is the default I recommend.
Generative AI
AI that produces new content, such as text, images, audio or code, instead of only sorting or scoring existing data. ChatGPT, Claude, Gemini and Copilot are generative AI assistants.
Find the first job to hand to AI
In a free call we go through your week together. You leave knowing which job to hand to AI first.
- Free
- 45 minutes
- No pitch