The Owner's AI Glossary: 20 AI Terms Explained in Plain English

You are twenty minutes into a demo call. The vendor says their "agentic platform leverages a fine tuned LLM with an extended context window", and you nod, because everyone nods. The ideas behind those words are simple, but jargon moves the power in the room toward whoever is selling.

This page moves it back. Here are the 20 AI terms owners hear most, explained in plain English with the practical takeaway underneath. Technical references like Google's machine learning glossary run to hundreds of entries; you need these 20, in language you can repeat on a call. Bookmark it.

What do AI terms like LLM, agent, and token actually mean?

In plain English: an LLM is software trained on vast amounts of text so it can write and answer questions, an AI agent is AI that takes actions in your tools instead of just chatting, and a token is the small chunk of text AI reads and bills by. Every other term is a variation on three questions: what the AI knows, how you talk to it, and what it may do without you.

If you only have one minute, start with these five:

If a word cannot survive translation into one plain sentence, it was doing sales work, not explanation work.The rule we apply on every audit call

The basics

AI (artificial intelligence)

AI is software that performs tasks which normally require human judgment: understanding language, recognizing patterns, making predictions, producing text or images. Instead of following explicit rules, modern AI learns patterns from millions of examples, which is why it handles messy input like a rambling email. It is a capability, not a single product.

Why it matters: when a product says "AI powered", ask which task the AI performs. The label alone tells you nothing.

Generative AI

Generative AI is the branch of AI that creates new content: text, images, audio, video, or code. Instead of only sorting or scoring existing data, it produces original material following the patterns it learned during training. ChatGPT drafting an email is generative AI at work.

Why it matters: for most businesses the value is drafts. Treat the output as a fast first version that a human polishes.

LLM (large language model)

An LLM is the engine behind tools like ChatGPT, Claude, and Gemini: a model trained on enormous amounts of text to predict the most likely next word. At scale, that simple trick produces software that can write, summarize, translate, and answer questions fluently. LLMs work with probability, not a database of verified facts.

Why it matters: most AI tools are a layer on top of a handful of LLMs. Judge the workflow, not the engine.

Model

A model is the trained result of the machine learning process: a very large file of numbers encoding everything the AI learned. Models differ in size, speed, cost, and skill, the way engines differ in horsepower. When a vendor says "our model", it often means "the model we rent from a large AI lab".

Why it matters: ask which model powers the tool, and what happens to your setup when it is upgraded or retired.

Chatbot

A chatbot is a conversation interface: you type, the AI replies, and nothing happens until you type again. A chatbot does no work between conversations. It is the front door to AI, not the whole house, a distinction we unpack in why chatting is not automation.

Why it matters: a chatbot saves you typing; it does not send the invoice or book the job.

Talking to AI

Prompt

A prompt is the instruction you give an AI: the question or request, plus any context you include. Output quality tracks prompt quality, because the model can only work with what you hand it. A good prompt reads like a good brief to a freelancer: the task, the audience, the tone, and one example.

Why it matters: you do not need a prompting course. You need the habit of briefing AI as clearly as you would brief a new hire.

Token

A token is the small chunk of text an AI model actually reads and writes, usually a word or a piece of one. In English, 1,000 tokens is roughly 750 words. Models are priced, limited, and measured in tokens, which is why AI bills look odd until you do that conversion.

Why it matters: a plan offering a million tokens means about 750,000 words of input and output combined.

Context window

The context window is how much text a model can consider at once: your prompt, the conversation so far, and any documents you attach. Think of it as working memory. Anything outside the window is invisible to the model, which is why a long chat can forget an instruction from an hour ago.

Why it matters: if the AI "forgot" something, it probably fell out of the window. Restate key facts, or use a tool with permanent brand memory.

Hallucination

A hallucination is a confident, fluent, wrong answer. Because LLMs predict plausible text rather than look up verified facts, they can invent a statistic, a citation, a price, or a policy that sounds real. It is a known property of the technology, not a rare glitch.

Why it matters: never let unreviewed AI output state facts, prices, or promises to a customer. Drafts first, approval always.

Brand memory (knowledge base)

Brand memory, often called a knowledge base, is the stored set of facts an AI consults before answering: your prices, services, policies, tone of voice, and standard answers. The AI checks your documents before relying on the general internet. This grounding is the main cure for generic answers and many hallucinations.

Why it matters: an AI without brand memory sounds like everyone else. Feeding it your real FAQs is the highest leverage hour you will spend.

AI that works for you

AI agent

An AI agent is AI that can take actions, not just produce answers. Given a goal, it can plan steps, use tools like your inbox, calendar, or customer database, check the results, and adjust. We explain agents in our plain language guide to AI employees.

Why it matters: agents turn AI from a writing aid into hours saved. Because they act for you, scope and approvals matter more than features.

AI employee

An AI employee is an AI agent given a defined role: a scope of work, access to specific tools, standards to follow, and a supervision routine. Think job description rather than gadget. The term is a business framing of agents, and vendors use it with varying honesty.

Why it matters: the useful question is not "is it really an employee" but "which tasks does it own, and who checks its work".

Automation

Automation is any system that completes a task without a human performing each step. Classic automation follows fixed rules: when an invoice is seven days late, send this exact email. AI adds judgment: read the reply, notice the customer is disputing the charge, and route it to you instead of nudging again.

Why it matters: many "AI" products are ordinary rules based automation with a new label. Still a fine buy, just not at AI prices.

Workflow

A workflow is the full path a task travels from trigger to finished result: a review arrives, a draft reply is written in your tone, you approve, it posts, the outcome is logged. Tools automate steps; workflows deliver outcomes. At Eva we map the workflow first and pick technology last.

Why it matters: buy outcomes, not features. Ask any vendor to demo one complete workflow before you talk price.

Integration

An integration is the connection that lets one piece of software work inside another: your AI reading the inbox, writing to the calendar, pulling numbers from your invoicing tool. Without integrations an AI can only talk about your work; with them it can do your work. Most setup effort lives here.

Why it matters: list the three tools your business runs on and confirm the AI connects to them today, not on a roadmap.

Words from the sales calls

These five terms appear most often when money is about to change hands. Regulators have noticed: the FTC has warned companies to keep their AI claims in check. Your job is simpler: make each word earn a plain sentence.

Credits

Credits are how many AI tools meter usage: each message, task, or generation costs a set number of credits, and your plan includes a monthly allowance. Sintra, for example, includes around 250 credits a month on entry plans, and real work burns them faster than the marketing suggests, leaving you to wait or upgrade.

Why it matters: budget on your actual monthly volume, not the sticker price. We do the math in our guide to what an AI employee costs.

API

An API, or application programming interface, is the doorway that lets one program use another program's abilities. When a tool "uses the OpenAI API", it sends your request to that company's model and returns the answer. API usage is typically billed per token: metered pricing, like electricity, rather than a flat subscription.

Why it matters: metered pricing can be far cheaper or far dearer than a flat plan. Ask what a typical customer's monthly bill looks like.

Human in the loop

Human in the loop means a person reviews or approves the AI's work before it takes effect: a drafted reply waits for your sign off, a refund needs your click. It is the standard safety pattern for anything customer facing, and the NIST AI Risk Management Framework treats human oversight as a core control.

Why it matters: ask exactly where the humans sit in the loop. "Fully autonomous from day one" is a red flag, not a feature.

Fine tuning

Fine tuning means training an existing model further on your own examples so it absorbs your style or specialty. It is genuinely useful in narrow, high volume cases, and genuinely expensive to maintain. For a typical small business, brand memory plus clear instructions delivers most of the same effect at a fraction of the cost.

Why it matters: if a vendor leads with fine tuning for a five person business, ask which problem a good knowledge base cannot solve.

AI audit

An AI audit is a structured review of your business to find where AI would actually pay off. A good one inventories your repeated tasks, measures the time they take, scores them for automation fit, and hands you a prioritized plan with honest costs. Our guide to where to start with AI walks through the method.

Why it matters: free audits vary wildly. Ask up front whether you keep the findings in writing, whether or not you buy.

Take this to your next sales call

A glossary is only useful if it changes what you ask. Before your next demo or renewal:

• Name the engine. Which model powers the tool, and what happens when it changes?
• Locate the humans. Where does approval sit for anything a customer sees?
• Price the usage. What does one typical task cost in credits or tokens?
• Demand one workflow. Watch one complete workflow with your tools, trigger to result.
• Test the translation. Pick any term from the pitch and ask for it in one sentence.

Twenty terms, three questions underneath: what the AI knows, how you talk to it, what it may do on its own. Keep those in your pocket and the vocabulary stops being a moat. When you choose a first project, start with the tasks worth delegating to AI agents, not the fanciest word in the pitch.

Frequently asked questions