AI terminology for beginners: the business team glossary
TL;DR
AI has a language problem, not because the concepts are hard, but because nobody explains them plainly. This guide covers 30+ terms, from LLMs and tokens to agentic AI and MCP Servers. Once you know these words, you stop being intimidated by AI. You start being dangerous with it. The economics alone, £500, £2,000 a year in tokens versus a £25,000 employee doing the same work, should be enough to get your full attention.
What even is AI?
Artificial Intelligence is a machine doing something that normally requires human thinking. That is it. Not magic. Not a robot army. Just software that can understand language, spot patterns, and make decisions based on what you ask it. If you are using an AI tool, you are using artificial intelligence.
Automation is the next layer. Taking a task a human does and making software do it instead, not just speeding it up, replacing it. You define the rules, the system follows them, and a job that took an hour takes a second. Invoices, emails, customer data sorting: all candidates.
Business Process Reengineering (BPR) is the step before automation. You look at how you do something now, ask why you do it that way, and redesign it from scratch rather than making the old way faster. With AI, you can do things differently because AI removes constraints that never existed before. A team that re-engineers with AI in mind often gains 10 to 100 times more capacity than one that simply bolts AI onto the old process.
What is an LLM?
An LLM, Large Language Model, is software trained on millions of words and documents so it can predict what word comes next and understand context. It is not alive. It is not thinking. But it behaves as if it understands what you are asking, and that is what matters in practice.
LLMs are the engine under almost everything modern AI does: writing, summarising, translating, analysing, problem-solving. They are the workhorses.
Claude (made by Anthropic), ChatGPT (made by OpenAI), and Perplexity are all distinct AI systems, different sports cars on the same road. Claude is stronger at reasoning, safer with your data, and better at understanding what you actually want rather than just what you typed. ChatGPT is the most famous. Perplexity is built for research. They think differently because they were trained differently.
Manus is a different kind of tool, not an LLM itself, but a platform for real-time human-AI collaboration. Think of it as a whiteboard where you and Claude work together on thinking, designing, and solving problems. It is designed for strategy sessions and brainstorming where you need to think out loud with a partner that never gets tired or annoyed.
How does AI actually know things about your business?
Two concepts trip people up here.
Vector Database, do not let the word scare you. A vector is just a way to turn meaning into numbers. Your knowledge base, procedures, templates, FAQs, gets converted into vectors so the AI can find the right information instantly when you ask. Without a vector database, the AI guesses about your business. With one, it remembers exactly how you operate.
Training an LLM means teaching the AI about your business. You show it examples of how you write, how you think, and what good looks like. You are not changing how Claude works at a fundamental level. You are feeding it your knowledge so it sounds like you and follows your rules. A trained system sounds like your business. A raw system sounds generic and makes mistakes.
Bias through training is the risk. If you train the AI with bad data or poor examples, it learns the wrong thing. Show it 100 examples of how your business treats customers badly and it will think that is normal. Bad inputs equal bad outputs. Be deliberate about what you teach it.
What is prompt engineering and why does it matter?
A prompt is the question or instruction you give the AI. Context is everything you tell the AI to help it understand what you want, the background, the brief, the rules. Think of it like briefing a consultant before they start work. Tell them nothing, they guess. Tell them everything, they are brilliant.
Context is the difference between the AI being useless and the AI being 10 times better than a human.
Prompt engineering is the art of asking the right question in the right way. A vague prompt gets a vague answer. A clear, specific prompt gets a brilliant one. It is like the difference between asking a chef to cook something and asking them to cook something specific the way your grandmother did it. Good prompt writing multiplies the power of AI. Bad prompt writing wastes it.
Context engineering goes deeper: building the background knowledge the AI needs to give good answers consistently. You are telling it who you are, how you work, what matters to you, and what the rules are.
Intent engineering is the next layer: making sure the AI understands the goal, not just the task. You ask it to write an email, but what you really need is an email that closes a deal. Intent engineering bridges that gap between what you said and what you actually meant.
See where AI fits in your business. Free.
A 45-minute audit. We map the highest-value automations and what they're worth in time and money. No pitch, no pressure.
What do MCP Server, API, and webhook actually mean?
These are the plumbing words. You will hear them constantly.
MCP Server, a bridge between the AI and the tools you use every day: email, spreadsheets, calendars. With an MCP Server, Claude can reach into your systems and do work without you copying and pasting. The AI becomes part of your workflow, not a separate tool you move data to and from.
Integration, plugging AI into your existing systems so they work together. Your CRM, your accounting software, your email platform. A system that cannot talk to your other tools is useless. Integration is how AI becomes part of your actual business operations.
API (Application Programming Interface), the way two pieces of software talk to each other. If you want your email system to tell your CRM about a new contact, they talk through an API. It is the translator between systems. Without APIs, nothing integrates. With them, everything can work together.
Auth and OAuth, authorisation. Making sure the AI only accesses what it is allowed to access. OAuth is the modern secure way to say: this system can access that system, but only these parts. Think of it as a security pass at a building. You do not want the AI to have access to everything, auth makes sure it only touches what it needs.
Webhook, a way for one system to automatically tell another system something happened. A customer submits a form, a webhook tells the AI to start processing it immediately. Without webhooks, everything has to be scheduled or triggered manually. With them, AI responds instantly to what is happening in your business.
JSON, a way to format data so computers understand it. Structured like a filing cabinet: folders have documents, documents have fields. You do not need to understand it to use it. Just know it exists, it is how systems organise information so the AI does not get confused.
What are tokens and why should your CFO care?
A token is a chunk of text, not a word, a chunk. Sometimes a word, sometimes part of a word, sometimes punctuation. Think of tokens like Scrabble tiles. Every time the AI reads something and writes something, tokens are consumed. That is the currency of AI.
You pay for two things:
- Input tokens, what you send to the AI
- Output tokens, what the AI sends back
A typical conversation might use 500 input tokens and 300 output tokens. A long document might use 10,000 tokens to summarise, ten times more expensive.
Claude Opus costs approximately £0.015 per 1,000 input tokens and £0.045 per 1,000 output tokens. On the surface those numbers sound tiny. Scale them up and the picture changes.
A junior employee costs £25,000 a year. That is equivalent to roughly 1.7 billion tokens of Claude output. If you use AI smart, one person can do what used to take five.
For the first time in business, you can make a genuine choice: pay a token cost or pay a staff cost. Running an AI agent that does the same work as a £25,000 employee might cost £500 to £2,000 a year in tokens, a 10 to 50 times cheaper alternative. The CFO should track token spend the same way they track headcount.
If you run the same prompt inefficiently every day, you bleed money. Good prompt design and context engineering save tokens. Bad design wastes them.
Local models vs cloud models: what is the actual difference?
A local model runs on your own computer. No token cost per use. No ongoing cloud charges. But it is slower, weaker, and often cannot handle complex thinking. Good for simple tasks. Think of a calculator: fast, cheap, but only good for maths.
A cloud model, Claude, ChatGPT, runs on someone else's servers. Pay per token. Much smarter. Faster updates. Better reasoning. Think of hiring a consultant: you pay per hour, but you get brilliance.
The smart approach is hybrid: use local models for simple screening and filtering, cloud models for thinking work that requires real intelligence. Save tokens on repetitive tasks. Spend tokens on work that matters. Done right, this approach can cut AI costs by 40 to 60 per cent while actually improving results.
The trap is using the most expensive model for everything. It is like hiring a heart surgeon to pick up your post, overkill and wasteful. If you have 100 tasks a day and 60 of them could be handled by a cheaper model, you have just cut your token spend in half with the same output.
What problems can AI develop over time?
AI brain rot, what happens when you ask the AI the same questions over and over and stop thinking for yourself. The AI thinks for you and your own thinking gets worse, not better. It happens fast. AI should make you think better, not replace your thinking. Use it as a sparring partner.
AI prompt drift, over time, the way you ask the AI things changes subtly. You get lazy. You stop being specific. The AI starts giving sloppy answers. Quality drifts down because you have stopped telling it what you need. It is like a map that gradually gets more wrong. Regular review of your prompts and context is how you prevent it.
What is agentic AI and why is it the real shift?
Agentic AI does not just answer questions. It sees what needs to be done and does it. It can open your email, write a response, check your calendar, and send the email, all without asking permission each time. It acts on your behalf.
This is what doubles your capacity. Not just faster answers. Systems that work while you sleep.
Claude Code, OpenClaw, and NemoClaw are examples of agentic AI systems that can write code, build systems, and automate things without a human having to touch the keyboard. You do not need coders. You need people who understand the problem. The AI handles the code.
This is also where the Chief Agent Officer role emerges, a title you are going to see everywhere in the next few years. Not a coder. A strategist who designs, oversees, and maintains the AI agents doing the work. Not replaced by AI. Elevated by it.
Will AI take your job?
Let us be honest. Some work will be replaced. Not your job, the work. The best people in five years will not be the ones who avoided AI. They will be the ones who learned to work with it, design with it, and think with it.
Four things keep you relevant:
- Know the language. You do not need to be technical. You need to understand what is possible. That is what this guide is for.
- Think in processes. Start seeing your work as a series of steps. Which steps could AI do? Which require human judgement? The people who redesign their processes win.
- Own the judgement. AI can do the work. You provide the judgement. Is this the right decision? Does this fit our values? Is this good for our customers? Those questions still need humans.
- Build context. The best humans in an AI world understand the business deeply. Teach the AI about your industry, your customers, your rules. You become invaluable.
One Mac Mini per employee, a small computer running agentic AI. A person with an agent handling routine work, plus that person thinking, is 10 to 100 times more effective than a person doing routine work alone. The shift is not more employees. It is more agents.
What to do this week
Pick one task you do regularly that feels like busywork, something repetitive that takes focus but not creativity. Then:
- Write it out as a process. What are the exact steps, in order?
- Identify which steps require human judgement and which are purely mechanical.
- Test a prompt with Claude that handles one mechanical step.
- Note what context the AI needed to give you a good answer, that is your knowledge base starting to form.
- Track your token usage for that task so you have a baseline cost to optimise against.
Start tracking how many tokens you use on what. Start asking: could we do this more efficiently? Could we use a cheaper model? The best learning happens when AI solves a real problem for you. Find that problem this week.
Where to from here
Book a free 60-minute AI audit, we'll explore exactly what workflows are worth augmenting with AI.
Live with passion & AI,
Brett
Host a podcast? Have Brett on as a guest.
Straight talk on implementing AI in real SMEs, no jargon, plenty of receipts from the businesses we run.
Frequently asked questions
What is an LLM and how does it work?
+
An LLM, Large Language Model, is software trained on millions of words and documents that can predict what word comes next and understand context. It is not alive and it is not thinking, but it behaves as if it understands what you are asking. LLMs are the engine under almost everything modern AI does: writing, summarising, translating, analysing, and problem-solving.
What is a token in AI and why does it cost money?
+
A token is a chunk of text, sometimes a word, sometimes part of a word, sometimes punctuation. Every time AI reads something and writes something, tokens are consumed. You pay for both input tokens (what you send to the AI) and output tokens (what the AI sends back), which means every email written, every document summarised, and every analysis run has a real cost.
What is prompt engineering?
+
Prompt engineering is the art of asking the AI the right question in the right way. A vague prompt gets a vague answer; a clear, specific prompt gets a brilliant one. It is like the difference between asking a chef to cook something and asking them to cook something specific the way your grandmother did. Good prompt writing multiplies the power of AI, bad prompt writing wastes it.
What is agentic AI?
+
Agentic AI does not just answer questions, it sees what needs to be done and does it. It can open your email, write a response, check your calendar, and send the email, all without asking permission each time. It acts on your behalf, which is what makes it capable of doubling your effective capacity.
What is an MCP Server in plain English?
+
An MCP Server is a bridge between the AI and the tools you use every day, email, spreadsheets, calendars. It lets an AI like Claude reach into your systems and do work without you copying and pasting between applications, making the AI part of your actual workflow rather than a separate thing you feed data to.
What is the difference between local AI models and cloud AI models?
+
A local model runs on your own computer with no token cost per use, but it is slower and weaker, good for simple tasks. A cloud model like Claude or ChatGPT runs on remote servers, is much smarter, and costs money per token. The smart approach is hybrid: use local models for simple screening and filtering, cloud models for complex thinking work that requires intelligence.
What is AI brain rot and how do you avoid it?
+
AI brain rot is what happens when you ask the AI the same questions over and over and stop thinking for yourself, the AI thinks for you, and your own thinking gets worse. The fix is to use AI as a sparring partner, not a replacement. The best people in this space are the ones who think better because of AI, not instead of it.
Brett is a four-time founder (Darra Tyres, Gladfish, EzyTrac, Anaboo) and the operator behind AIOS, Anaboo's AI Operating System. He writes from inside the build, installing AI in his own businesses first and reporting back what actually moves the numbers. Based between Singapore, the UK and Australia.
