You do not need a computer science degree to make sound decisions about AI. You do need to know what the words mean, because they get used loosely, and a vague word is an easy place for a weak product to hide.
Here are the terms worth knowing, drawn from our free guide AI in plain English, which goes further with a 20-term glossary and a safe-use checklist.
The words for how it works
Large language model (LLM). The technology behind tools such as ChatGPT, Claude, Gemini and Microsoft Copilot. It has learned patterns from huge amounts of text, and it produces writing by working out which word is likely to come next, then the next. This one fact explains a lot. A model on its own is built to produce plausible text rather than to check facts, which is why a reply can read beautifully and still be wrong.
Prompt. Whatever you type to instruct the tool. A good prompt states the task, gives the context and says what format you want. A vague one earns a vague reply.
Microsoft Copilot. Microsoft's assistant, which can draw on the files a person is already allowed to open. Copilot Chat comes with many Microsoft 365 plans at no extra charge, while some features need a separate Copilot licence. Because of how it finds content, permissions are worth reviewing first, and our SharePoint oversharing check shows how.
Chat, automation and agents
These three get mixed up all the time, yet the difference decides how much checking you need. The simplest test is to ask who decides the steps.
| Term | Who decides the steps | Everyday example | What keeps it safe |
|---|---|---|---|
| Chat | You, one question at a time | Asking for a first draft of a quote email | You read and correct the reply |
| Automation | You, in advance, and the steps then run by themselves | Logging each new web enquiry and sending the same acknowledgement | Testing the steps before they go live |
| Agent | The AI, inside limits you set | Drafting a reply to a new enquiry, then waiting for approval | Clear rules, narrow permissions and a person approving what goes out |
The more freedom a system has, the tighter your rules should be. If someone offers you an "agent", ask what it can do without asking first, which systems it can reach and how you can see what it did. Vague answers are a warning sign.
Where a job does call for one, APPS 365 builds custom AI agents, with the goal, tools, limits and approval steps set out in your written quote. Custom builds start from £999, with a fixed quote after a call, and monthly running costs apply.
The words for when it goes wrong
Hallucination. The word for an AI stating something false as though it were fact. The UK's National Cyber Security Centre uses it when warning about these models. It sounds exotic, but in practice it means an invented reference, a wrong date or a product feature you do not offer, all written in the same assured tone as everything else.
Three other faults are worth watching for:
- Wrong sums. Check any figure that matters, and rework it yourself if money depends on it.
- Stale knowledge. A model may not know about recent changes, so confirm anything that depends on dates, prices or rules.
- Agreeing with you. Ask whether your idea is good and it may simply say yes. Ask for the case against as well.
Draft first, a person checks. Not a technical term, but the most useful habit here. The safer everyday jobs follow this pattern: quote emails, replies to reviews, meeting actions and spreadsheet formulas. AI does the first pass, and a person who knows the subject decides what goes out.
The words for your data
Public AI tool. A tool open to anyone, often free. The company running it can see what you type, so do not put customer or staff details into one.
UK GDPR. The main data protection law here, and it carries on applying when you use AI. Share only what a task needs, and strip out names before pasting.
DPIA. A data protection impact assessment, which is a written look at the privacy risks of some kinds of data use before you begin. The ICO explains when you need one, and it lists AI among the innovative technologies that can count towards that. This is general guidance, not legal advice.
Three questions that cut through the hype
- What does it do on its own, and what needs a person?
- Where does our data go, and who can see it?
- How will we know when it has got something wrong?
A supplier who answers these plainly is worth talking to, whatever words are on their website. If you would like to test your own readiness next, try our 10-point AI-readiness check.
Read the full guide
For the complete explanation, including everyday uses by role and the full glossary, ask for the free AI in plain English guide, a 10-page PDF sent by email. If you have a particular job in mind and want a straight answer on whether AI suits it, book a free 30-minute call with APPS 365.