How do I get started with AI?

Start hereTen minutes · Checked Aug 2026

The basics, without the embarrassment.

Everyone around you talks as if they already know this. Most of them learned it last month.

What AI actually is

The AI everyone means right now is software that has read an enormous amount of writing and learned to predict what comes next. Ask it a question and it writes the most likely good answer, one word at a time. That is the whole trick, and it turns out to be a very useful trick.

It is strong at language jobs: summarising, drafting, rewriting, translating, sorting and finding the pattern in a pile of text. It knows nothing about your company, and it looks nothing up unless you or a tool tells it to.

It can be confidently wrong. It will write a plausible date, quote or figure that never existed. Treat everything it gives you the way you would treat a bright new hire’s first draft: useful, and checked.

Good at

  • Summarising a long document, thread or transcript.
  • A first draft of almost anything written.
  • Rewriting for a different reader: shorter, plainer, kinder.
  • Finding the themes in a pile of feedback or notes.
  • Explaining a term you would rather not ask about out loud.
  • Turning rough notes into a plan, an agenda or a checklist.

Not good at

  • Facts you did not give it. It will guess, and sound sure.
  • Anything inside your own systems. It cannot see them unless a tool connects it.
  • Arithmetic on a big spreadsheet, unless the tool says it runs calculations.
  • Events after its training ended, unless it searches the web.
  • Knowing what matters to your business. That part is still you.
  • Decisions that are yours to make. It can prepare one for you.

Which tool you already have

These four are more alike than different. Start with the one your company already pays for. That is the version IT has checked, and the one allowed to see your documents.

  • ChatGPT

    OpenAI

    The one most people mean when they say AI. Free to try, and many companies buy a team version.

  • Microsoft Copilot

    Microsoft

    Sits inside Word, Outlook, Excel and Teams if your company runs Microsoft 365. IT decides what it can see.

  • Gemini

    Google

    The same idea inside Gmail, Docs and Sheets for companies on Google Workspace.

  • Claude

    Anthropic

    Known for long documents and careful writing. Sold to individuals and to companies.

Not sure which one you have? Ask IT one question: “Which AI tool is approved for company documents?” That answer is worth more than any comparison chart.

Try one thing before lunch

Turn the thread you are dreading into a to-do list

10 min

Open whichever tool your company gave you. Paste the prompt below, then paste the email thread where it says so.

You are a sharp chief of staff. Below is an email thread. Give me four things: one paragraph on what it is actually about; the decisions already made; the questions still open; and what I personally need to do next, as a checklist. Use plain language. Do not invent anything that is not in the thread. If something is unclear, say so. Thread: [PASTE THE THREAD]

One check first. Read the answer against the thread. If anything in it is not in the thread, it was made up. On a personal account, take out names and anything confidential before you paste.

Five habits that keep you out of trouble

  1. Give it a role and a reader.

    “You are an HR editor. Write for staff reading on a phone.” One sentence like that changes the whole answer.

  2. Show it the material.

    It knows nothing about your business. Paste the policy, the thread, the notes, the numbers.

  3. Ask for a shape.

    Five bullets. A table. One paragraph. A checklist. You get the shape you name.

  4. Check every fact.

    Names, dates, numbers and quotes. It can invent all four while sounding certain.

  5. Keep secrets out of tools IT has not cleared.

    Customer data, pay, contracts, anything you promised to keep private. Ask which version your company has approved. The free one and the company one treat your data differently.

The words you keep hearing

One sentence each. If a vendor uses a word that is not here, ask them to say it again in plain English. That is a fair question and a good test of the vendor.

  • AI — software that learned from examples instead of being programmed step by step. On this site it means the tools on this page.

  • Generative AI — AI that makes new things: text, images, code, summaries. ChatGPT is the famous example.

  • Model — the engine inside the tool. GPT, Claude and Gemini are models. The app you type into is the car built around one.

  • Large language model, or LLM — a model that learned from an enormous amount of text. When someone says “the LLM”, they mean the engine behind the chat window.

  • Prompt — what you type. A good one names the job, the reader and the shape of the answer.

  • Chat window — the box you type into. ChatGPT, Copilot and Claude all start here. Useful, and only the beginning.

  • Hallucination — a confident answer that is simply wrong. A made-up quote, a date that never happened. The reason you check.

  • Copilot — Microsoft’s name for its AI helpers, and now a general word for AI that sits inside software you already use.

  • Agent — AI you give a job to rather than a question. It can open your email or calendar and act, with your permission.

  • Automation — a fixed set of steps that runs on its own. An agent can handle a step nobody wrote down. An automation cannot.

  • Specialised tool — software built for one job in one department: reading contracts, matching invoices, answering support tickets. The function pages on this site are full of them.

  • RAG — software that looks things up in your own documents before it answers. It is how a tool learns your policies without being rebuilt.

  • Training data — everything the model read while it was being built. If a fact is missing from both that and your prompt, it is guessing.

  • Guardrails — rules that stop a tool doing what you did not ask for. Approvals, spending limits, topics it may not touch.

  • Human in the loop — a person checks before anything final happens. The phrase to use when a vendor says “fully automated”.

  • Pilot — a small trial with one team, one measure and an end date. The right size for a first project.

  • API — a plug that lets two pieces of software talk. When a vendor says “we have an API”, they mean it can connect to your systems.

Where to go next