What Claude AI is, in plain English

A plain explanation of what Claude is, who makes it, how it differs from a search engine, and the one mental model that saves people a fortnight of confusion.

A board update mentions the licences, everybody nods, and the agenda moves on. Nobody asks what Claude AI is, because by the summer of 2026 it feels late to. So, plainly: Claude is an AI assistant built by a company called Anthropic. You give it a request in ordinary language and it writes back, whether that is a draft, a summary, an analysis or a plan. That is most of it.

Anthropic publish their own documentation on what the models do and where they fall short, and they revise it constantly. That is the useful thing to know about them, and it is nearly the only one. When a client asks me whether Claude can handle something, I go and look there instead of at somebody's summary of it, and I would suggest the same habit.

Two different machines One question a search engine finds pages that already exist you read them Retrieval Claude composes an answer that did not exist before you check it Composition Not optional A search result arrives with its source attached. A composed answer arrives without one.
The difference is provenance, not quality. A page you were pointed at can be judged by where it came from. A paragraph written for you on the spot has no such history, so the last step on the lower route is the one you do yourself.

It composes, it does not retrieve

The most expensive misunderstanding I meet is treating Claude as a cleverer search engine. A search engine retrieves. It finds a page that already exists somewhere and points you at it. Claude composes. It produces new text on the spot, shaped by what you asked for, and none of it was sitting in a drawer waiting to be found.

This is why the output feels written for you, and also why it needs reading properly. A search result carries its source with it, so you can judge it. A composed answer arrives in the same confident prose whether it is right or badly wrong. Understanding that one distinction changes how people work more than any prompt trick I could teach in an afternoon.

The layer underneath this one, the way it produces an answer at all, accounts for most of the strange behaviour you are going to notice. If you would rather start at the practical end, the work it handles well is the shorter route.

The mental model worth keeping

Picture a very well-read colleague on their first morning. They have read an extraordinary amount, they write fluently about almost anything, and they have never met your organisation. They cannot look anything up unless you put it in front of them, and by tomorrow they will have forgotten this conversation entirely.

Nearly every disappointment I see traces back to skipping the part where you put it in front of them. Someone asks for a summary of a client relationship without pasting in a single email, gets something generic back, and concludes the tool is overrated. The colleague was willing. Nobody gave them the file.

Where you will meet it

There is a website, desktop apps for Mac and Windows, and apps for iPhone and Android, plus a browser extension called Claude for Chrome and an integration with Microsoft 365 if your organisation lives in Office. Claude Cowork, a desktop app since January 2026 and on web and mobile since July, exists so you can hand over a longer task and collect the result later, which suits report compilation and reconciliation. Claude Code exists too. It is for developers and you can ignore it.

A tool with a grain

Claude has a grain, in the way timber does. Work along it and the thing goes quickly. Work across it and you get a rough edge and blame the material. The people who get the most out of it are rarely the most technical people in the building; they are the ones who spent a few weeks paying attention to how it responds.

So give it three weeks of ordinary work before you decide what it is worth to you. The first afternoon tells you nothing, and most of the people I meet who have written it off had only ever had the first afternoon.

Sources

  1. Anthropic — Claude product overview
  2. TechCrunch — The coding agent wars are spilling into the rest of the office: Claude Cowork