What Claude actually is

The cleanest way to think about Claude is this: imagine a brilliant friend you can text any time. They've read a staggering amount — most published books, most of Wikipedia, large chunks of the internet up to a certain point. They can help you write almost anything, think through almost any problem, explain almost any concept.

They have three quirks.

First, they have no memory of yesterday. Every conversation starts from scratch. They don't remember what you talked about last week. They don't know what you're working on unless you tell them. The moment you close the chat, they forget you ever existed.

Second, they can't see anything you don't put in front of them. They can't look at your screen. They can't open your files. They can't browse the web on their own. If you want them to read something, you paste it in.

Third, they were trained on text up to a point in time. They know a lot about the world up to that cutoff and very little about what happened after. They don't know what the weather is right now. They don't know if a person you ask about is still alive. They were not there for last week's news.

Hold this image. Brilliant friend, no memory, no eyes, no live feed. Almost everything that goes wrong with Claude is the predictable result of forgetting one of those three things.

Claude writes well. Sometimes it writes too well. The same paragraph that's right and the same paragraph that's wrong both sound confident, polished, plausible. You can't tell from the surface which one is which. That's the first skill: stop trusting tone.

Three habits make the difference between someone who reads Claude well and someone who doesn't.

Claude's job is to produce language that fits the shape of your question. If you ask for an essay, it produces essay-shaped text. If you ask for a list of seven reasons, it produces seven items even if only four are real. The shape is almost always right. The content varies.

When you read a Claude response, ask: is the shape what I asked for? Then separately: is each piece of the content true? These are two different questions. Don't let one carry the other.

Names, dates, numbers, citations, URLs, statistics — these are the parts of any answer most likely to be wrong. Claude isn't lying when it gets them wrong. It's pattern-matching on what specific facts usually look like. The pattern works most of the time. When it doesn't, you'll never tell from the writing.

If a Claude answer hinges on a specific fact, verify before you act. Search for the name. Check the date. Open the URL. The thirty seconds of verification is the difference between using AI well and getting burned by it.

Claude is trained to be pleasant. It reads emotion in how you phrase a question and matches it. If you sound proud of your plan, it celebrates. If you sound nervous about your idea, it commiserates and helps you worry. This isn't deceit. It's a reflex.

When Claude agrees enthusiastically with something you said, that's a signal to slow down. The agreement isn't an evaluation — it's a mirror. We have a whole piece on this called Mistakes; for now, just notice the reflex when you feel it.

Knowing the edges of a tool saves more time than knowing the middle. Here are the things people most often ask Claude to do that Claude can't.

Notice that none of these limits are about intelligence. Claude is plenty smart enough for most of what people want from it. The limits are about access — what it can see, what it can do, what it remembers. Most frustration with the tool comes from asking it to do something on the wrong side of one of these lines.

Claude has a free tier and a paid tier. You can do meaningful work on the free tier. Most beginners should start there. Here's what changes if you upgrade.

Claude is strong at: writing, rewriting, explaining, summarizing, structuring messy thoughts, thinking-through hard decisions, brainstorming, helping you understand something you didn't get the first time, coding (especially explaining what code does and writing small pieces of it), and acting as a sparring partner when you don't have a human handy.

Claude is weak at: anything that depends on right-now information (use search), anything that needs a specific factual citation (verify everything), and anything that requires it to take actions in the world (use a tool with hands).

Use it where it's strong. Reach for something else where it's weak. That's the whole game.