Claude Terms Decoded: What Beginners Actually Need to Know

You opened Claude. It asked what you needed. You typed something. It answered.

And then someone said "agentic" and you smiled and nodded and had no idea what they meant.

That's fine. Here's the decoder ring.

This is not an exhaustive tech glossary. It's the eight terms that actually matter when you're just getting started. The ones that will make the rest of the conversation make sense.

Claude is the AI itself. The brain. Made by a company called Anthropic (more on them next). When people say "I'm using Claude," they mean they're talking to this AI, the same way people say "I'm Googling something." Claude is the thing.

The company that built Claude. Founded by researchers who left OpenAI because they wanted to build AI that was genuinely safe and useful. They are serious about that. It shapes how Claude behaves — it's trained to be honest, to say when it doesn't know something, to push back when something seems off.

Worth knowing. Most people use Claude without knowing who made it.

This is where people get confused, so pay attention.

Claude.ai is the browser chat interface. You go to claude.ai, you type, it answers. One message at a time, like texting back and forth. This is where most people start.

Claude Cowork is the desktop app. This is where it gets interesting. Cowork can take actions — look through your files, do research, operate other programs on your computer. You give it a task. It works. You come back when it's done.

Claude Code is for developers. It lives in a terminal or code editor and can read, write, and modify code across entire projects. If you're not a developer, you probably don't need this one right now.

Same AI. Three very different tools. Knowing which one you're using (or which one you need) matters.

This word gets used a lot. It means Claude is taking action on its own, not just answering a question.

Think of the difference between asking someone what they think about a plan versus handing them the plan and saying "handle it." The first is a conversation. The second is agentic. Cowork and Code are both agentic tools. You give them a goal, and they work toward it without you managing every step.

Claude's working memory. Everything in your current conversation — your messages, Claude's responses, any files you shared — lives in the context window.

It's large but not infinite. If a conversation goes on long enough, earlier parts can drop out of that window. That doesn't mean Claude is getting dumber. The session has simply filled up its available memory. Start a new conversation and you're back to full capacity.

This is how Claude reads and writes. Not word by word — in small chunks called tokens. Roughly three quarters of a word each.

You probably won't need to think about tokens much unless you're running large-scale operations or managing costs. But when someone mentions "token limits" or "token costs," this is what they mean.

A skill is a set of instructions that teaches Claude how to do one specific thing really well. Instead of explaining the same task from scratch every time, you build a skill once and Claude runs it.

I use skills constantly in my own work — one for writing in my voice, one for building content calendars, one for SEO. They're like trained assistants who already know the job before they walk in the room.

MCPs let Claude talk directly to other apps. Slack. Gmail. Google Drive. Instead of copying and pasting between tools, Claude can pull information from and push actions into those apps — if you've connected them.

The protocol is the bridge. The connected apps are the destinations. Once you have a few MCPs set up, the way Claude can operate across your whole workflow changes pretty dramatically.

---

That's eight terms. That's the vocabulary you need to follow most conversations about Claude without feeling like you walked into the middle of someone else's meeting.

There are more terms worth knowing once you get further in, especially if you start working with Claude Code or building more advanced workflows. The full glossary, including developer-specific terms and command references, is on the resources page.