AI Language
Landscape.
This index exists because people get left behind not because the concepts are hard, but because nobody stopped to explain what they actually mean.
Every term in here is defined the way a real person would explain it over coffee. The fundamentals are simple, the rest builds on them. Skip ahead if you already know a section, or come back when something doesn't make sense.
The Fundamentals.
Three terms that show up everywhere. Get these and the rest gets easier.
Token
The smallest unit a LLM uses to process text. Not quite a word, not quite a character, somewhere in between.
One token is about three-quarters of a word in English. The word "fantastic" is one token. The word "uncharacteristically" might be three.
Claude doesn't read your message the way you do. It processes everything as a stream of tokens. Every word you type, every word Claude generates, and every piece of context loaded into a session (CLAUDE.md, skill files, connector data) is counted in tokens.
When the session runs out of space, older content gets pushed out to make room for new content. This is known as the context window. Tokens also determine cost when using the API.
Context Window
The total amount of text Claude can "hold in mind" at once during a single session. Think of it like a whiteboard.
Every message you send, every response Claude gives, every file it reads, and every skill it loads gets written on that whiteboard. When the whiteboard fills up, older content gets erased from the top to make room.
This is why very long sessions can start to feel like Claude is "forgetting" things, it hasn't forgotten them exactly, it's just that those messages have scrolled off the whiteboard.
Current Claude models have large context windows of 200,000 tokens (roughly 150,000 words), but they do have limits. This is why the workflow guides recommend breaking large projects into chunks and keeping a CLAUDE.md handoff file, so that when a new session starts, Claude can reload the essential context from a file rather than relying on conversation history.
Markdown .md
A plain text file that uses simple symbols to add formatting. The .md extension just means "markdown."
No special software required, any text editor (TextEdit on Mac will do) can open and edit a .md file.
Why it matters in Cowork: CLAUDE.md (your project memory file), SKILL.md (your skill instruction files), and most reference documents in a project are markdown files. They're lightweight, universally readable, and Claude can read and write them directly.
# Heading 1 ## Heading 2 **bold text** *italic* - bullet point - another point
• another point
The Claude Collection.
The suite from Anthropic you'll be using. Same model under the hood, very different surfaces.
Claude Chat
The LLM interface at claude.ai. You type a message, Claude responds. It's a conversation.
Powerful for thinking through problems, drafting content, answering questions, summarising documents, and general-purpose AI assistance.
What it doesn't have: persistent projects, connected tools, skills, or the ability to read and write your actual files. Every conversation starts fresh. You can upload files for a session, but Claude can't reach into your Notion workspace or create tasks in ClickUp on your behalf.
Claude Cowork
The desktop app built on top of Claude. It adds a layer of infrastructure that turns Claude into a co-working assistant.
The same Claude model runs under the hood. The difference is that Cowork can now actually DO things with your tools and files rather than just talking about them.
Claude Code
A version of Claude designed specifically for software developers. You run it in your terminal (CLI) and it can read your codebase, write code, run tests, fix bugs, create files, and execute commands.
Claude Code is for people who build software. It's deeply integrated with the development workflow: it understands version control, file structure, code dependencies, and can make changes across multiple files in a project simultaneously.
If you're not a developer, you probably won't use Claude Code directly. If you are a developer, it's the fastest way to use Claude for coding work.
Inside a Project.
Once you're in Claude Cowork, this is the landscape of what's actually in the room with you. Files, instruction sets, integrations, the lot.
Project
In Claude Cowork, a project is a connected context, a folder on your computer linked to a Cowork session, along with any skills and connectors associated with it.
When you open a project, Claude automatically loads the CLAUDE.md file from that folder (if it exists), which gives it the context of the work. Projects remember their settings across sessions. You can have multiple projects for different areas of your work and switch between them.
Think of a project as a workspace with memory. Everything relevant to that work lives in one place, and Claude picks up where you left off.
CLAUDE.md
The memory file of a project. A markdown file that lives in the root of your project folder and tells Claude the context of the project every time a new session starts.
What goes in it: project name and purpose, connected tools, installed skills and when to use them, naming conventions, current status, standing rules, and a running handoff note from the last session.
CLAUDE.md is loaded automatically at the start of every session. It's the difference between a Claude that asks "what are we working on today?" and a Claude that already knows.
Skill
A pre-built instruction set written in a markdown file (SKILL.md) that tells Claude how to handle a specific type of task. Skills load automatically when their trigger conditions are matched.
Examples: a formatting skill that enforces consistent heading styles in Notion, a workspace structure skill that gives Claude the exact folder hierarchy of your ClickUp space, a brand guidelines skill that keeps every piece of content on-brand.
Skills are the knowledge layer of a Cowork setup. Connectors give Claude access to tools, skills give Claude the context to use that access correctly.
skill-name/
├── SKILL.md
└── assets/
└── asset-name.md
The / Command
Inside Claude Cowork, typing / followed by a skill name manually triggers that skill. This is called invoking a skill with a slash command.
Example: typing /clickup-structure loads the ClickUp structure skill immediately, even if Claude wouldn't have triggered it automatically.
Most of the time you don't need to do this, skills load on their own when relevant. But if Claude isn't picking up a skill you know should apply, or if you want to be explicit about which skill to use, the /skill-name command forces it. The same / pattern is used for built-in commands (like /clear or /help).
Connector
An installed integration that gives Claude direct read and write access to an external tool or service. Connectors are built on MCP and are installed through the Cowork settings or plugin marketplace.
A Notion connector lets Claude create pages, update database properties, and search your workspace. A ClickUp connector lets Claude create tasks and filter lists. A Shopify connector lets Claude read product inventory and manage collections. With a connector: Claude reads and writes to the tool directly.
MCP
The technical standard that makes connectors work. MCP is an open protocol that defines how AI assistants like Claude connect to external tools and services.
This universal standard lets Claude plug into any tool that supports it. Developers build MCP-compatible connectors for their apps, Claude uses those connectors to talk to the apps.
As a Cowork user, you interact with MCP through connectors. You don't need to understand the protocol itself, just know that when someone says "MCP connector," they mean a live integration between Claude and an external app.
Artifact
A self-contained piece of output that Claude creates and saves, something designed to be used, not just read. In Cowork, an artifact is typically a live .html page, a dashboard, or an interactive widget that Claude builds and hosts persistently.
Unlike a regular response that disappears when the chat ends, an artifact persists. You can open it again, share it, and it will pull fresh data from your connected tools each time it loads.
Examples: a project status dashboard that shows live task data from ClickUp, a weekly metrics page that updates from your analytics tool, a brand asset viewer that displays your style guide. Artifacts are built with HTML, CSS, and JavaScript, but you don't need to write any of it. You describe what you want and Claude builds it.
Asset
A file or resource attached to a project or skill. Assets are supporting materials, reference documents, images, data files, tables, or specifications that are too detailed to embed directly in SKILL.md or CLAUDE.md.
In a skill, assets live in an assets/ subfolder alongside the SKILL.md file. The SKILL.md points to them, it doesn't duplicate the content. In a project folder, assets might live in an assets/ directory containing images, exports, and reference documents.
The principle: keep assets separate, keep them referenced, keep them updated independently of the main file they support.
Dispatch
A feature that lets you remote-control Claude Cowork (Desktop) from your mobile. Dispatch creates a single, persistent conversation thread that syncs between your phone and your computer.
You send Claude a task from mobile, Claude works on it using your desktop's full setup (local files, connectors, plugins, computer use), and pushes a notification to your phone when it's done. The key idea: you don't have to be at your desk.
Going Deeper.
You don't need these to use Cowork day-to-day, but you'll see them on docs pages, in tutorials, and in conversations with developers. Worth recognising.
API
The way two pieces of software talk to each other.
When you log into Spotify on your phone, your phone's app is talking to Spotify's servers via their API. When Claude connects to Notion through a connector, it's using Notion's API. When a developer builds an app powered by Claude, they're using Anthropic's API.
Plain English: an API is the door between two systems. It defines what requests can be made, what format to make them in, and what gets sent back.
SDK
A package of tools, code libraries, and documentation that developers use to build something on top of an existing platform.
Analogy: if an API is the door between systems, an SDK is a toolkit that makes it easier to build things using that door. Instead of writing all the connection code from scratch, a developer uses the SDK and a lot of the hard work is already done.
Anthropic has a Claude SDK that developers use to build applications powered by Claude, products like Claude Cowork are built using it.
CLI
A way of interacting with a computer by typing text commands into a terminal window, rather than clicking through a visual interface.
Instead of clicking File → New → Folder, a CLI user types mkdir my-folder and presses Enter. Same result, different method.
Claude Code is a CLI tool, you use it from the terminal. Most non-developers never touch a CLI, but developers live in one. If someone tells you to "run a command in your terminal" or "use the CLI," they mean: open your terminal app and type something.
Quick Reference.
What You'll Need.
Anthropic
A Pro or Max subscription unlocks the full Cowork desktop app, projects, skills, connectors, and Dispatch. The free tier of Claude Chat works for the basics, but the workflows in this index assume Pro at minimum.
Want the next one?
New guides posted weekly. The settings, prompts, and workflows I actually use.
Follow on Instagram →