Agentic
Core Concept
Describes an AI that takes actions on its own rather than simply answering questions. When Claude operates agentically, you give it a goal and it works toward that goal — opening files, running searches, executing steps — without you directing each one. Claude Cowork and Claude Code are both agentic tools. Claude.ai chat is not.
Anthropic
Company
The company that built Claude. Founded by researchers who left OpenAI with the goal of building AI that was genuinely safe and useful. Anthropic's approach to safety shapes how Claude behaves — including its tendency to acknowledge uncertainty, push back on requests that seem off, and be honest about its limitations.
Claude
Core Concept
The AI itself. Made by Anthropic. Claude is the underlying model — the intelligence — behind all of Anthropic's products and interfaces. When people say "I'm using Claude," they mean they are interacting with this AI, regardless of which tool or interface they're using to access it.
Claude.ai (Chat)
Interface
The browser-based chat interface at claude.ai. You type a message, Claude responds, you reply. One exchange at a time, similar to texting. This is where most people start with Claude. It does not take autonomous actions on your computer or files — it answers, drafts, analyzes, and converses.
Claude Code
Interface · Developer Tool
A terminal-based and code-editor-based version of Claude built for software developers. Claude Code can read, write, edit, and navigate entire codebases autonomously. It understands project structure, runs commands, and makes changes across files. Non-developers rarely need this tool — it's purpose-built for software engineering workflows.
Claude Cowork
Interface
The desktop application that lets Claude operate autonomously across your computer. Cowork can look through your files, conduct research across multiple tabs, draft and save documents, and take actions in other apps. You give it a task and come back when it's done. This is where Claude moves from a conversation partner to an actual working tool. See also: Computer Use, Dispatch, Scheduled Tasks.
CLAUDE.md
Developer Feature
A special instruction file placed at the root of a software project. Claude Code reads it automatically at the start of every session. It works as persistent project memory — storing coding conventions, architectural decisions, context Claude needs to understand the codebase, and instructions that should apply every time. Without it, developers have to re-explain project rules every session.
Commands (Slash Commands)
Feature · Claude Code
Shortcut instructions typed directly into Claude Code with a forward slash. Common examples:
/help — shows available commands and features
/clear — clears the current conversation context
/init — initializes a new project and creates a CLAUDE.md file
Commands are Claude Code-specific. They do not work in Claude.ai chat or Claude Cowork.
Computer Use
Feature · Claude Cowork
A Claude Cowork capability that lets Claude see your screen and interact with it directly — clicking buttons, typing into fields, scrolling, navigating between apps. Claude operates software the same way a person would, using the visual interface. This is what makes complex, multi-app workflows possible without building custom integrations for every tool.
Context Window
Core Concept
Claude's working memory for a session. Everything in the current conversation — your messages, Claude's responses, uploaded files, shared documents — lives inside the context window. It is large (Claude's context window holds hundreds of thousands of tokens) but not infinite. As a very long conversation fills the window, the earliest parts may drop out. Starting a new session resets the window completely. This is not forgetting — it's a capacity limit, the same way a whiteboard has a finite amount of space.
Dispatch
Feature · Claude Cowork
A Claude Cowork feature that lets you send tasks to Claude from your phone. Instead of sitting at your computer to start a task, you send it via Dispatch — a mobile-friendly way to assign work to Claude Cowork. Claude picks it up and works on it. You come back to results.
MCP (Model Context Protocol)
Feature · Integration
A standard protocol that lets Claude connect to and communicate with external apps. MCPs are pre-built connectors for tools like Slack, Gmail, Google Drive, Notion, and others. Once an MCP is set up, Claude can pull data from those apps and take actions in them directly — without you having to copy, paste, or manually bridge information between tools. The protocol is the standard. The connected apps are the destinations. A single well-configured MCP setup can dramatically expand what Claude can do inside your actual workflow.
Plan Mode
Feature · Claude Code
A Claude Code safety feature that lets Claude propose what it intends to do before it does anything. You see the plan — which files it will change, what steps it will take — and you approve or adjust it before Claude executes. Plan Mode exists because agentic tools working across large codebases can make consequential changes quickly. The ability to review before committing reduces the risk of unintended modifications.
Scheduled Tasks
Feature · Claude Cowork
A Claude Cowork feature that lets you set tasks to run automatically on a recurring schedule. Instead of manually starting the same task every week, you schedule it once and Claude runs it on whatever cadence you set — daily, weekly, at a specific time. Useful for recurring research, monitoring, content drafting, or any repeatable workflow.
Skills
Feature
A saved set of instructions that teaches Claude how to do one specific thing well and consistently. Instead of explaining the same task from scratch every session, you build a skill once. Claude runs it on demand, following the instructions each time. Skills can encode anything: a writing style, a research process, a content format, a workflow. They function like trained assistants who already know the job before the session starts.
Tokens
Core Concept
The unit Claude uses to read and write. Not words — chunks of text that are roughly three quarters of a word each. Claude processes your input in tokens and generates its response in tokens. Token count determines two practical things: the size of the context window (measured in tokens, not words) and the cost of using Claude via the API (billed per token). Most users working directly in Claude.ai or Claude Cowork do not need to think about tokens day-to-day. Developers and anyone building AI-powered tools at scale will encounter them constantly.
This glossary covers the core terms from The Beginner's Glossary: Claude Cowork + Claude Code by The Prepared Performer (mollymahoney.com), reproduced and expanded with permission for use on tamistewartconsults.com.
Content adapted from The Prepared Performer / mollymahoney.com. Used with permission.