Anthropic ships new Claude features fast, and the naming doesn't always make it obvious what something actually does. Is a Skill different from a Plugin? Is a Routine just a Scheduled Task with a new name? What does a Subagent actually do that Cowork doesn't already do?
This is a plain-English reference for the terms you'll run into across Claude's apps and docs. No history lessons, just what each thing is and when you'd reach for it.
Ways to work with Claude
Chat
The basic conversation window. One question, one answer, back and forth. Where almost everyone starts, and plenty of work never needs to leave it.
Projects
A persistent workspace with its own memory. Attach reference files and custom instructions once, and every new chat inside that Project starts already knowing your context.
Cowork
An agentic mode built into the Claude desktop app. Hand it a multi-step task, point it at a folder, and it plans the work, runs it in an isolated environment, and hands you the finished result with minimal supervision.
Claude Code
Claude built for developers, running in your terminal or IDE. It reads and writes real files, runs commands, and can be extended with Skills, Plugins, and MCP servers.
Claude Design
An Anthropic Labs product for building visual work by talking, slides, prototypes, one-pagers, and more, on its own canvas with its own export options (PDF, PPTX, or straight into Canva).
What you get back
Artifacts
A piece of content Claude generates that lives in its own panel next to the chat, a document, a chart, a small tool, that you keep refining in place instead of scrolling back through chat history to find "the good version."
Extending what Claude can do
Skills
Instructions you write once (a folder with a SKILL.md file), and Claude follows them automatically whenever they're relevant, in every chat after that. Try the free SKILL.md Scaffolder to generate a starter file.
Plugins
A packaged bundle of Skills, MCP servers, and other components, installed as one unit. Think of it as a way to distribute a whole toolkit instead of setting up each piece by hand.
Connectors
Integrations with the apps you already use, Gmail, Drive, Slack, Notion, and others, that let Claude read and act on your real accounts instead of working from what you paste in.
MCP
Model Context Protocol, the open standard that Connectors and most tool integrations are actually built on. You'll rarely interact with MCP directly, but it's the plumbing underneath almost everything in this section.
Automation and multi-agent work
Scheduled Tasks
A prompt or workflow you set to run on a recurring schedule, a daily digest, a weekly report, run for you automatically and delivered when it's done.
Routines
Claude Code's version of scheduled automation: a prompt, repo, and set of connectors configured once and triggered on a schedule, an API call, or an event. Runs on Anthropic's infrastructure, so your laptop doesn't need to be open.
Dispatch
A Claude Cowork feature where your phone controls and your desktop executes, effectively a walkie-talkie to a Cowork session running on your computer, so you can kick off or check on work from your phone.
Subagents
Smaller Claude instances a main session can delegate side tasks to, each with its own context window and tools. Used to keep a "side quest" (research a library, verify a finding) out of your main conversation instead of cluttering it.
Context that carries over
Memory
Claude's ability to remember details across separate chats, your role, preferences, and ongoing projects, so you stop re-explaining yourself every time. Project-based, editable, and can be turned off entirely in settings.
How to actually use this: you don't need all fifteen of these on day one. Chat, Projects, and Artifacts cover most day-to-day work. Everything from Skills onward is for when you're ready to connect Claude to real tools or automate something that used to take you an hour every week.