Three modes, and when the browser tab is the wrong place

You have been using Claude.ai in the browser. That is the right place to start. But once Claude becomes part of how you actually work, the browser tab starts to feel like a bottleneck — you switch windows to find a screenshot, you paste messy context, you lose the thread between meetings. The Claude desktop app solves that by giving you three different surfaces for three different kinds of work: Chat, Cowork, and Code.

This lesson is a quick tour of each mode, what it is for, and how to pick the right one.

The three modes at a glance

Chat is the familiar claude.ai experience, running natively on your machine. Ask questions, brainstorm, draft, iterate. Quick and conversational.

Cowork is an agentic mode. You give Claude a goal, connect it to your tools and files, and let it work independently — research, analysis, multi-source reports, end-to-end document creation. This is Claude with stamina.

Code is for software development. Claude works directly in your codebase — reading, writing, running commands, tracking changes with git. Aimed at developers, but worth knowing exists even if you do not write code.

Here is the unlock that ties it together. Cowork and Code run on the same engine — both are Claude Code underneath. That is what gives them their agentic qualities: the ability to spin up sub-agents, sustain long tasks, and work through big pieces of work without a human steering every step.

Chat — quick, conversational, always within reach

Chat is what you already know from the browser, plus a handful of things you only get from running on your machine.

• Quick entry. Double-tap the Option key on Mac to pull Claude up as a compact window over whatever you are working on. It stays on top while you switch between apps. You never have to leave what you are doing to ask a question.
• Screenshots and window sharing. Capture what is on your screen so Claude sees exactly what you see. Faster than describing a dashboard in words.
• Dictation. Talk through a problem instead of typing. Useful when you are thinking out loud or working away from the keyboard.
• Desktop connectors. Connect local tools so Claude can work with what is already on your machine — Apple Notes, Gmail, calendars, and more.

When to use Chat. You are staring at an unfamiliar chart and want to know what it means. You are between meetings and want to talk through a structure. You are drafting, exploring ideas, or learning something new by dialogue. Anything under ten minutes of back-and-forth belongs here.

Cowork — real work, start to finish

Cowork is built for the tasks that take real effort: pulling from many sources, making sense of the mess, and producing something finished. Research briefs, cross-source financial analysis, contract reviews, polished slide decks from material scattered across tools.

Before Cowork starts, it often asks a short set of questions to pin down scope, format, and constraints. Then it builds a plan you can review in the sidebar, and as it works you can watch the task come together — sources it is drawing from, files taking shape, progress through the plan.

A few features make Cowork feel qualitatively different from Chat:

• Folder access. Point Claude at a folder on your computer. It reads what is relevant, figures out the shape of your material, and saves finished work back to the same place.
• Scheduled tasks. Claude can run recurring work on a schedule — a morning inbox triage, a daily briefing from Slack and calendar, a weekly roundup of what shipped. You define it once; it runs every time.
• Sub-agents. For complex requests, Claude breaks the work into subtasks and hands each to a background worker with its own context, then coordinates the results into a single deliverable. You get one polished answer, not ten fragments.
• Browser use. Claude can navigate websites, interact with pages, and pull what it finds directly into its work. This is how Cowork checks competitor pricing across ten sites or gathers data from pages that do not have an API.
• Computer use (research preview). When there is no connector or plugin, Claude can drive your computer directly — clicking, typing, opening apps. There is a permission prompt before it touches each app, and you can blocklist anything off-limits.
• Protected environment. Cowork runs in a contained workspace on your machine. Claude can read and write within the folders you share, but cannot reach anything outside them.

When to use Cowork. You want to query your tools the way you would query a database. You are researching a new market across ten tabs and five PDFs. You have fifty documents to review and need the pattern that only emerges from reading them all. You keep doing the same morning setup and want to wake up to the finished version.

Cowork is available on Pro, Max, Team, and Enterprise plans. Think of it as the mode for work that takes time, but should not take your time.

Code — building software, with guardrails

The Code tab is Claude Code running inside the desktop app. It is a full development environment for building software: Claude reads your codebase, writes and modifies files, runs commands, and tracks every change through git so you can roll back whenever you want.

You choose where the work happens:

• Local. Claude works directly with files on your machine, can access local tools, and can run a development server you preview in your browser.
• Remote. Claude works in a cloud environment tied to a GitHub repository. Sessions keep running even when you close the app — useful for big refactors or larger codebases.

You also choose how much autonomy to grant:

• Ask. Claude proposes every change and waits for your approval. You see a visual diff before anything is modified.
• Code. Claude applies file changes automatically but checks before running terminal commands.
• Plan. Claude outlines its full approach before touching anything — good for complex work where you want to see the strategy first.

If you do not write code, you can skip Code for now. But know it exists — the same agentic engine that powers Cowork powers Code, which is why Claude can build, debug, and ship real software under your supervision.

Picking the right mode

A rough decision rule:

• Chat for quick questions, exploration, drafting, and anything under ten minutes of back-and-forth.
• Cowork for complex or sustained work that touches multiple sources and needs a finished deliverable.
• Code for software development — inside a codebase, running commands, shipping changes.

If you are about to paste a long brief and expect Claude to work on it for a while, that is a Cowork signal. If you catch yourself opening five tabs because Claude cannot reach the information, that is a connector or Cowork signal. If you are iterating on a sentence or exploring an idea, stay in Chat.

Before you move on

Think about the last three things you used Claude for. Pick one that would have been better in Cowork instead of Chat. That is the first real task you should try in the desktop app.