Claude Chat, Cowork, and Code — which is which?
Chat for conversations, Cowork for file work, Code for developers — what each product does and when to use it.
When people hear "Claude," they think chatbot. That's accurate — but only partly. Behind the name are three distinct products built for fundamentally different situations. Using the wrong one means working inefficiently. Knowing the right one changes how you delegate work.
Claude Chat — for conversations and generation
Claude Chat is the web interface at claude.ai. Open a browser, ask something, get an answer. No download, no setup, no terminal. Available to everyone, including a free tier.
The core principle: you stay in the conversation. Claude responds, you react, something takes shape together. That makes Chat the right tool for anything that needs back-and-forth — working through texts, thinking out loud, researching, asking questions.
Two features lift Chat above a basic chatbot:
Projects — Persistent workspaces. Instead of re-explaining your context in every conversation, you set up a project once: custom instructions for Claude, uploaded documents as a knowledge base, everything stored centrally. Every conversation within the project starts with that knowledge. Available to all users; free accounts can create up to five projects.
Artifacts — When Claude creates something that stands on its own — code, an HTML page, a document, a diagram — it appears in a dedicated panel beside the chat. Visible immediately, editable directly, no copy-paste loop. Claude revises the artifact on request, and earlier versions are preserved.
Typical tasks:
- Writing, editing, translating texts
- Research and summarization
- Explaining code or developing smaller snippets
- Analyzing documents (upload up to 20 files per chat)
- Working through concepts iteratively
Limits: Claude Chat cannot access local files on your machine (upload only), cannot execute code, cannot handle multi-step tasks autonomously, and cannot create git commits.
Claude Cowork — for autonomous tasks without code
Cowork became available to all paid Claude plans in April 2026 and is the least-known of the three products — despite being the most relevant for many office workers.
The principle is a direct answer to Chat's limits: you don't need to stay in the conversation. You describe what you need — and Claude handles it. You come back when it's done.
What Cowork does differently:
Direct file access — No manual upload. Cowork reads and writes files directly on your machine. You point to where the files are; Claude takes it from there.
Multi-step autonomy — Claude completes several steps in sequence without requiring you to confirm each one. A task can have ten sub-steps — you wait for the result, not for every decision along the way.
Professional outputs — Cowork doesn't just produce text. It creates finished working documents: Excel files with formulas, PowerPoint presentations, structured reports. No manual post-processing required.
Scheduling — Tasks can be triggered on demand or set to run on a recurring basis: daily, weekly, whenever needed.
Isolated execution — Cowork runs in an isolated environment. What Claude executes there does not touch your main system.
Concrete example: You have sixty contract attachments as PDFs. You need an Excel table with all deadlines, counterparties, and amounts. In Chat, you'd upload file by file, wait for responses, and piece everything together manually. In Cowork, you describe the task once — Claude reads all the files directly, structures the data, and hands you a finished spreadsheet.
Typical users: Finance teams, legal, HR, operations, consultants — anyone who regularly processes large numbers of documents, extracts data from PDFs, or builds reports from raw material. No programming knowledge required.
Limits: Cowork is designed for file and document work, not software development. Writing code in a repository, running tests, or creating git commits is outside its scope.
Requirements: macOS and Windows only (Claude Desktop app). Paid plans only (Pro, Max, Team, Enterprise). The desktop app must remain open while a task is running.
Claude Code — for software development
Claude Code is a coding tool that integrates directly into the development environment. It runs in the terminal, as an extension for VS Code and JetBrains, as a desktop app, or in the browser — and can be reached via Slack. And it does something fundamentally different from Chat: it understands the entire codebase — and acts on it.
The core difference from Chat: in Chat you describe a problem and get code back that you copy and paste into place. In Claude Code you delegate a task — Claude reads the relevant files, plans the approach, makes all necessary changes, runs tests, fixes errors, and creates a commit at the end. You review the result, not every intermediate step.
What Claude Code actually does:
Codebase understanding — Claude Code doesn't read one file; it reads the whole project: architecture, dependencies, conventions. A change in file A that affects files B and C is planned together from the start.
Direct file editing — Claude Code edits files directly. No code block to copy, no manual insertion.
Test integration — Runs tests, interprets failures, fixes the code, runs again — until everything passes.
Git workflow — Creates commits with meaningful messages, opens pull requests, integrates into CI/CD pipelines (GitHub Actions, GitLab).
MCP integration — Via the Model Context Protocol, Claude Code connects to external tools: reading Jira tickets, sending Slack messages, pulling in Google Drive documents — so Claude can close the relevant ticket while fixing the bug, without you switching context.
Parallel sub-agents — Claude Code can run multiple tasks simultaneously in isolated git worktrees. One agent refactors the auth module, another writes tests for it, a third updates the documentation.
Routines — Recurring tasks on Anthropic infrastructure that keep running even when your machine is off: review new PRs each morning, audit dependencies weekly, update docs after every merge.
Concrete example: You say: "Migrate the auth module from session tokens to JWTs. All tests should pass afterwards. Open a PR." Claude Code reads the relevant files, plans the migration, makes all the changes, runs the tests, fixes what's broken — and returns a finished pull request.
Typical users: Developers and technical teams. Claude Code assumes you know what a terminal is and how git works.
Limits: Claude Code is optimised for software development, not document processing or conversation.
Requirements: Paid Claude plan (Pro minimum). Access to a terminal or IDE.
When to use which
| Situation | Tool |
|---|---|
| Quick question, writing something, thinking something through | Chat |
| Storing knowledge for a project permanently | Chat with Projects |
| Processing many files, building reports, extracting data | Cowork |
| Automating recurring document tasks | Cowork |
| Writing code, fixing bugs, running tests | Code |
| Creating commits, opening pull requests, CI integration | Code |
| Automating recurring coding tasks | Code |
The simplest rule of thumb:
- Is it a conversation or a question → Chat
- Is it file processing without code → Cowork
- Is it software → Code
The three products are not alternatives to each other — they cover different layers of work. Knowing all three means you choose deliberately.
