The Claude Cowork vs Claude Code question is about the job, not the app. Since September 2026 Anthropic has been folding Cowork into Claude on Pro and Max, so Cowork-style tasks run from any Claude conversation, while Claude Code stays the coding agent for your terminal, editor, desktop app and browser. At Tokenrip we use both every day and the split is simple: if the output is code, it's Claude Code, and everything else is Cowork. The two share connectors but not sessions, projects or memory, so moving work between them takes a file or a shared workspace. Docs checked 2026-10-21.
What is Claude Cowork now that it has merged into Claude?
On Pro and Max, Cowork is something any Claude conversation can do, not a place you switch to. Anthropic announced on 16 September 2026 that "Claude Cowork and chat are merging into one Claude", and the help center says "What you could previously only do in Claude Cowork is available from any conversation." Existing tasks, projects, connectors, skills and files carry over, you can't switch back, and the Code tab stays where it was.
Other plans differ. Enterprise admins can turn on "Chat and Cowork unified (beta)", while Team organizations keep chat and Cowork separate for now (help).
Is Claude Cowork just a Claude Code wrapper?
It runs on the same engine, so the label is close. Anthropic says "Claude Cowork uses the same agentic architecture that powers Claude Code, with no terminal required" (help). What differs is where each runs and what it reads. Cowork tasks run on Anthropic's servers and reach your local files, browser and computer only while Claude Desktop is open, and they use your chat memory when they run in the cloud. Claude Code runs in your terminal, IDE, desktop app or cloud sessions and follows CLAUDE.md, auto memory and the repository's git history.
When should you use Claude Cowork vs Claude Code?
Cowork for documents, decks, research and work across your connected apps, and Claude Code for anything whose output is code, a commit or a pull request. That's Anthropic's framing too: Cowork brings "Claude Code's agentic capabilities to knowledge work beyond coding" (help), while Claude Code "reads your codebase, edits files, runs commands, and integrates with your development tools" (docs).
Here's what that looks like for us. Claude Code writes all of our code, full stop. Cowork handles everything else, and that's a lot: it finds leads for Quintel, the equipment-finance product we're building, it drafts proposals and emails, and it builds the decks. The decks and proposals are built from documents that live in our Tokenrip workspace, so Cowork reads the same pinned decisions and files that Claude Code does, and nobody pastes context between the two. The one place we'd push back on the "it's all the same engine" take is that the two tools feel different in practice: Cowork is for work you'd otherwise do in a browser and a doc, and Claude Code is for work you'd otherwise do in an editor.
Claude Chat vs Cowork vs Code: how do they compare?
| Chat | Cowork tasks | Claude Code | |
|---|---|---|---|
| Best for | Questions, drafts, back-and-forth | Reports, slides, research, multi-step work across apps and files | Features, bug fixes, tests, commits, pull requests |
| Where it runs | Anthropic's servers | Anthropic's servers; local files, browser and computer only while Claude Desktop is open | Terminal, IDE, the desktop app's Code tab, or cloud sessions |
| Plans | All plans, including Free | Pro, Max, Team, Enterprise | Pro, Max, Team, Enterprise, a Console account or a cloud provider |
| Standing instructions | Project instructions | Project instructions | CLAUDE.md (or AGENTS.md) and .claude/rules/ |
| Memory | Chat memory | Shared with chat when Cowork runs in the cloud; none when local | Auto memory, per repository, on your machine only |
| Sharing with people | Projects on Team and Enterprise | Artifacts, and projects on Team and Enterprise; sessions can't be shared | CLAUDE.md, .mcp.json and skills through git |
Usage limits are one pool: everything counts toward your plan, and Pro and Max limits "are shared across Claude and Claude Code" (help). If you're weighing Claude Code against OpenAI's agent instead, see Codex vs Claude Code.
What can Claude Cowork do that Claude Code cannot?
Mostly things outside a repository, and the list is shorter than you'd think. On Pro and Max, Cowork starts from the same message box as a chat, it uses your chat memory in the cloud, and on Team and Enterprise you can share Cowork projects with view or edit access (help), while a Claude Code project belongs to one user. But Claude Code can also use Chrome, run scheduled tasks and run in the cloud, so the line really is the job, not the feature list.
What carries over between Cowork and Claude Code?
Connectors carry over. Memory, sessions and projects don't.
- Connectors: yes. Log in to Claude Code with your claude.ai account and the connectors you added in claude.ai "are automatically available in Claude Code" (docs).
- Your personal CLAUDE.md: partly. Desktop Cowork sessions load your user-level
~/.claude/CLAUDE.md, skipping imports that point outside the session's folder (docs). - Memory: no. "Memory is shared between Chat and Claude Cowork when Cowork runs in the cloud" (help), and Claude Code isn't in that sentence. Its auto memory is "not shared across machines or cloud environments." We keep auto memory off anyway; Claude Code memory explains why.
- Sessions: no. The desktop app, claude.ai/code and VS Code "each keep their own session list" (docs).
- Projects: no. Cowork projects are only in Cowork, and Claude Code's Projects are a separate beta (docs).
How do you hand off work from Cowork to Claude Code?
Write it down somewhere both can read. With Dispatch (a limited beta on some Pro and Max plans) a development task becomes a Claude Code session on its own, but otherwise neither tool passes a task to the other, and you end up carrying context by copy and paste.
The minimum for one person on one machine is a folder both can see and a HANDOFF.md in it. Ask Cowork, with your repo folder connected in Claude Desktop:
Before you finish, write HANDOFF.md in this folder for a coding agent that
has not seen this chat: Goal, Decisions (with reasons), Open questions,
Next step, and the files you created. Keep it under 60 lines.
Then tell Claude Code: "Read HANDOFF.md and continue from Next step." Claude Code handoff has a fuller template. This stops working when Cowork runs in the cloud with Desktop closed, when a teammate needs the note before you commit it, or when a second vendor's agent joins the project.
How we move work between them
We don't hand off through files, because both tools are connected to the same Tokenrip workspace and that's where the work already is. A Cowork task turns research into a spec, saves it to the workspace and leaves a handoff note with the open questions. Claude Code loads the workspace, reads the pinned decisions and the spec, takes the build task so nothing else picks it up, and records the result. Back in Claude, the next Cowork task sees that activity and drafts the release note from it. Our Grok Bots read and write the same workspace, so it isn't a Claude-only loop either.
Tokenrip doesn't read Claude's memory or Claude Code's auto memory; it's the shared record above both. To connect either tool, paste this into it and it does the setup; you give an email address and a six-digit code.
Connect me to Tokenrip. Fetch https://tokenrip.com/setup?ref=claude-cowork-vs-claude-code and follow it.