Comparison · Updated

Claude Cowork vs Claude Code: what each is for and when to use which

Claude Cowork vs Claude Code as of October 2026: what each agent is built for, where tasks run, plugins and skills, scheduling, usage, and which to use when.

Use Claude Code when the work lives in a code repository: it edits files across a codebase, runs your tests and commands, and opens pull requests, from the terminal, your IDE, the desktop app's Code tab or the web. Use Cowork when the output is a document, spreadsheet, deck or a task across your apps: Anthropic built it for non-coding knowledge work, on the same agentic architecture as Claude Code but with no terminal. Both come with paid Claude plans, and since September 16, 2026, Cowork is merging into the main Claude app, so you reach it from the same message box as chat.

Cowork and Claude Code at a glance

As of October 2026, from Anthropic's product pages, help center and Claude Code docs.

CoworkClaude Code
Built forKnowledge work: research, analysis, documents, spreadsheets, slidesSoftware: features, bug fixes, tests, refactors, pull requests
Where you use itClaude app on desktop, web and mobile; Claude in Chrome side panelTerminal, VS Code, JetBrains, desktop app (Code tab), web, mobile app, Slack
PlansPro, Max, Team, EnterprisePro, Max, Team, Enterprise, or an API key
Where tasks runIn the cloud on Pro and Max; local files, browser and computer use need the desktop app openOn your machine, or in the cloud on claude.ai/code
Works withFolders you connect, connectors, built-in browser, computer use (research preview)Your repository, shell commands, git, MCP servers
ApprovalsManual (default), Auto, or SkipAuto (default in recent versions), Manual, accept edits, plan, or bypass
ExtensionsPlugins, skills, connectors, sub-agentsPlugins, skills, MCP, subagents, hooks
SchedulingScheduled tasks in the cloud, on demand or on a cadenceRoutines in the cloud, desktop scheduled tasks, /loop
Typical outputA finished file or deliverableA commit, branch or pull request

Same engine, different job

Anthropic's help center puts it plainly: Cowork uses "the same agentic architecture that powers Claude Code, with no terminal required." Both plan a multi-step task, split it into subtasks, call tools, and check their own work. What differs is the toolset around the loop.

Claude Code's tools are a developer's: it reads and edits source files, runs shell commands and test suites, works with git, and talks to services through MCP servers. It reads CLAUDE.md for project rules and can run hooks on 33 lifecycle events, so you can, for example, run the linter after every edit or block writes to .env.

Cowork's tools are an office worker's: files in folders you connect, connectors to apps like Google Drive or Slack, a built-in browser that is separate from yours, and, in research preview, the ability to use your computer's apps. It produces spreadsheets with working formulas, slide decks and formatted documents. Since September 16, 2026, Claude Docs and Claude Slides (both in beta on paid plans) let you edit and present those outputs inside Claude.

How tasks run: your computer or the cloud

Cowork moved to the cloud in October 2026. On Pro and Max, new Cowork tasks run on Anthropic's servers and follow your account across desktop, web and mobile, so a task keeps going after you close your laptop. Anything that needs your machine (local files, local connectors, browser use, computer use) still needs the desktop app open on that computer. On Team, Cowork on web and mobile is in beta; on Enterprise, an admin turns it on.

Claude Code gives you the choice per session. Run claude in a folder and it works on your machine with your tools. Start a session on claude.ai/code or in the mobile app and it runs in a cloud container against a GitHub repository, and you can pull it back to your terminal with claude --teleport.

Extending each one

The extension models overlap more than they differ. In both, skills package know-how, connectors or MCP servers reach external apps, and plugins bundle all of it into one install. Anthropic says a plugin made for Cowork also works in chat and Claude Code, so a team can write a skill once (say, how to build the monthly board report) and use it in either.

Claude Code adds hooks and subagent definitions you control in files, which matter for software work where some steps must happen every time. See Claude Code agent teams for how it runs several agents in parallel.

Scheduling

Both can work on a schedule without you:

  • Cowork scheduled tasks run daily, weekly or monthly in the cloud, for example a Monday summary of last week's sales spreadsheet.
  • Claude Code routines run in Anthropic's cloud on a schedule (hourly at most), from an HTTP call, or on a GitHub pull request or release, for example a nightly review of open PRs. Desktop scheduled tasks run on your own machine, and /loop repeats a prompt inside a session.

Usage and approvals

Both draw on your Claude plan. Anthropic notes that Cowork uses your limits faster than chat, and that its Auto mode uses even more, so long research tasks add up. Claude Code usage is metered the same way, in five-hour windows plus weekly limits on paid plans.

The approval models look alike. Cowork's default is Manual (you allow or deny each action); Auto lets Claude keep working while a safety check reviews each action; Skip runs with no checks. Claude Code works much the same way: since version 2.1.283, interactive terminal and VS Code sessions start in auto mode, where a second model reviews actions instead of you, and Manual mode asks before most edits and commands. The --dangerously-skip-permissions flag skips permission prompts altogether, apart from a few protected actions; our skip permissions guide covers what it does and doesn't skip.

Which to choose

  • Fix a bug, add a feature, write tests, open a PR: Claude Code.
  • Turn a folder of PDFs into a comparison spreadsheet: Cowork.
  • Research competitors and draft a slide deck: Cowork.
  • Upgrade dependencies across three repositories overnight: Claude Code, as a routine or cloud session.
  • Fill in a web form or work inside a desktop app: Cowork, with browser or computer use.
  • Not sure whether a question needs a task at all: start in the Claude app; see Claude chat vs Cowork and Claude vs Claude Code.

Questions

Is Cowork just Claude Code without the terminal?

Close. Anthropic describes Cowork as using the same agentic architecture as Claude Code, with no terminal required, aimed at knowledge work such as research, documents and spreadsheets rather than software in a repository.

Do Cowork and Claude Code cost extra?

No. Both are included in paid Claude plans (Pro, Max, Team and Enterprise) and count against that plan's usage limits. Neither is in the free plan.

Can a plugin work in both Cowork and Claude Code?

Yes. Anthropic's Cowork guide says plugins bundle skills, connectors and sub-agents into one package that also works in chat and Claude Code.

Can Cowork edit code?

It can read and write files in folders you connect, so it can touch code, but it has no terminal workflow, git integration or test loop. For changes to a codebase, Claude Code is the right tool.

Does Cowork keep working when my laptop is closed?

On Pro and Max, yes: since October 6, 2026, new Cowork tasks run in the cloud. Tasks that need local files, the built-in browser or computer use still need the desktop app open.

More comparisons

All comparisons

Know when your agents need you

Briefkin puts your Claude Code, Codex and Cursor sessions on one board: working, needs you, done or failed. Telegram pings you when one is waiting on you, and scheduled skills post their reports next to them.

Free during the open beta. No card.