Claude Cowork: The Complete Guide

Most AI tools wait for your next prompt. Claude Cowork waits for nothing — you hand it a goal, and it works across your files, folders and apps until there is a finished deliverable to review. Anthropic calls it "agentic AI for knowledge work," and it is the company's attempt to take the autonomy that made Claude Code useful to developers and hand it to everyone else: analysts, operations teams, legal, finance, and anyone whose day is full of document-shaped work.

This guide covers what Cowork actually is, how it works, how to set it up and use it, what it is good at, where its limits are, and who should bother. Every capability below is drawn from Anthropic's own product pages and reputable coverage — not speculation.

What Claude Cowork is

Cowork is a mode of the Claude desktop app that sits alongside Chat and Code. Where Chat is a conversation and Code is a developer's terminal agent, Cowork is a working session: you describe the outcome, Claude plans and executes the steps, and you steer along the way. Anthropic's framing on the product page is blunt — "give it a goal and Claude works on your computer, local files, and applications to return a finished deliverable."

The lineage matters. Per TIME and Wikipedia, Cowork launched in January 2026 as a research preview and is essentially Claude Code's agentic engine wrapped in a graphical interface for non-technical users. Claude Code, by contrast, launched as a developer preview in February 2025 and went generally available in May 2025. Cowork reached general availability across all paid plans in spring 2026, according to coverage of the rollout (TestingCatalog reported a 9 April 2026 GA date).

The model under the hood has moved with Anthropic's release cadence. Reporting and release notes indicate Cowork ran on Claude Opus 4.5 at preview launch, and that Claude Sonnet 4.6 later became the default model across Free and Pro plans, Claude Code, and Cowork. Anthropic does not pin Cowork to a frozen model — it tracks the current default.

Why it matters

The pitch is that a large slice of knowledge work is time-consuming but not technically hard: renaming and sorting a folder of files, pulling numbers out of contracts, assembling source material into a draft, rebuilding the same weekly report. A chatbot can tell you how to do those things; it cannot do them, because it has no hands on your filesystem. Cowork's difference is exactly that permission — as the Claude.com product page puts it, in Cowork "Claude has permission to read, edit, and create files in folders you specify," so it completes the task rather than describing it.

That shifts the unit of interaction from the prompt to the outcome. You are not coordinating each step; you are reviewing a result.

How it works

Anthropic describes a three-step loop:

1. Describe the task. From the desktop app (or your phone), you state the goal and, for recurring work, the cadence.
2. Claude works. It moves between local files, folders and applications, synthesizes information across sources, and executes multi-step work — "looping you in before anything significant." For consequential actions it shows you the plan and waits for approval.
3. Review the deliverable. You check a finished output rather than a stream of step-by-step updates, and you can redirect, refine, or change approach at any point.

Access is scoped by you. Cowork only touches the folders and connectors you grant it, and Anthropic stresses the design keeps "consequential decisions with the user." When there is no direct integration for an app, Cowork can control the screen to navigate it, and it can use Chrome for web research.

How to set it up and use it

1. Get on a paid plan and install the desktop app. Cowork is available on all paid plans through the Claude desktop application; there is no separate purchase.
2. Open the Cowork mode alongside Chat and Code in the app.
3. Grant folder access. Point Claude at the specific folder(s) it should read and write — keep this tight; it is your main control surface.
4. Connect your tools. Cowork supports connectors including Slack and, for organizations, Google Drive, Gmail, DocuSign, FactSet and a Zoom MCP connector for meeting transcripts and action items.
5. Describe the outcome, not the steps. "Turn these 12 receipts into a formatted expense spreadsheet" beats a click-by-click script.
6. Approve the plan, then let it run. Review the proposed plan, approve, and check the deliverable at the end.

For recurring work, use scheduled tasks: Cowork can check your email each morning, pull metrics into a weekly report template every Friday, or run a weekly Slack digest on a set cadence.

Key features at a glance

Limitations and tips

• It is not a chatbot. If you want a quick answer or a conversation, use Chat — Cowork is built around producing a deliverable, which is overkill for a one-line question.
• Permission is the safety boundary. Cowork can read, edit and create files. Scope it to a dedicated working folder rather than your whole drive, and review the plan before approving — the same "least privilege" discipline that applies to any autonomous agent.
• Consequential actions still need you. By design, Cowork pauses for approval on significant steps; it is an assistant under supervision, not an unattended robot.
• Connector availability varies by tier. Several integrations (Google Drive, Gmail, DocuSign, FactSet) are framed around organizational deployment; check what your plan exposes.
• Mobile is catching up. Anthropic has been extending Cowork to the phone — you can message it from mobile — but the desktop app remains the primary surface.
• Tip: write outcome-first prompts. State the finished result and the format you want; Cowork plans the steps better than you can script them.

FAQ

Is Claude Cowork the same as Claude Code? No. They share an agentic engine, but Code is a terminal tool for developers and Cowork is a graphical app aimed at non-technical knowledge workers.

Do I need to be a developer? No. Anthropic built Cowork specifically for people without a technical background — researchers, analysts, operations, legal and finance teams.

Can it touch my files? Yes, but only the folders you grant. Within those it can read, edit and create files; everything else is off-limits.

What does it cost? It is included on all paid plans through the Claude desktop app; there is no separate Cowork subscription. Enterprise tiers add spend controls and analytics.

Which model runs it? Cowork tracks Anthropic's current default — Opus 4.5 at preview launch, later Sonnet 4.6 across consumer plans — rather than being fixed to one model version.

Can it run tasks on a schedule? Yes. Scheduled tasks handle recurring work like a daily email check or a weekly report on a cadence you set.

Bottom line

Claude Cowork is the clearest sign yet that AI assistants are moving from answering questions to finishing work. For anyone who spends their day wrangling documents, data and files, the value is concrete: hand over a folder and a goal, approve a plan, and review a deliverable. Treat folder access as the safety dial, keep humans on the consequential decisions, and Cowork becomes less a chatbot and more a supervised coworker. If you already lean on Claude Code, this is the same power pointed at the rest of your workday.

Compare Claude Code with Cursor and Codex

Sources and further reading