Claude Cowork: Every Feature Explained (2026 Guide)

Claude Cowork is Anthropic's "AI coworker" — a mode inside Claude that works directly with your files, folders and connected apps and hands back finished work, instead of just talking about it. Where Chat is a conversation and Claude Code is a coding agent, Cowork is a working session for everyday knowledge work: you describe a goal, Claude plans and executes it across many steps, you steer along the way, and it ships a real Excel model, a PowerPoint deck, or a formatted document you can actually use.

It went generally available on April 9, 2026, it's the third tab in Claude (Chat · Code · Cowork), and it's included on every paid Claude plan — there's no separate add-on to buy. This guide walks through every feature: how sessions work, the folder-access model, the three permission modes, scheduled unattended tasks, plugins and integrations, the security model, pricing, current limits, and exactly who it's for.

What Claude Cowork Actually Is

Regular Claude Chat answers questions and hands you text. Cowork takes the next step: you give it a goal and it does the task. In Anthropic's words, "Claude Cowork completes tasks you can steer from anywhere. Give it a goal, and it works across your files and tools." It's aimed squarely at non-coding knowledge work — research, analysis, document creation and other multi-step jobs that used to eat an afternoon.

The mental model that matters: Cowork is a working session, not a chat thread. You describe the outcome, Claude proposes a plan, you approve it, and then it executes step by step — opening files, running tools, making decisions — while showing its work. "Claude shows each step: the files it opens, tools it uses, and choices it makes." For big jobs it even parallelizes: "Claude Cowork splits big projects into chunks that run together," coordinating sub-agents so several parts of a project progress at once.

Where It Runs: Desktop, Web and Mobile — One Continuous Session

Cowork runs on the Claude desktop app for macOS, Windows, ChromeOS and Linux, with web (claude.ai) and mobile (iOS/Android) in beta and rolling out gradually. The desktop app must stay open when a task needs your local files, your browser, or other computer functions.

The standout here is continuity. Because the work runs remotely on Anthropic's servers, a session persists even when the desktop app closes or your computer sleeps — "Close your laptop, Claude Cowork keeps going." You can start a task at your desk, check its progress from your phone, and open the finished deck on the web later, all in the same session. That "start here, finish there" behavior is one of the biggest practical differences from a normal chat.

How Claude Cowork Handles Your Files

Access is scoped and opt-in. You choose the specific folders and tools Claude may use, and — in Anthropic's phrasing — "Claude can't reach anything else." Inside the folders you authorize, it can read, edit and create files; outside them, it's blind. You can also attach folder instructions that give Claude project-specific context (naming conventions, house style, what lives where), and Claude can update those instructions as a project evolves.

Supported file types cover most day-to-day office work:

One hard guardrail worth repeating: Claude requires your explicit permission before permanently deleting any file. This "deletion protection" is on regardless of which permission mode you're using.

The Three Permission Modes: Manual, Auto and Skip

Cowork's autonomy is a dial, not a switch. Three modes decide how often it stops to ask you:

Two things to know. First, Auto mode consumes more of your usage limit than the others because of the extra safety-review overhead. Second, on Team and Enterprise plans, admins can enforce per-task approval for write-capable connectors, so an individual can't switch to Skip on company data.

What It Produces: Real Deliverables, Not Just Text

This is the point of Cowork. Instead of describing how to build a report, it builds the report. Documented outputs include:

• Excel spreadsheets with "working VLOOKUP, conditional formatting, and multiple tabs" — not a screenshot of a table, an actual working workbook.
• PowerPoint decks generated from your notes or a meeting transcript.
• Formatted documents in Markdown/Word, editable inline with the "Edit with Claude" feature.
• CSV and structured data files for downstream use.

The workstreams it's built for line up with those outputs: research synthesis (combining web searches and articles into a brief), transcript analysis (pulling themes and action items out of a call), statistical analysis (outlier detection, cross-tabulation, time-series), and the data visualization and transformation that turns raw numbers into something a stakeholder can read. Because these are compute-heavy, multi-step jobs, they use more tokens than a normal chat — more on that in pricing, and you can sanity-check model costs on our Token Calculator.

Scheduled & Unattended Tasks

Cowork isn't only reactive — it can run on a timer. Type /schedule inside any task to set it up, then manage everything from the "Scheduled" sidebar. Scheduled tasks run remotely and unattended: no device or app needs to be active, and you can pick any cadence — a one-off run at a set time, or a recurring "every Monday at 8am" job. This is what turns Cowork from an assistant into something closer to a standing process: a weekly numbers pull, a daily inbox triage, a recurring competitive-research digest that's waiting for you when you sit down.

Plugins, Connectors and Computer Use

Out of the box Cowork is a generalist; plugins turn it into a specialist. A plugin bundles "any skills, connectors, and sub-agents together to turn Claude into a specialist" for a role or team. Anthropic ships domain plugins including Brand Voice, Legal and Finance, and organizations can build customizable ones across areas like financial analysis, engineering and HR.

Beyond plugins, Cowork extends through:

• Connectors / MCP — connect external tools (CRM, calendar, email, messaging apps, call recordings, Google Drive, Gmail, DocuSign, FactSet, and more) via Settings → Connectors. Network access follows your configured egress settings.
• Claude in Chrome — browser automation for "clicking, typing, navigating, and filling forms." Computer use is in research preview; Claude reaches for your connectors first and falls back to the browser when there's no direct integration.
• Live Artifacts — a Team/Enterprise, desktop-only feature for sharing work artifacts inside your organization.

Security & Isolation: How Anthropic Contains It

Handing an agent your files raises the obvious question, so the security model is worth stating plainly. Cowork's work runs on Anthropic's servers in an isolated environment, separate from your computer and network. Specifically:

• Session isolation — shell commands and code execute in a sandbox, not on your machine.
• Scoped file access — Claude can only touch the folders you connect, nothing else.
• Controlled network egress — outbound access is governed by your egress configuration.
• Deletion protection — permanent deletes always require your explicit approval.

For companies, there's an enterprise-grade layer on top: admin controls to manage feature access, control spend and track usage across the org; deployment through your Claude account or via Amazon Bedrock, Google Cloud or Microsoft Foundry; and observability where "activity streams to your SIEM through OpenTelemetry." Privacy is a real strength of the Claude ecosystem generally — see how it scores on Claude's tool page and our 8-dimension methodology.

Claude Cowork vs Claude Code — Which One Do You Need?

They share the same agentic DNA and people constantly confuse them, but they're built for different jobs. The one-line rule: if your output is source code, use Claude Code; if it's documents, data and decks, use Claude Cowork.

Both are included in the same paid Claude plans, so it's not either/or — most people will use Cowork for the report and Code for the app. If coding is your world, start with our complete guide to Claude Code in VS Code or the Grok Build 0.1 vs Claude Code head-to-head.

Claude Cowork Pricing in 2026

There is no free Cowork tier. It's bundled into every paid Claude plan, and the key nuance is that the plans sell usage, not features — every paid tier runs the same full Cowork, and the higher tiers just give you a bigger allowance for these token-hungry agentic tasks.

Prices are approximate and vary with annual vs monthly billing — always check the live Claude tool page before subscribing. Practical advice: start on Pro and only move up if your daily use hits the limit. Because agentic tasks "consume more capacity than regular chat," watch Settings → Usage, batch related work into one session, and keep simple questions in normal Chat.

Current Limits & What to Watch

Cowork is powerful but young, and a few rough edges are worth knowing before you rely on it:

• Memory doesn't fully carry over. What Claude knows from Chat doesn't automatically flow into Cowork yet (projects are the exception), so you may need to restate context.
• No session sharing (yet). Sharing a live session isn't available; the closest thing is Live Artifacts on Team/Enterprise.
• Some plugins need the desktop app. Plugins that rely on a local MCP server won't run on web/mobile.
• Web and mobile are still beta. The rollout across paid plans is ongoing, so parity with desktop isn't complete.
• It's usage-hungry. Multi-step agentic work burns tokens far faster than chat — the biggest practical constraint on a Pro plan.

Who Should Use Claude Cowork

Worth it if you spend real time turning inputs into deliverables — an analyst building recurring Excel models, a marketer producing decks and briefs, an ops person triaging inboxes and reconciling files, a founder who needs a research memo by morning. If you already pay for Claude Pro or Max, Cowork is included, so there's no reason not to put your most repetitive "take these files and produce that document" jobs through it.

Hold off if your Claude use is casual Q&A and light writing — you'll get more from Chat and won't touch Cowork's usage ceiling, and there's no free tier to experiment with. And if you're a developer whose output is code, Claude Code is the tool you actually want. Not sure which AI fits your workflow at all? Pickurai's 6-question wizard matches you in about 30 seconds.

The bottom line: Claude Cowork is the clearest sign yet that the "chatbot" era is giving way to the "AI coworker" era — an agent that reads your files, runs multi-step work unattended, and hands back something you can send. It's scoped, sandboxed and admin-governable enough to trust with real work, and because it's bundled into plans you may already pay for, the only real cost of trying it is the usage it burns.

Frequently Asked Questions

What is Claude Cowork?

Claude Cowork is Anthropic's agentic "AI coworker" mode — a third tab alongside Chat and Code — that works directly with your files, folders and connected apps instead of just answering in a conversation. You give it a goal, it plans and executes across multiple steps, and it returns finished deliverables like Excel spreadsheets, PowerPoint decks and formatted documents. It went generally available on April 9, 2026 and runs on Anthropic's servers in an isolated environment.

Is Claude Cowork free?

No — there's no free Cowork tier. It's included on every paid Claude plan (Pro, Max, Team, Enterprise), and the plans sell usage rather than features, so every paid tier runs the same full Cowork. The cheapest way in is the ~$20 Pro plan.

What's the difference between Claude Cowork and Claude Code?

Claude Code is a coding agent for developers working in a terminal or IDE; Claude Cowork is a general knowledge-work agent for non-coding tasks — research, analysis, documents, spreadsheets and decks — working across your files and business apps. If your output is source code, use Code; if it's documents and data, use Cowork.

What files can Claude Cowork access?

Only the folders and tools you explicitly connect — it can't reach anything else. Inside those folders it reads, edits and creates .docx, .pdf, .txt, .md, .xlsx, .pptx, images, code files and JSON/CSV/YAML. Local file access needs the desktop app open, and Claude asks before permanently deleting anything.

Does Claude Cowork keep running when I close my laptop?

Yes. Tasks run remotely in an isolated environment, so a session continues even if you close your laptop or your computer sleeps. The same session is available across desktop, web and mobile, and /schedule can run tasks unattended on any cadence with no device active.

Is Claude Cowork safe to give access to my files?

It's built around scoped permissions and isolation: Claude only reaches folders you connect, its code runs sandboxed away from your machine, network egress is controlled, and deletion needs explicit approval. Three permission modes (Manual, Auto, Skip) set its autonomy, and Team/Enterprise admins can enforce per-task approval and stream activity to a SIEM.