On this page (8)
Claude Cowork is the part of Claude that carries out multi-step work for you instead of answering one prompt at a time. You describe the outcome, and Claude plans the steps, works in the files, connectors and websites you allow, and comes back with finished documents, spreadsheets or research. Anthropic built it on the same agentic architecture as Claude Code, without the terminal, for knowledge work that is not coding.
This guide explains Cowork in Claude, powered by Anthropic, for the people deciding whether and how to turn it on: what it does, how it differs from chat and Claude Code, which plans include it, where sessions run and where the data goes, and the admin settings to decide before launch. Details are as of October 2026, from Anthropic's product pages and help center.
What does Cowork actually do?
Anthropic describes Cowork as a way to hand Claude multi-step work: you give it a goal, and it works in the folders and tools you choose. On the desktop app, Claude can read and write local files without manual uploads or downloads. Complex jobs are split into smaller tasks that run in parallel, and the output can be a real deliverable, such as an Excel workbook with working formulas, a PowerPoint deck or a formatted document.
- Research and synthesis across documents, connectors and the web, returned as a written brief
- Building spreadsheets, decks and documents from source files, instead of describing how to build them
- Organizing, renaming and summarizing folders of files on the desktop
- Browser tasks through Claude in Chrome, which Anthropic still treats as a higher-risk activity
- Repeatable work driven by skills, connectors and plugins, the same ones used elsewhere in Claude
How is Cowork different from chat and Claude Code?
| Chat | Cowork | Claude Code | |
|---|---|---|---|
| What it is for | Questions, drafting and analysis, one exchange at a time | Multi-step knowledge work that ends in a finished file or result | Software engineering: writing, debugging and shipping code |
| How you work with it | Prompt and reply | Describe the outcome, approve key steps, review the output | Terminal, IDE, desktop, web or Slack, inside a code repository |
| What it can act on | Uploaded files and connectors | Local files on desktop, connectors, the browser, skills and plugins | Your codebase, shell commands, tests and MCP servers |
| Usage | Lowest per task | Anthropic says Cowork consumes limits faster than chat | Billed like other agentic work; heavy sessions use more |
The line between chat and Cowork is fading. On October 23, 2026, Anthropic announced that chat and Cowork are one Claude on web, desktop and mobile for Claude Enterprise, so people no longer pick a mode first: a conversation starts as chat and brings in Cowork when the task needs to work with files. Anthropic says the unified experience is still rolling out on other plans, where Cowork sits next to chat until it arrives.
Which plans include Cowork?
As of October 2026, Anthropic lists Cowork on every paid plan: Pro, Max, Team and Enterprise. It is included in Team Standard and Premium seats, which cost $20 and $100 per seat per month billed annually, or $25 and $125 billed monthly. On Enterprise the $20 seat covers access, and Cowork tokens are billed at API rates on top, like chat and Claude Code. Because Cowork uses tokens faster than chat, budget for it separately and set spend limits by group.
Where does Cowork run, and where does the data go?
Cowork can run in two places. In a local session, Claude's work runs on the user's own computer, with code in an isolated virtual machine, and conversation history is stored on that computer. In a cloud session, tasks run on Anthropic's infrastructure, so work keeps going after someone closes their laptop. As of October 2026, cloud sessions are on by default for Team and off by default for Enterprise, where an owner turns them on and grants them to groups through custom roles.
For records, Anthropic says Cowork sessions are captured in the Compliance API, including local sessions, which Enterprise admins can retrieve. The Compliance API returns sessions from the most recent 30 days, and deletion endpoints for local sessions are not available yet. Team and Enterprise owners can also stream Cowork events to a SIEM through OpenTelemetry, though Anthropic says this does not replace audit logging for compliance.
What can admins control?
- Turn Cowork off for the whole organization; it is on by default
- On Enterprise, enable it for specific teams with groups and custom roles
- Decide whether cloud sessions are allowed, and for which groups on Enterprise
- Choose whether Automatically approve is available and whether new tasks start in it
- Set spend limits at organization, group and member level
- Manage connectors, private plugin marketplaces and the built-in browser from Organization settings
- Stream Cowork activity to a SIEM or observability tool through OpenTelemetry
Is Cowork safe to use with company data?
It can be, with the right scope. Anthropic is direct about the risks: Cowork has internet access and acts on its own, so the risk of prompt injection, where hidden instructions in an email, web page or document redirect Claude, is not zero. Cowork asks before permanently deleting files, and admins choose between manual approval, automatic approval with Claude reviewing each action, or skipping approvals entirely.
Is Cowork covered under HIPAA?
Only after extra configuration. In organizations that have enabled HIPAA, Cowork is off by default, and Anthropic says that without the HIPAA configuration for Cowork, it is available but not covered under your BAA. With that configuration applied, Cowork in the cloud is not available and the built-in browser is turned off. Our HIPAA guide explains who can apply the configuration and what it changes.
How should a company roll out Cowork?
- Decide the approval mode and whether cloud sessions are allowed before anyone starts
- Pilot with one team whose work is file-heavy and repeatable, such as finance, operations or research
- Connect only the systems that team needs, and package its procedures as skills
- Give the pilot group its own spend limit and read actual usage per person after two weeks
- Turn on Compliance API retrieval or OpenTelemetry export before regulated teams join
- Expand group by group, adding each team's skills and connectors as it joins
Questions people also ask
What is Claude Cowork?
Cowork is the part of Claude that carries out multi-step work instead of answering one prompt at a time. You describe an outcome, and Claude plans the steps, works in the files, connectors and sites you allow, and returns finished documents, spreadsheets or research. Anthropic says it uses the same agentic architecture as Claude Code, with no terminal required, and builds it for knowledge work that is not coding.
What is the difference between Claude Cowork and Claude Code?
Claude Code is for software engineering and Cowork is for everything else. Both use the same agentic architecture, according to Anthropic. Claude Code works in a terminal, IDE or the web inside a code repository, running commands and tests. Cowork works in the Claude apps on documents, spreadsheets, folders, connectors and the browser, with no terminal. Engineers usually want Claude Code; analysts, finance and operations teams usually want Cowork.
How is Cowork different from Claude chat?
Chat answers a prompt; Cowork carries out a task. In chat, Claude replies one exchange at a time and describes steps. In Cowork, Claude takes the steps itself, working with your files and tools until the task is done. On Claude Enterprise the two became one experience on October 23, 2026, so a conversation starts as chat and brings in Cowork when the task needs it.
Which Claude plans include Cowork?
All paid plans. As of October 2026, Anthropic lists Cowork on Pro, Max, Team and Enterprise, and it is included in both Team Standard and Premium seats. On Enterprise, the seat covers access and Cowork usage is billed at API rates on top. Anthropic says Cowork consumes usage limits faster than chat, so plan seats and spend limits with that in mind.
Does Cowork run on our computers or in the cloud?
Either, depending on your settings. Local sessions run on the user's computer, with code in an isolated virtual machine, and keep conversation history there. Cloud sessions run on Anthropic's infrastructure and keep going after the laptop closes. As of October 2026, cloud sessions are on by default for Team and off by default for Enterprise, where an owner enables them and grants them to specific groups.
Can admins see what people do in Cowork?
On Enterprise, largely yes. Anthropic says Cowork sessions, including local ones, are captured in the Compliance API, which returns sessions from the most recent 30 days. Team and Enterprise owners can also stream Cowork events to a SIEM through OpenTelemetry, although Anthropic says that does not replace audit logging for compliance. Set up retrieval before regulated teams start using Cowork.
Can we turn Cowork off?
Yes. Cowork is on by default, and organization owners can switch it off for everyone. On Enterprise, groups and custom roles let you enable it for specific teams instead of all or nothing. In organizations that have enabled HIPAA, Cowork starts off by default. A sensible approach is to leave it off at launch and turn it on for one pilot team once approval and spend settings are decided.
