
Use Claude Code when the work lives in a code repository: features, bug fixes, tests, scripts and anything that ends in a commit. Use Claude Cowork when you want to hand Claude a multi-step task across your files, connected apps and browser and get a finished deliverable back. Use chat for questions and drafts you steer turn by turn. Since 16 September 2026 Anthropic has been merging Cowork and chat into one Claude app, plan by plan. Claude Code stays separate.
We implement Claude for client teams and run Claude Code in our own operations. Every product fact below was checked against Anthropic's own pages on 4 November 2026.
What is the difference between Claude Code and Claude Cowork?
Claude Code works inside a code repository and ends in a commit, while Cowork works across your files, connected apps and browser and ends in a finished document or spreadsheet. The two are closely related: Anthropic's help center says Cowork "uses the same agentic architecture that powers Claude Code, with no terminal required." Both can take a goal, plan the steps, run them and show you what they did.
Claude Code is an agentic coding tool. In Anthropic's words it "reads your codebase, edits files, runs commands, and integrates with your development tools." It runs in the terminal, in VS Code and JetBrains, in the Code tab of the Claude desktop app, and on the web at claude.ai/code. It works with git directly: it stages changes, writes commit messages, opens pull requests and can run in CI through GitHub Actions or GitLab.
Claude Cowork is for handing off a task. Claude reads, edits and creates files in the folders you connect, uses your connectors and a built-in browser, runs code in an isolated environment, and delivers a finished output for you to review. Anthropic's product page aims it at marketers building reports, finance teams reconciling data, legal teams reviewing contracts and sales teams working through call notes.
A practical test: if the result of the task is a change to a codebase, use Claude Code. If the result is a document, a spreadsheet, a deck or a set of updated records, use Cowork.
Anthropic's newest agent works the other way round. Claude Duet, released as a US-only research preview on 21 October 2026, checks your connected apps and comes to you when something needs attention. We compare it with Cowork in Claude Duet vs Cowork: what it is and who can use it.
Claude chat vs Cowork vs Code: side-by-side comparison
| Chat | Cowork | Claude Code | |
|---|---|---|---|
| Best for | Questions, drafts, analysis you steer turn by turn | Multi-step tasks that end in a finished file or updated records | Work inside a code repository: features, fixes, tests, scripts |
| Where you use it | Web, desktop and mobile apps | Desktop app (macOS, Windows, Linux in beta), web and mobile (beta on Team; on Enterprise when an owner enables it), and the Chrome side panel | Terminal, VS Code, JetBrains, desktop app Code tab, web, mobile |
| What it works on | The conversation, uploaded files, connectors | Folders you connect, connectors, a built-in browser, computer use on desktop | Your codebase, shell commands, git, MCP servers |
| Where the work runs | Anthropic's servers | A cloud sandbox on Anthropic's servers by default. On Team and Enterprise, also the member's computer, with code in a local virtual machine | Your machine (terminal, IDE, desktop) or Anthropic's cloud (web, cloud scheduled tasks) |
| Plans | All, including Free | Pro, Max, Ultra, Team, Enterprise | Pro, Max, Ultra, Team, Enterprise |
| Main admin control (Team/Enterprise) | Connector controls; roles and custom retention on Enterprise | Org-wide on/off toggle, cloud-session toggle, auto-approve toggles | Managed settings that override every developer's own settings |
Sources: claude.com/pricing, Get started with Claude Cowork, Cowork architecture overview, Claude Code overview. All checked 4 November 2026.
Claude chat vs Cowork: are they still separate?
It depends on your plan and your admin: on Pro, Max and Ultra they are becoming one product, and on Team and Enterprise they stay separate until the unified experience is turned on. Anthropic's 16 September 2026 announcement, Claude Cowork and chat are now one Claude, says the two "are merging into one Claude." The stated reason is that users found it frustrating to decide where a task belonged, and work started in one place did not carry into the other. Claude now works out what a task needs, so Cowork's capabilities are available from any conversation, with the same connectors and skills.
Where the rollout stands on 4 November 2026:
- Pro, Max and Ultra: rolling out in the Claude app on web, desktop and mobile. Nothing to turn on.
- Team: in beta, available where the unified experience is turned on.
- Enterprise: generally available since 23 October 2026, where the unified experience is turned on.
- Free: Anthropic's announcement says it "will follow soon." Handing off tasks is still listed as a paid feature on the pricing page.
You can check your own account: Anthropic's help center says that if your message box no longer shows "Chat" and "Cowork" options, you have the new experience. Organizations that haven't turned it on keep the two as separate options, and the admin toggles described below still apply.
Plan the training before you turn it on. People who only ever asked Claude questions can then start handing it tasks from the same window, and Anthropic's own guidance is that multi-step tasks, "like ones that run code, create files, or use your connected apps and browser, use more of your usage than a quick question." So Cowork beats chat for a multi-step job that ends in a file, and costs you more of your limits for a quick question.
Check the approval default as well, because Anthropic's pages describe it in two ways. The announcement says "By default, Claude asks before taking an action." The admin article lists a setting that decides whether new Cowork tasks start in "Automatically approve" mode, and it is on by default on Pro, Max, Ultra and Team and off by default on Enterprise. On Team, read that setting in your own organization before you rely on the announcement.
When should a business team use Cowork?
Cowork fits work that has three traits: it takes several steps, its inputs sit in files or connected apps, and the output is something a person reviews before it goes anywhere. A weekly pipeline summary, a month-end reconciliation of two exports, a first-pass review of a contract folder, a deck built from a report.
Two capabilities matter most for delegation. Cowork can schedule a task at any cadence and run it unattended, and it keeps going after you close your laptop when the session runs in the cloud. Both are also why Anthropic's safety guidance tells you not to "schedule tasks that access sensitive files, send messages on your behalf, make purchases, or take other actions that are difficult to undo." Start scheduled Cowork tasks on work where a wrong output costs you a re-run, not a customer email.
Cowork is the wrong tool when the task is really a piece of software: when you want the same logic to run the same way every time, be reviewed as code, and be fixed in one place when it breaks. That is where Claude Code comes in, even if nobody on the team calls themselves an engineer.
When is Claude Code the better choice, even outside engineering?
Choose Claude Code when the work depends on files you maintain over time: scripts, instructions, configuration. We run part of our own marketing operations this way. A scheduled Claude Code routine works from a folder of Python scripts and a written runbook, and three properties of Claude Code made it the right home for that job.
- Rules live in files. Claude Code reads permission rules from settings files, and a shared
.claude/settings.jsonsits in the project where anyone can review it. Our routine only runs commands that match an allowlist exactly. When an early version chained several commands into one, the chained command matched nothing on the list and the run stalled waiting for approval. The fix was to run every script as its own command, which is now written into the runbook. - Gates are scripts. A quality check runs before anything is published and exits with an error if the draft fails. The routine cannot talk its way past an exit code.
- Verification is part of the job. Each run ends by fetching the live page and checking the change is actually there, then writing one log line, including on runs that ship nothing.
Cowork could produce a similar draft. What Claude Code gave us is an operation we can read, diff and hand to someone else. If your team's recurring task has started to look like "run these five scripts in order and check the result," it has become a Claude Code job. Our guide to rolling out Claude Code across an engineering team covers the repository instructions, permission rules and review habits in detail.
What are examples of Claude Cowork tasks, and which belong in Claude Code?
Anthropic's Claude Cowork examples are a weekly marketing readout, a spreadsheet reconciliation, a first-pass contract review and a call-note analysis. All four work well as Cowork tasks while a person reads each result. The right-hand column below is our judgement of when each one has outgrown Cowork.
| Job | The Cowork task, from Anthropic's pages | When it moves to Claude Code |
|---|---|---|
| Weekly report | "Every Monday morning, build last week's marketing readout": pull numbers from Amplitude and Drive, flag anything that moved more than 10%, save the slides to Drive | People argue about how a metric is calculated, so the calculation needs to live in one reviewed script |
| Month-end reconciliation | Point Claude at the regional exports and budget files, and get a working spreadsheet with variances flagged and a note on what didn't reconcile | The matching rules must be identical every close, and an auditor may ask when a rule changed |
| Contract review | Point Claude at a folder of agreements and your playbook, and get a memo per contract with each departure cited to the clause | Rarely. A lawyer reads every memo, so a task that varies by document is fine |
| Call-note analysis | Point Claude at a quarter of call notes, and get a memo on where every open deal stands, with objections quoted from the notes | The output starts writing back to your CRM on a schedule, with no one reading it first |
| Expense report | Turn a folder of receipts into a formatted expense report | Rarely. It is a one-off with a person checking the total |
Sources: Claude Cowork product page and Get started with Claude Cowork, checked 4 November 2026.
We made that move for our own weekly search report. Anthropic's example prompt asks Claude to note "anything you couldn't pull", which works when someone reads the deck on Monday. Our report decides what we publish next, so it runs in Claude Code from Python scripts that save every data pull to a file and record a failed pull as failed. On our 1 November 2026 run, 7 of 46 tracked search prompts returned errors from the data provider. The retry recovered 5, and the report shows the other 2 as unmeasured.
Which one should you use? A decision flow
Most teams end up using at least two of the three. Operations, finance and sales staff work in chat and Cowork. Engineers, and whoever owns the scripted parts of your operations, use Claude Code. Standardising on one product for the whole company forces every task through a tool built for a different kind of work.
What should admins decide before turning Cowork or Claude Code on?
On Team and Enterprise plans, the two products are governed in different places.
For Cowork, Anthropic's Team and Enterprise admin article lists the controls:
- Cowork is on by default, and an organisation-wide toggle under Organization settings turns it off for everyone. In organizations that have enabled HIPAA, it starts off.
- Cloud sessions are on by default for Team and off by default for Enterprise, where an admin grants them to a group through custom roles.
- Separate toggles decide whether members may use "Automatically approve" mode and "Always allow" for connector tools. The connector setting is off by default. A further setting decides whether new tasks start in "Automatically approve": on by default for Team, off for Enterprise.
- Cowork sessions through Claude, Claude Desktop and Claude Mobile are captured in the Compliance API, and events can stream to your SIEM through OpenTelemetry.
- Local sessions store conversation history on the member's computer. Anthropic states this data "is not subject to Anthropic's standard data retention policies, and admins cannot centrally manage or delete it." Enterprise admins can retrieve it through the Compliance API, but deletion endpoints for local sessions aren't available yet.
That last point decides a lot for regulated teams. If you need central deletion, favour cloud sessions, or keep sensitive folders out of Cowork until deletion endpoints for local sessions exist. Anthropic's safety guidance also recommends dedicated working folders over broad access to sensitive files.
For Claude Code, control sits in settings files that stack in a fixed order. Managed settings, deployed by your organisation through a managed-settings.json file, device management or the claude.ai console, sit at the top and override anything a developer sets locally. That is where company-wide permission rules belong.
On Enterprise, role-based access with fine-grained permissions decides which groups get which capabilities; Cowork's cloud sessions, for example, are granted to a group through custom roles. Our Claude Enterprise rollout guide covers identity, roles and spend limits in the order we set them up. If an outside firm will do that setup, our guide to verifying an Anthropic partner shows how to confirm its status.
Does Cowork or Claude Code cost extra?
Neither is a separate purchase on paid plans. Prices below are from Anthropic's pricing page on 4 November 2026, before tax.
| Plan | Price | Claude Code | Cowork |
|---|---|---|---|
| Free | $0 | No | No |
| Pro | $17/month billed annually, $20 monthly | Yes | Yes |
| Max | $100/month | Yes | Yes |
| Ultra | $200/month | Yes | Yes |
| Team, standard seat | $20/seat/month billed annually, $25 monthly | Yes | Yes |
| Team, premium seat | $100/seat/month billed annually, $125 monthly (5x standard usage) | Yes | Yes |
| Enterprise | US$20/seat/month billed annually, 20-seat minimum, plus usage at API rates | Yes | Yes |
The seat price is the smaller question. Anthropic's help center says multi-step tasks use more of your usage than a quick question, and its Enterprise pricing notes that usage cost "scales with model and task." The people who delegate the most work hit limits first. On Team, that is the case for giving premium seats to heavy users and standard seats to everyone else. On Enterprise, where usage is billed at API rates on top of seats, it is the case for spend limits per group. Our Claude for business adoption page sets out how we plan a controlled pilot, access boundaries and the operating model before a wider rollout, and Claude Code for companies covers the engineering side.
Frequently asked questions
Is Claude Cowork the same as Claude Code?
No. They share the same agentic architecture, but Claude Code is built for codebases and runs in a terminal, IDE, the desktop Code tab or the web. Cowork is built for tasks across your files, connected apps and browser, with no terminal involved.
Can non-developers use Claude Code?
Yes, through the desktop app's Code tab or the web, without installing anything in a terminal. It makes sense when the task depends on scripts or instruction files you maintain. For one-off tasks on documents and apps, Cowork is the simpler fit.
Is Cowork available on the free plan?
No. Anthropic lists Cowork on paid plans only: Pro, Max, Ultra, Team and Enterprise. Claude Code is also not included on Free.
Does Cowork run on my computer or in the cloud?
The cloud, by default: each session gets a temporary sandbox on Anthropic's servers and keeps going after you close your laptop. Team and Enterprise can also run local sessions, which keep the agent on the member's device and execute code inside an isolated virtual machine. On Pro, Max and Ultra, only tasks started on the device before 6 October 2026 stay local. Cloud sessions are in beta on Team and need an owner to enable them on Enterprise.
What happened to Cowork in September 2026?
On 16 September 2026 Anthropic began merging Cowork and chat into a single Claude app, starting with Pro, Max and Ultra. Existing chats, projects, connectors and skills carry over. The unified experience reached general availability on Enterprise on 23 October 2026 and is in beta on Team.
Do we need both Cowork and Claude Code?
If you have an engineering team, you will likely use both, and the same Team and Enterprise seats include both. Decide which work goes where and who governs each.
How this article was produced
Written by the ATI implementation team. First published 29 September 2026. Product, plan and admin details were verified again on Anthropic's pricing page, help center, product pages, blog and Claude Code documentation on 4 November 2026. The operating details and the 1 November run figures about our own Claude Code routine come from ATI's internal runbook and logs. AI assistance was used in research and drafting; a human owns every claim and recommendation in it.