Anthropic Unifies Claude Chat and Cowork for Enterprise: The November 16 Switch and the $20 Seat

On October 23, 2026, Anthropic removed the last menu choice between Claude chat and Cowork for its Enterprise customers. Every conversation begins as a chat, and when the work needs code run or files touched, Claude carries the same conversation into an isolated cloud sandbox managed by Anthropic, then brings the finished work back for review. The change lands on web, desktop, and mobile at once — and for existing organizations it is off by default until November 16, 2026, when it switches itself on for any organization that already has chat and cloud code execution enabled.

That date is the part worth circling. Anthropic announced the merge on a Friday, gave admins three and a half weeks, and set a default flip that fires silently. If your organization has both capabilities on and nobody touches the setting, your people will be working in the unified experience on the morning of November 16 whether or not anyone planned for it. The work itself does not change — chats, projects, connectors, and skills carry over — but where the work runs, and what it can reach, does change.

In short: Anthropic merged Claude chat and Cowork into one Enterprise experience on October 23, 2026, so every conversation starts as a chat and moves into an isolated cloud sandbox when a task needs code or files. The switch is off by default and turns on automatically on November 16, 2026, for organizations that already have chat and cloud code execution enabled. Enterprise costs $20 per seat per month billed annually, with a 20-seat minimum, and AI usage is billed separately at Claude API rates.

What actually changed on October 23

Cowork was Anthropic's name for the agentic side of Claude: hand Claude a multi-step job, and it would plan the work, run code, work through files, and come back with a result. Chat was the other side: ask a question, get an answer. The October 23 change deletes that choice for Enterprise. One conversation can now do both, sliding from a quick answer into a cloud session and back.

Enterprise is the last major tier to get the merge. Pro and Max users have had the same combined experience since September 16, 2026. The Enterprise release is the one that matters most for governance, because it is the tier where an AI that can run code unsupervised touches company files — and where the admin controls, not the interface, decide what is allowed.

One thing did not merge: Claude Code stays separate, with its git integration, test running, and CI hooks. The unified experience is for the work around the code — documents, analysis, reporting, file wrangling — not for repository work.

How the unified experience works

The flow Anthropic describes is simple, and the simplicity is the point. You type a question. If the answer is a paragraph, you get a paragraph. If the task needs a file read, a script run, or a document built, Claude continues the same conversation in a cloud session — an isolated virtual machine on Anthropic's side — and returns the finished work for review. Sessions keep running after you close your laptop, because the work lives in the cloud.

That "isolated" wording is Anthropic's own description of the sandbox. Each agentic conversation gets its own machine, and the sandbox's network access is off by default for Enterprise organizations — admins can open it to package managers only, a custom domain list, or all domains, but the default is closed. Connector calls, the way Claude reaches your other tools, are enforced by Anthropic outside the sandbox on every call, under the organization's connector policies.

There is no independent test of the sandbox isolation published yet, and Anthropic has published no benchmark for how often a long task goes off the rails. That is normal for a three-day-old enterprise feature, but the honest way to read the launch is as a capability with guardrails, not a proven system.

How to turn it on — or keep it off

The setting lives in Organization settings > Capabilities, under "Chat and Cowork unified." For existing organizations it is off by default. Owners and Primary Owners see a note reading "Will be automatically enabled on November 16, 2026." From there, the choices are: Enable now, or Keep disabled. If the setting was already turned on or off at some point, that choice stays — the November 16 flip only applies to organizations that never touched it.

There is a catch that Anthropic states plainly in its own setup guide. The unified experience needs two capabilities on: Chat, and Cloud code execution and file creation. If either one is off, nothing turns on on November 16, and — in Anthropic's words — there is no notice. After November 16, the experience turns on as soon as both are on. So an organization that enables cloud code execution for another reason in December will find the unified experience arriving with it, with no fresh prompt.

For a controlled rollout, roles are the lever. Once the organization setting is on, everyone with the User, Admin, Owner, or Primary Owner role has access. To limit it, move people who should not have it to a custom role with "Chat and Cowork unified" turned off before enabling it. Access is checked on every request, so removing a role or turning the setting off cuts access immediately.

Three groups cannot get it at all yet: organizations using zero data retention, the HIPAA-ready configuration, or customer-managed encryption keys. Anthropic lists the exclusion in its setup guide without a date for when it lifts.

What it costs: the $20 seat and the usage catch

Cowork was never sold as a separate product, and the merge adds no new line item. You pay for a Claude plan, and the capability comes with the tier. The part that catches budgets is the same one that has always applied to agentic work: a long, multi-step task consumes far more of a plan's usage allowance than a chat, because every step is its own model call. Anthropic publishes no multiplier, so the only honest way to plan is to watch usage after enabling it.

Enterprise pricing, from Anthropic's own pricing page, is $20 per seat per month billed annually, with a 20-seat minimum for the self-serve tier. That is $4,800 a year in seat fees for a 20-person organization before a single token is used. Usage is billed separately at standard Claude API rates — the tokens are the variable part, and on agentic work they are the part that moves.

The API rates that set the usage bill, from Anthropic's API pricing documentation, are $2 per million input tokens and $10 per million output tokens for Claude Sonnet 5.5, $0.10 and $0.50 for Claude Haiku 5.5 on prompts up to 100,000 tokens ($0.25 and $1.25 above that), and $4 and $20 for Claude Opus 5.5. Prompt caching softens the input side: a cache hit costs 5% of the standard input price on Sonnet 5.5 and Opus 5.5 — $0.10 and $0.20 per million tokens. Those are the rates a long Cowork session draws down, call by call.

Anthropic's answer to the cost question is Efficient mode. When it is on, organization members can use only Sonnet and Haiku models in the unified experience, and new sessions start on Sonnet 5.5 at low effort. Owners turn it on for the whole organization on the Models page, or for specific custom roles. The trade is plain: cheaper sessions, with the strongest model — Opus 5.5 — off the table for unified work while the mode is on.

One accounting detail matters for anyone reading the invoice. Usage from the unified experience gets its own line in reports, under "Chat and Cowork unified," and in the Cost and Usage API under the product value chat_cowork_unified. Anthropic is explicit that this is only a filing change: moving usage to the new line does not change how it is counted or billed.

The admin controls that matter before November 16

The merge is an interface change, but the controls around it are the real story for Enterprise admins — and they now govern a system that can keep working while the user is away.

Network egress is the first control to check. It is off by default for Enterprise, and the sandbox cannot reach the internet until an admin allows package managers only, a custom domain list, or all domains. Approvals are the second: admins can turn off "Skip all approvals" mode and "Automatically approve" mode, which stops users from letting Claude act without asking — including using tools, editing files, and browsing — and the restriction applies to scheduled tasks and the built-in browser. Sharing is the third: a conversation that has used its sandbox can be shared only when both "Share chats" and "Share cloud sessions" are on, and the more restrictive setting applies.

The context is not theoretical. PYMNTS Intelligence research cited in its coverage found that 57.1% of companies with at least $1 billion in annual revenue reported incidents or losses involving adversarial bots and agents, against 32.4% of companies with $50 million to $250 million in revenue. That is PYMNTS's own research, not an industry-wide count, but it explains why the sandbox settings, not the chat interface, are where an admin's time goes this month.

Retention and compliance carry over unchanged: custom data retention settings govern conversations, projects, and artifacts, the Compliance API covers the unified experience with existing keys, and SSO, SCIM provisioning, and tenant restrictions still apply, with no separate sign-in. What the merge adds is a reason to review connector permissions before November 16 — a connector approved for chat answers becomes a connector a cloud session can call while nobody is watching. It is the same governance question as Walmart letting AI agents shop with spending limits built in: the agent is useful exactly to the extent its permissions are bounded.

Claude Enterprise vs Microsoft 365 Copilot vs ChatGPT Enterprise

Every major vendor is now selling the same idea — one assistant that answers questions and does multi-step work — and every one of them has moved the heavy agentic features to usage-based billing on top of a seat price. The seat prices look close. The totals are not, because the base license and the usage meter differ.

Two caveats sit under that table. The ChatGPT Enterprise figures are reported, not official — OpenAI does not publish Enterprise pricing. The Microsoft all-in figures come from independent cost analysis, not Microsoft, and enterprise agreements commonly cut 20 to 40 percent off the $30 list price. The Claude figures are the exception: the $20 seat and the API rates are on Anthropic's own pages. One more difference matters: Microsoft's Copilot Cowork went generally available on June 16, 2026, and its usage-based billing stays off by default until an admin creates a spending policy — Microsoft leaves its agentic meter off until an admin opts in, while Anthropic turns its unified experience on unless an admin opts out. That is why the Claude date matters more than the Copilot one.

Pros and cons

What is genuinely good: one conversation for questions and tasks removes a real piece of friction, cloud sessions that continue after a laptop closes are useful for long work, and the admin controls are unusually complete for a launch: egress off by default, approval modes that can be locked down, role-based access checked on every request, and existing SSO, retention, and Compliance API coverage carrying over without a new setup. Efficient mode gives cost-conscious organizations a real lever.

What to watch: the November 16 default flip fires silently for organizations that never touch the setting, and enabling cloud code execution later switches the experience on with no notice. Agentic sessions burn usage allowance far faster than chat, and Anthropic publishes no multiplier, so the first bill after enabling it is a discovery exercise. Zero data retention, HIPAA-ready, and customer-managed-key organizations are excluded with no date for inclusion, and there is no independent verification yet of the sandbox isolation or of long-task reliability.

Your pre-November 16 checklist

First, open Organization settings > Capabilities and look at "Chat and Cowork unified." If it says it will be automatically enabled on November 16, 2026, decide now: enable early and test with a pilot group, or keep disabled and revisit on your own schedule. A choice already made stays made; only the untouched setting flips. Second, check the two prerequisites — Chat, and Cloud code execution and file creation — and understand that turning the second one on after November 16 also turns on the unified experience. Third, set network egress deliberately, and decide the approval modes: whether users may skip approvals or auto-approve actions, including in scheduled tasks. Fourth, move anyone who should not have the experience to a custom role with the capability off before you enable it, and review the connector policies. Fifth, turn on Efficient mode if cost matters more than having Opus 5.5 in the unified experience, and tell finance that usage will now appear under a new "Chat and Cowork unified" line without changing how it is billed.

FAQ

Does the merge change what I pay? No. There is no Cowork add-on fee and no new price. Enterprise stays at $20 per seat per month billed annually with a 20-seat minimum, and usage is billed separately at API rates. What can change is how much usage you consume — agentic sessions run many more model calls than a chat.

What happens to my existing chats and Cowork tasks? They carry over. Scheduled tasks and session history keep working, and nothing needs to be recreated.

Can I keep chat and Cowork separate? You can keep the unified experience disabled — select Keep disabled and it stays off. But the direction is set: Pro and Max users have had the merged experience since September 16, 2026, and Anthropic says the unified experience is meant to replace the separate settings over time.

Why can't some organizations get the unified experience? Organizations using zero data retention, the HIPAA-ready configuration, or customer-managed encryption keys are not eligible yet. Anthropic gives no date for when the exclusion lifts.

Does Claude Code merge into the unified experience too? No. Claude Code stays outside it, with its own git integration, test running, and CI support. The unified experience covers the work around the code — documents, analysis, and file tasks — not repository work.

Key takeaways

• Anthropic merged Claude chat and Cowork for Enterprise on October 23, 2026 — one conversation that starts as a chat and moves into an isolated cloud sandbox when a task needs code or files.
• The switch is off by default but flips itself on November 16, 2026, for any organization with chat and cloud code execution enabled — and enabling cloud code execution later turns it on with no notice.
• Enterprise costs $20 per seat per month billed annually with a 20-seat minimum, and usage is billed separately at API rates — Sonnet 5.5 at $2/$10 per million tokens, Haiku 5.5 at $0.10/$0.50.
• The admin controls are the real story: egress off by default, approval modes that can be locked down, role-based access checked on every request, and existing SSO, retention, and Compliance API coverage carrying over.
• Zero data retention, HIPAA-ready, and customer-managed-key organizations are excluded for now, Claude Code stays separate, and there is no independent verification yet of the sandbox.

Anthropic gave Enterprise admins three and a half weeks between the October 23 announcement and the November 16 default flip. The merge itself is a small interface change; the decision it forces — what an AI agent may do, unsupervised, with your organization's files — is not.