Anthropic has folded chat and Cowork into a single Enterprise surface. The unified Claude experience, generally available for Enterprise plans since October 23, means teams no longer choose a mode before they start work: every conversation opens as a chat, and when a job needs code run or files handled, the same thread continues in an Anthropic-managed cloud sandbox and returns finished work for review.
Pro and Max subscribers received the combined experience in September; Enterprise, where procurement, compliance, and admin control decide everything, is where the consolidation either earns its keep or stalls. Anthropic has put a date on that verdict: existing organizations that already have chat and cloud code execution switched on will move to the unified experience by default on November 16, unless an owner actively keeps it disabled.
The practical change is the removal of a decision users used to make up front. Previously, a team member picked chat for questions and Cowork for delegated work. Now the conversation itself decides: a quick question gets a quick answer, and a request that needs files built, code run, or documents assembled escalates into a cloud session without starting over or switching apps.
Continuity is the selling point Anthropic leans on hardest. Because the agentic work runs in the cloud rather than on the employee's machine, it keeps going after the laptop closes, and the same conversation can be picked up on the web, desktop, or mobile. Chats, projects, connectors, skills, and plugins carry over into the new experience, so nothing needs rebuilding.
What is actually new here is less the chat interface than the execution model behind it. Each agentic conversation runs in its own isolated virtual machine managed by Anthropic. For Enterprise organizations, network access from that sandbox is off by default, and connector credentials are kept outside it — so a task that touches company systems does so through controlled channels rather than from inside an open environment.
That architecture is aimed squarely at security teams. It builds on the enterprise controls organizations already use — single sign-on and the Compliance API — and Anthropic says data stays encrypted in transit and at rest and is not used for model training by default. The pitch, in short: instead of agents running loose on employee laptops, agent work happens in one managed place where IT can see it and set rules around it.
Admins control the rollout by role, deciding which members get the unified experience, and can limit model choices and spending. A new Efficient mode confines the unified experience to Sonnet and Haiku models, giving cost-conscious organizations a way to cap usage without banning the tool outright. Pricing stays usage-based, with organization, group, and individual spend limits alongside it.
The controls matter because agentic work is inherently open-ended: a task that runs in the cloud, unattended, can consume far more tokens than a chat reply. Efficient mode is Anthropic's answer to the finance team's first question, and the role-based rollout is its answer to the security team's.
The rollout has a clear boundary, and it is worth stating plainly: organizations using zero data retention, the HIPAA-ready configuration, or customer-managed encryption keys do not get the unified experience yet and stay on their current setup. Anthropic's own setup guide lists those three exclusions, and they cover exactly the regulated customers who would otherwise be first in line.
There is a second, quieter prerequisite. The November 16 default applies only where both chat and cloud code execution are enabled; if either is off, nothing turns on, and the experience activates later as soon as both are on. Owners can also enable early or keep the experience disabled. Admins weighing the switch should review connector permissions, network egress, and approval settings before enabling work that can continue while its owner is away.
Anthropic's move is best read as infrastructure, not a feature launch. The company is betting that the enterprise agent era will be decided by who runs the work, who controls it, and who pays for it — and it has now put all three answers in one place.