Claude's Cowork Agent Reaches Web, Mobile, and Government

Anthropic has taken Claude Cowork, its AI coworker that runs multi-step tasks on a user's behalf, off the desktop and onto the web and mobile. The company also gave the agent write access to Microsoft 365 and launched a version cleared for government use. Taken together, the changes push Claude further into the everyday office work that sits well beyond writing code.

The desktop-only limit had been the awkward part of Cowork since launch. As TechCrunch put it, the coding-agent wars are now spilling into the rest of the office, and an assistant you can only reach from one machine does not fit how most people actually work.

Sessions that follow you around

With the update, a Cowork session and its files now move across devices. You can start something on a laptop, approve a step from a phone, and let the agent keep working in the background on scheduled tasks. Shared chats and projects carry over too, so the same work is visible wherever you pick it up. The rollout starts with the Max plan and reaches other tiers over the coming weeks, according to 9to5Mac.

The more consequential addition is write access to Microsoft 365. Until now an assistant like this could read your email and calendar and suggest what to do. With write tools switched on, Claude can draft and send messages, organize a mailbox, manage calendar events, and create or update files in OneDrive and SharePoint. That is a real change in what the agent is trusted to touch. Anthropic has put a gate in front of it: a Microsoft Entra administrator has to consent to the new permissions and enable them for the organization before anyone can use them.

A separate track for agencies

Alongside the consumer rollout, Anthropic put Claude Code and Claude Cowork into public beta inside Claude for Government Desktop, running in a FedRAMP High authorized environment. The government build adds tamper-evident audit logs, stronger admin controls and spend governance, the sort of oversight a public agency needs before it lets an agent act on its own.

That continues a run of public-sector deals for the company, after California put Claude into its state agencies earlier this summer. It also lands at an awkward moment abroad, with Anthropic still working through a public rift with Alibaba over Claude Code.

The pattern across all of it is the same. Anthropic is widening what Claude is allowed to do without a human watching each step, and pairing that reach with permission gates and audit trails. How comfortable people are with an agent sending their email and editing their files, rather than just suggesting the edits, is the question the next few weeks will start to answer.