The most useful kind of AI is not always the one that gives the fastest answer. Sometimes it is the one you can hand a real task to, walk away from, and check back on later. That is the idea behind Claude Cowork, Anthropic’s agent-style workspace, which is now available on mobile and the web across Claude’s paid plans.

For Claude users, this is a meaningful change. Cowork began life as a more limited desktop experience, built around the idea that Claude could do more than respond to one prompt at a time. Now, a Pro subscriber can start a longer task from the web or phone, monitor it while away from a desk, and return to the finished work without treating the process like a conventional chatbot session. Anthropic says Cowork is available to Pro, Max, Team, and Enterprise customers, although Enterprise access can depend on whether an administrator has enabled it.

That sounds like a small platform expansion at first. In practice, it is a glimpse at where AI software is heading: away from a single chat box and toward a persistent work layer that follows users across devices.

The key difference is simple. In a standard Claude chat, a person asks for something, gets an answer, follows up, and repeats. Cowork is designed for a broader assignment: describe an end result, approve the approach, and let Claude work through the steps needed to get there.

Anthropic positions Cowork as bringing some of the agentic mechanics behind Claude Code to non-technical work. It can plan a task, split more complex work into smaller pieces, coordinate parallel workstreams where appropriate, and deliver outputs such as organized files, research summaries, formatted documents, spreadsheets, and presentations.

That makes Cowork less like asking an assistant to draft an email and more like delegating the first pass of a project. A content team might use it to turn scattered source notes into a structured briefing. A small business owner could ask it to sort receipts, create an expense report, and build a spreadsheet with formulas. A researcher could hand it a collection of documents and request themes, gaps, and a readable synthesis.

Of course, the word “agent” can make these systems sound more autonomous than they really are. Cowork still works within the permissions and tools the user grants it. But it can run for longer than a typical chat exchange, which changes the feel of the product. The point is not merely to generate text. The point is to get through a sequence of work.

The mobile and web rollout matters because long-running work does not naturally fit inside one device.

A task that starts in the office should not need to stay trapped on an office laptop. With Cowork’s cloud sessions, users can begin work on desktop, check progress on the phone, answer a question while traveling, and open the finished result in a browser later. Anthropic says these sessions and associated files are tied to the user’s Claude account, allowing them to move across desktop, web, and mobile.

That is the more practical promise behind the update. You might initiate a research task from a laptop in the morning, notice an issue in the afternoon from your phone, and redirect the work without reopening the original machine. For people who already work across Chrome, mobile apps, and desktop software throughout the day, that continuity may prove more valuable than another increment in raw model capability.

The rollout also gives Pro subscribers access to a feature that initially carried a distinctly power-user flavor. Claude’s paid plans begin with Pro, while the company also offers higher-usage Max tiers along with Team and Enterprise options. Making Cowork broadly available across paid subscriptions turns it into part of the everyday Claude product, rather than a premium experiment reserved mainly for heavy users.

Cloud work, with caveats

There is an important technical distinction in how Cowork works across devices. Its cloud sessions can keep going while the user steps away, closes a laptop, or lets a computer sleep. Anthropic says the agent’s work runs in an isolated environment on its servers, and that users can reopen the same session from another surface to watch progress or change direction.

But that does not mean a phone can magically access every file sitting on a home or office computer. For tasks involving local files, browser controls, or computer use, the Claude Desktop app still needs to be open and connected on the relevant Mac or Windows PC. In other words, cloud continuity handles the task itself, while the desktop app remains the bridge to local resources.

This is a sensible compromise, but it is also the kind of detail that will shape real-world expectations. Cowork on mobile is not a full replacement for desktop when the job requires direct access to a local folder, a browser session, or applications installed on a computer. It is more accurately a remote control and continuation layer for work that may still rely on a desktop connection.

For many use cases, that will be enough. The ability to check in, approve a next step, review an output, or redirect a task from a phone is exactly what turns a long process into something less tethered to a desk.

The permission question

The more capable an AI becomes, the more important it is to understand what it can access and act upon.

Anthropic acknowledges that Cowork has distinct risks because it can use the internet and, when permission is granted, interact with files and connected tools. The company says users can choose different approval modes, from reviewing actions manually to allowing more automated execution. It also requires explicit permission before Cowork permanently deletes files.

That is a welcome safeguard, but it should not be mistaken for a reason to switch off attention. The company itself advises users to stay close to tasks with real consequences, including financial activity, messages sent on a person’s behalf, or important files.

The practical rule is straightforward: use Cowork freely for repeatable, low-risk work, but treat sensitive material and irreversible actions with more caution. Ask it to organize a downloads folder or outline a report, sure. Give it access to confidential client data, payroll materials, or systems that can send messages and make purchases, and the review process should become much more deliberate.

This is especially relevant for businesses. The promise of AI agents is productivity, but the responsibility remains human. A polished document can contain mistakes. A plausible research summary can omit a crucial source. An automated workflow can move faster than the person supervising it. Cowork may reduce busywork, but it does not remove the need for judgment.

Anthropic’s expansion comes as major AI companies race to make their products more useful beyond a prompt-and-response loop. The competitive question is increasingly not just which model writes better or reasons longer. It is which platform can reliably help people complete work across their actual tools, files, devices, and routines.

Cowork is Anthropic’s answer to that question. Its scheduled-task feature, for example, can run saved work automatically or on demand in the cloud, without requiring a computer to remain online. It also supports project spaces, which group related tasks with their own files, links, instructions, and memory.

That combination begins to resemble a lightweight work operating system. The AI is not simply waiting for the next question. It has a task list, a workspace, a place for context, and a way to produce tangible outputs. The more friction Anthropic removes between chat, files, research, automation, and device switching, the more Cowork could become a habitual part of a user’s daily workflow.

There are still limits. Anthropic notes that general chat memory does not yet transfer into Cowork sessions, except through project-based memory, and that some features remain desktop-only, including live artifacts and plugins that use local MCP servers. Cowork sessions also cannot currently be shared with other people, though Team and Enterprise users have some organization-level collaboration options for live artifacts.

Those constraints matter because collaborative work is where many AI tools either become indispensable or hit a wall. A solo user may be happy to run a private research or file-management job. Teams often need visibility, shared context, approvals, and repeatable governance. Anthropic has the beginnings of that framework, but Cowork is still clearly evolving.

The real test

The headline is that Claude Cowork now runs on the web and mobile for all paid plans. The more interesting development is that Claude is becoming a tool users can hand work to, rather than merely talk with.

For journalists, marketers, analysts, founders, and anyone buried in documents and recurring research, that distinction is easy to understand. Chat is useful when you need a thought partner. Cowork aims to be useful when you need to get a messy, multi-step job moving.

Whether it earns a regular place in people’s workflows will depend on reliability, trust, usage limits, and how well it handles the awkward middle ground between automation and human oversight. Anthropic notes that Cowork consumes more usage than ordinary chat because its multi-step tasks require more computing resources, so users may need to reserve it for work where sustained execution genuinely adds value.

Still, giving every paid subscriber the ability to start on a laptop and keep tabs from a phone brings the idea of agentic AI closer to ordinary working life. The next phase of the AI race may not be about who has the cleverest answer. It may be about which assistant can take the work off your plate – and still let you stay in control.