Anthropic’s Claude Cowork has taken a step that feels small on the surface—web and mobile access for Max subscribers—but it signals something bigger about how “coding agents” and other AI work assistants are evolving. Until now, Cowork largely lived where most people already do their most focused work: on a laptop. That made sense for an agent that needs a stable environment, access to local context, and a place to review and iterate on outputs. But it also created a practical bottleneck. If you started a task and then had to leave your desk, the agent’s momentum was tied to the device you launched it from.
With the latest update, Claude Cowork is available on web and mobile for Max subscribers, allowing users to initiate work from one device, monitor progress from another, and return later to retrieve results—even if the original laptop is closed. That shift changes the rhythm of using an AI assistant. Instead of treating the agent like a chat window you must keep open, it becomes closer to a background workflow you can hand off and check on when it matters.
For readers who have watched the “agent” category mature over the last year, this is a familiar pattern: the technology improves, then the product catches up with how people actually operate. Most knowledge workers don’t spend their day tethered to a single screen. They move between meetings, calls, quick checks, and deep work blocks. The difference between an assistant that works only while you’re actively at your laptop and one that can continue across devices is the difference between “cool demo” and “daily tool.”
What’s new, specifically, is the ability to start a task from your desk and receive status updates on your phone. That matters because it addresses the biggest psychological friction in agent-based workflows: uncertainty. When you launch an automated process, you want to know whether it’s still running, stuck, or ready for review. Phone notifications turn that uncertainty into something manageable. You don’t have to keep a tab open or repeatedly check your computer. You can step away and still stay informed.
Then there’s the second half of the experience: picking up the finished output later. The claim that you can retrieve results even if your laptop is closed is important because it implies that the system isn’t dependent on a live session on that device. In other words, the agent’s work is decoupled from the physical machine you used to start it. That’s a subtle but meaningful architectural and product shift. It suggests that Cowork is moving toward a model where tasks are orchestrated in a way that persists beyond a single client session, which is exactly what you’d need for reliable “start now, review later” workflows.
This is also where the “coding agent wars” framing from TechCrunch becomes more than just a catchy phrase. The competition among AI coding assistants has increasingly moved from raw capability—can it write code?—to operational usefulness—can it manage work end-to-end in a way that fits real schedules? Many tools can generate code snippets. Fewer can reliably carry out multi-step tasks, track progress, and deliver outputs you can trust enough to integrate. And even fewer can do all of that without forcing you to remain present on the same device for the entire duration.
By expanding Cowork to web and mobile, Anthropic is effectively reducing the cost of interruption. That’s a big deal for developers and technical teams, because interruptions are not rare—they’re constant. A typical day includes context switching: you might start a refactor, get pulled into a meeting, then come back to decide whether the changes are correct. If the agent’s work is paused or lost when you close your laptop, you end up either delaying the task until you have uninterrupted time or accepting that you’ll have to babysit it. Cross-device continuity makes it easier to treat the agent as a collaborator rather than a process you supervise.
There’s also a broader implication for how people will think about “ownership” of the work. When an assistant runs only in the foreground, the user feels like they are driving the process. When it runs in the background and reports back later, the user shifts into a reviewer role. That change in mental model can improve outcomes. Reviewers tend to be more deliberate: they look for correctness, edge cases, style consistency, and integration details. If the agent can handle the heavy lifting while you’re away, you can spend your attention where it’s most valuable—on decisions, not on waiting.
The web and mobile surfaces also hint at a future where AI work becomes more event-driven. Status updates on a phone are essentially a lightweight interface for task lifecycle events: started, running, awaiting input, completed, or failed. Even if the current implementation is simple, it sets the stage for richer interactions. For example, a future version could allow users to approve changes, request clarifications, or adjust priorities without returning to a laptop. That would further reduce friction and make the assistant feel less like a tool you “use” and more like a system that “keeps working.”
From a product standpoint, this expansion is likely to broaden the audience for Cowork. Max subscribers who don’t live in a developer environment all day—designers, analysts, operations staff, students, and general knowledge workers—may find it easier to adopt an agent when they can interact with it from wherever they are. Coding agents have often been marketed to developers, but the real value proposition is broader: automating structured work, turning instructions into artifacts, and iterating until the output meets a goal. Those are tasks that happen across roles, not just in IDEs.
At the same time, the move to web and mobile raises questions that power users will naturally ask. What exactly is the scope of tasks that can be started from mobile? How much control does a user have over the environment? Are there limitations on what the agent can access depending on the platform? How are long-running tasks handled if connectivity drops? While the announcement emphasizes the ability to start, receive updates, and retrieve outputs later, the deeper value will depend on reliability and transparency—whether users can trust that the agent is doing what it claims, and whether they can inspect intermediate steps when needed.
Even without those details, the direction is clear: Anthropic is aligning Cowork with the reality that modern work is distributed across devices. Laptops are still the primary workspace for editing and reviewing code, but phones are where people receive signals and make quick decisions. Web is the bridge between them. By placing Cowork in that ecosystem, Anthropic is making it easier for users to fit agent workflows into their day without sacrificing focus.
There’s also a subtle competitive angle here. Many AI assistants have tried to win by being “everywhere,” but the difference between being everywhere and being useful everywhere is the persistence of the work. A chat interface on mobile is not the same as an agent that continues running and produces an output you can pick up later. The latter requires careful handling of state, task management, and user notification. If Cowork delivers on that promise consistently, it becomes more than a convenience feature—it becomes a workflow upgrade.
Consider the typical sequence of using a coding agent today. You might prompt it to implement a feature, fix bugs, or generate tests. Then you wait. If the tool is limited to a single device session, you either keep your laptop open or you risk losing progress. With cross-device support, you can start the task, let it run, and then return when you’re ready to review. That reduces the “attention tax” of using an agent. Attention tax is the hidden cost of many AI tools: even when the assistant does the work, the user still spends time monitoring, checking, and managing the interaction. Lowering that tax is what turns experimentation into routine use.
This also changes how teams might coordinate around AI work. In a team setting, someone could kick off a task during a lull, then share the output when it’s ready. Status updates could help coordinate timing without requiring constant messaging. While the announcement is framed around individual Max subscribers, the underlying workflow improvements are the same ones teams care about: asynchronous progress, predictable delivery, and reduced dependency on a single workstation.
Another interesting angle is how this affects learning and skill development. Developers often use agents to accelerate tasks, but they also learn from the outputs and the reasoning behind them. If the agent can run while you’re away, you can allocate your return time to deeper review rather than shallow monitoring. That can lead to better feedback loops: you can examine the changes, understand why they were made, and then refine prompts or constraints based on what you see. Over time, that can improve both the quality of the agent’s future outputs and the user’s ability to direct it effectively.
Of course, the success of this kind of product update depends on the quality of the user experience. Status updates must be timely and accurate. The retrieval of outputs must be seamless. The system should make it easy to resume context—so you don’t have to reconstruct what you asked for when you come back. If the interface forces users to hunt for the right task or re-enter details, the benefit of cross-device continuity shrinks quickly.
That’s why the emphasis on “even if your laptop is closed” is more than marketing language. It’s a promise about continuity and resilience. Users want to know that starting a task is not a fragile act. They want to be able to close their laptop, walk away, and trust that the agent will finish. In a world where many digital workflows fail silently—downloads interrupted, sessions expired, tabs closed—reliable persistence is a differentiator.
Looking ahead, this update also suggests that Anthropic is thinking about Cowork as a long-running service rather than a single-session tool. Once you treat tasks as persistent objects with lifecycle states, you can build richer features on top: resumable work, multi-step approvals, audit trails, and possibly collaboration. Even if those aren’t part of this specific release, the product direction is consistent with that trajectory.
It’s worth noting that this kind of cross-device capability is particularly relevant