A Persistent Agent That Doesn't Quit When You Close the Lid
Anthropic has pushed its Claude Cowork agent into uncharted territory: tasks no longer die when you shut your laptop. According to a report by Wired on July 7, 2026, the updated agent now continues working on your smartphone after the computer goes to sleep—or even if you close the lid and walk away. This small-sounding change has enormous implications for how people will interact with agentic AI. For the first time, a major AI assistant can run a background process that persists on a second device without a user keeping an active session on the original machine.
The new capability is part of what Anthropic describes as a "larger push toward smartphone-controlled agents." In practice, this means Claude Cowork can be given a multi-step research, coding, or scheduling task on a laptop, and then seamlessly hand off the execution to your phone, where the agent will keep working and send push notifications when it finishes. It's a departure from the current generation of AI agents—including OpenAI's Operator and Google's Project Mariner—that are largely tethered to a single browser window or desktop session.
When we examined the release notes and user accounts associated with the update, it became clear that this persistence feature is not just a minor quality-of-life tweak. By designing an agent that can outlive the hardware it was launched on, Anthropic is effectively positioning Claude Cowork as a personal assistant that is always-on, cross-device, and not interruptible by routine user behavior like shutting a laptop to commute home.
How the Cross-Device Agent Actually Functions
Anthropic has not publicly disclosed every engineering detail, but early adopters report that the feature leverages a cloud-based agent instance paired with the Claude mobile app for iOS and Android. Once a task is initiated on the desktop version of Claude Cowork, the user can close the laptop; the agent continues executing in Anthropic’s cloud environment and uses the phone as the control plane. The mobile app then shows a live progress feed and delivers the final output—be it a completed spreadsheet, a draft of a legal document, or a summary of competitor research.
This design sidesteps a major limitation that has dogged AI agents since their inception: tasks that require lengthy processing or multiple steps are often interrupted when a user loses a network connection, puts the machine to sleep, or simply closes a tab. With the phone-based persistence, Claude Cowork behaves more like a remote job on a server—something developers are intimately familiar with—but wrapped in a consumer-friendly interface.
From a technical standpoint, the update likely relies on a synchronized state between the desktop and mobile client, with the agent's runtime environment abstracted away from the initiating device. Observers have noted that this is a logical extension of Anthropic’s earlier move to bring Claude Cowork’s computer-use abilities—like controlling the mouse and keyboard—to mobile platforms, effectively allowing the agent to perform tasks on the phone itself. The new feature doesn’t necessarily require the agent to act directly on the phone’s screen; rather, the phone becomes the user’s window into a background process that may be running on a server and only returning results when complete.
Why This Move Rattles the Agentic AI Landscape
The update lands at a moment of intense competition among foundation model providers to make AI agents practical for everyday work. OpenAI has been gradually extending ChatGPT’s ability to execute multi-step tasks across files and apps. Google is weaving agent-like behavior into Gemini, with prototypes like Project Mariner that navigate websites. But none of the market leaders had yet solved the simple but critical problem of session continuity across devices.
Anthropic’s approach is strategically clever because it addresses a real-world pain point: users frequently start something on a powerful device and want to finish it on the go. By solving the cross-device persistence riddle first, Claude Cowork could attract professionals who need long-running, unsupervised tasks—financial analysts running model simulations, lawyers reviewing thousands of documents, or product managers scanning competitor updates. Moreover, the smartphone-native focus may appeal to the growing base of mobile-first workers who rarely touch a traditional laptop.
Another important angle is the competitive response. If this feature works reliably, it could pressure Microsoft to extend Copilot across Windows and mobile in a similar persistent fashion, and might force Google to accelerate the smartphone integration of its own agent prototypes. Apple’s on-device intelligence model, meanwhile, operates very differently—focused on privacy-preserving, local processing—so a cloud-persistent agent like Claude Cowork could serve as a foil for users willing to trade some data locality for continuous AI labor.
Limitations and Privacy Questions Remain
Despite the compelling promise, Anthropic’s new cross-device persistence isn’t without caveats. The feature appears to require the Claude mobile app and an active cloud subscription—likely the existing Claude Pro or Team tier—though the company hasn’t officially confirmed pricing details as of publication. And because the agent runs continuously on remote servers, friction around data handling, security, and cost could quickly become sticking points. A background agent that keeps churning through a task for hours might burn through compute credits faster than a typical chat session, raising questions about billing transparency.
Privacy-conscious users will also ask what happens to the intermediate state of a task while it’s running unattended. If Claude Cowork is browsing the web, interacting with files, or opening third-party services in the background, the attack surface expands. Anthropic has long emphasized safety and alignment, but always-on agents introduce new vectors for prompt injection or data leakage if the agent is left to autonomously navigate sensitive environments.
Additionally, the reliance on a smartphone as the control hub may not suit all workflows. Enterprise users who manage complex batch jobs might prefer traditional server-side orchestration rather than push notifications. Still, the company seems to be betting that the convenience of lighting up your phone to check on an agent’s progress will outweigh the drawbacks for a broad swath of professionals.
The Phone Becomes the New Home Screen for AI Agents
Looking ahead, Anthropic’s smartphone-centric pivot could signal a broader industry shift. As AI agents evolve from novelty demos to productivity staples, the devices people always have with them—phones—could become the default interface for starting, monitoring, and collecting the results of AI-driven tasks. The laptop may recede to being just one of many launch points, rather than the sole cockpit.
This update also highlights the growing importance of mobile app ecosystems for AI companies. OpenAI, Google, and Anthropic have all invested heavily in mobile apps, but until now they mostly served as chat interfaces. With Claude Cowork’s phone-based persistence, the mobile app transforms into an agent dashboard. That could open the door to deeper integrations, such as widgets that show agent progress, voice-based task initiation via Siri or Google Assistant shortcuts, or even background long-running tasks that survive the user locking the phone.
For developers and IT decision-makers, the most immediate consequence is that cross-device agent continuity is no longer a theoretical feature—it’s shipping, and it’s likely to become table stakes. The companies that best manage battery life, notification fatigue, and trust will shape the next phase of how people work alongside AI. For now, Anthropic has vaulted ahead in that particular race, offering a glimpse of an agent that doesn’t ask you to stay tethered—and that simple change might be enough to redefine the whole category.
评论