Claude Cowork expands to mobile and web, enabling cross-platform asynchronous task execution.

What happened

Anthropic has expanded Claude Cowork beyond its initial desktop constraints, introducing full mobile and web support. Crucially, this update introduces asynchronous task persistence. Users can now initiate a complex task on one device, close the local client, monitor progress via mobile notifications, and retrieve the completed output later from any endpoint.

Technical details

Under the hood, this shifts Claude Cowork from a synchronous, client-dependent execution model to a decoupled, server-side asynchronous architecture. Previously, closing a laptop or dropping a websocket connection could interrupt or kill long-running generation or analysis tasks. Now, the context state, memory, and execution loop are maintained entirely server-side. The mobile and web clients essentially act as thin interfaces, polling or receiving push notifications for state changes in the background agent's lifecycle.

Why it matters

From an engineering standpoint, this is a massive infrastructure and usability upgrade for agentic workflows. As AI models are increasingly tasked with long-running operations—such as parsing massive codebases, generating multi-file projects, or conducting multi-step web research—the requirement to keep a local session alive was a significant architectural bottleneck. By pushing the execution state entirely to the cloud and making it accessible via multi-device endpoints, Anthropic is treating Claude less like a synchronous chatbot and more like a background compute job. This mirrors how developers already treat CI/CD pipelines or background worker queues.

What to watch next

Watch for how Anthropic handles state collision and concurrent interactions if multiple clients attempt to steer an ongoing Cowork task. Additionally, this infrastructure paves the way for more autonomous agentic behaviors, such as cron-scheduled tasks, webhooks for third-party integrations, and automated background processing. Expect competitors like OpenAI and Google to rapidly match this persistent, cross-platform execution model for their own agent frameworks.