Anthropic has expanded Claude Cowork, its AI agent product, beyond desktop to mobile and web platforms. Until now, Cowork remained confined to the desktop application, limiting its reach and utility for users working across devices.
The expansion changes how the agent operates. Cowork can now continue executing tasks in the background even when the user closes their laptop. When the agent encounters a decision point or needs human input, it sends a notification to the user's phone, allowing them to respond from anywhere. This cross-device continuity addresses a core limitation of previous AI agent implementations, which often stalled when users switched devices or went offline.
The move reflects Anthropic's strategy to blur distinctions between Claude's Chat interface and Cowork, its more autonomous agent mode. Chat handles conversational interactions. Cowork handles multi-step task execution with reduced human intervention. By bringing Cowork to mobile and web, Anthropic signals confidence in the agent's stability and positioning it as a core product, not an experimental feature locked to desktop.
The timing matters. AI agents remain computationally expensive and require careful handling to prevent errors. Desktop-only deployment let Anthropic test the technology with power users who could handle occasional failures. Mobile and web availability suggests the product has matured enough for broader deployment, though Anthropic has not disclosed specific performance metrics or reliability thresholds that triggered the expansion.
This move also creates competitive pressure. Other AI companies like OpenAI with its agents mode and Alibaba with its open-source frameworks are racing to deploy usable agents across platforms. Device fragmentation has slowed agent adoption, partly because orchestrating tasks across phones and computers requires robust background systems and notification infrastructure most AI startups lack.
The practical impact remains to be seen. Users will now delegate longer workflows to Claude while retaining oversight through mobile checkpoints. Whether this reduces cognitive load or creates notification fatigue depends on