Claude Cowork Adds Mobile and Web Access for Cross-Device Workflows
Anthropic's Claude Cowork collaborative agent workspace now runs on mobile and web, letting users hand off long-running tasks across devices without keeping a laptop open. The expansion targets knowledge workers who need to delegate work to AI agents and check in on results asynchronously.
Original sourceAnthropic has expanded Claude Cowork — its collaborative agent workspace — to mobile and web platforms, completing a cross-device loop that previously required a desktop session to stay alive. The core pitch is continuity: start a multi-step research or drafting task at your desk, monitor its progress from your phone on a commute, and return to finished output whenever you're ready, regardless of whether the originating machine is still running.
The timing reflects a broader maturation in how agentic AI tools are being designed. Early agent products assumed a user would sit and watch a task run. Claude Cowork's architecture separates task initiation from task monitoring and output retrieval — a meaningful structural choice that implies background execution infrastructure rather than a simple session-based model. Whether that infrastructure is genuinely async or just a long-lived server session isn't clear from the announcement.
The feature set positions Cowork against a growing cluster of agent orchestration tools including Notion AI's workflow features and Microsoft's Copilot Tasks, as well as more developer-oriented offerings like Devin and similar coding agents. Anthropic's differentiator appears to be the integration with Claude's broader reasoning capabilities, though the company has not released benchmarks comparing task completion quality across interruption scenarios.
For knowledge workers managing tasks that take minutes to hours — synthesizing research, drafting long documents, running analysis pipelines — device-agnostic access removes a real friction point. The question the product still has to answer is how gracefully it handles task failures, mid-run clarifications, and output editing once a user picks back up. Asynchronous delegation is easy to demo; the edge cases are where agent products either earn trust or lose it.
Panel Takes
The Builder
Developer Perspective
“The primitive here is async task delegation with cross-device state persistence — that's actually a hard infrastructure problem if it's real and not just a long-lived websocket with a mobile UI bolted on. The DX bet is that users want to delegate and disconnect, which is the right call, but I'd want to see what the handoff model looks like when a task needs mid-run input before I'd call this solved. If the 'mobile monitoring' is just a read-only status screen with no structured intervention surface, they've built a pager, not a coworker.”
The Skeptic
Reality Check
“The scenario they're selling — start at desk, check phone, pick up output later — is real and the friction is real, but the announcement is suspiciously light on what actually executes in the background and where the failure modes land. Every agent product demos beautifully on a linear happy path; what kills Cowork in 12 months is that Anthropic's own API roadmap ships persistent background tasks as a first-class primitive and this wrapper loses its reason to exist. The moat here isn't the mobile app, it's whether the orchestration layer has enough proprietary workflow logic that users can't just replicate it with a Claude API call and a push notification.”
The PM
Product Strategy
“The job-to-be-done is finally crisp: delegate a long-running task to an AI agent and retrieve the output without babysitting it. That's one sentence, no 'and' required — which is a good sign. The product completeness question is whether the mobile experience covers the full lifecycle: task creation, progress monitoring, output review, and crucially, mid-task steering when something goes sideways. If mobile is read-only and steering requires going back to desktop, users will keep a laptop tab open anyway and the whole value prop collapses.”
The Futurist
Big Picture
“The thesis here is falsifiable: within two years, the primary mode of AI interaction for knowledge work shifts from synchronous conversation to asynchronous delegation, and the interface layer that owns that handoff owns the relationship. Cowork is early to that specific bet — most competitors are still building chat-first with agent features appended. The second-order effect that nobody is talking about is calendar and attention management: if your AI agent works while your laptop is closed, the constraint on throughput stops being AI capability and starts being the human's ability to review and redirect output, which is a completely different product problem to solve next.”