Anthropic prepares Cowork support for mobile apps
Anthropic is preparing to shift its Cowork agentic system to the cloud, allowing tasks to run independently of the user's desktop hardware.
Deep dive
- Anthropic is extending its agentic 'Cowork' system to mobile platforms.
- Feature flags in the iOS build reveal a centralized dashboard for scheduling and viewing task results.
- The update moves away from the 'Dispatch' model, which required the user's computer to remain awake to run local tasks.
- Cloud-based execution will allow agents to operate independently of user hardware.
- Code analysis suggests an impending model update for Claude's voice mode, likely moving away from Haiku 4.5.
Decoder
- Cowork: Anthropic's agentic framework designed to automate knowledge work tasks.
- Agentic system: Software capable of autonomous decision-making and task execution to achieve user-defined goals.
- Dispatch: An earlier Anthropic feature that enabled mobile interaction with desktop-based AI agents.
Original article
Anthropic appears close to extending Cowork, its agentic system for knowledge work, beyond the desktop. A build of the iOS app carries a Cowork entry, gated behind a feature flag, that would surface in the side navigation. Tapping it reveals copy about scheduling Cowork tasks from a phone and picking up results across mobile, web, or desktop, plus a tab that gathers every scheduled action in one place.
Users will be able to trigger Cowork tasks on mobile and view scheduled tasks in the app.
That framing points to a shift rather than a debut. Cowork already reached phones in March through Dispatch, which lets someone message a desktop session remotely, but only while the computer stays awake, since the work runs locally. The new wording suggests moving execution to the cloud and the web, lifting Dispatch's central constraint so tasks can run without a machine left on. The copy is already written, hinting at a release as soon as this week, though nothing is operational yet.
A second finding concerns voice. New consent text in the app sources accompanies a model selector for voice mode, letting users choose the model behind the spoken experience. Claude Voice has leaned on Haiku 4.5 for a while, so the selector reads as groundwork for an underlying model refresh, on top of the multilingual rollout already underway across accounts.
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi