All posts

Claude Cowork expands to mobile and web for Max subscribers

Huma ShaziaJuly 7, 2026 at 11:01 PM4 min read
Claude Cowork expands to mobile and web for Max subscribers

Key Takeaways

Claude Cowork expands to mobile and web for Max subscribers
Source: The Decoder
  • Claude Cowork now works across mobile, web, and desktop, with background task execution even when devices are off
  • Over 90% of Cowork usage is non-coding work, primarily business operations and content creation
  • Desktop app remains required for local file access, browser control, and Computer Use features

Anthropic is expanding Claude Cowork, its AI agent for autonomous task completion, to mobile and web platforms. The feature was previously desktop-only. Max subscribers will get beta access first, with a gradual rollout over the coming weeks.

The cross-platform launch solves one of the obvious friction points with AI agents: you had to stay near your computer while they worked. Now users can start a task at their desk, check progress from their phone, and pull up results in any browser. Claude continues processing even when the laptop is closed or the phone is off.

Advertisement

Human approval now happens on your phone

Anthropic built Cowork with a "human in the loop" design. When the agent reaches a decision point that requires human judgment, it pauses and asks. That checkpoint system now works via smartphone notifications. Nothing gets sent or finalized without user approval, according to Anthropic.

This matters for teams worried about AI agents acting autonomously. The mobile approval flow means you can monitor and authorize agent actions without being tethered to a workstation.

What tasks are people actually running?

Anthropic says over 90 percent of Cowork usage is not software development. That tracks with how the company positioned the product: a general-purpose agent for knowledge work, not a coding assistant.

The two largest categories are business operations and content creation. Anthropic cites examples like reconciling quarterly spend, drafting variance memos, and building client decks from call transcripts. Together, these two categories account for roughly half of all usage.

For product teams, this signals where AI agents are finding traction. The value is in automating repetitive knowledge work that crosses multiple data sources, not in replacing specialized tools.

Advertisement

Desktop still required for local access

The desktop app is not going away. Anthropic says it remains essential for features that need local machine access: reading and writing files in connected folders, local connectors and plugins, browser control through Claude in Chrome, and Computer Use (where Claude clicks, types, and navigates on screen).

Users who could not install the desktop app before can now try Cowork through a web browser for the first time, though without those local capabilities. On web and desktop, Chat and Cowork will share a single home screen going forward. Projects and artifacts sync across platforms.

To mark the expansion, Anthropic is extending its doubled Cowork usage limits through August 5.

The agent-chatbot convergence

The lines between Chat and Cowork are blurring. Anthropic seems headed toward merging them into a single product. OpenAI is doing the same with Codex and ChatGPT. Mistral already pulled the trigger, replacing its Le Chat chatbot with Vibes, its agentic coding environment.

ChatGPT and Claude are bigger brand names, though, which makes that kind of move harder. Users have existing mental models of what "Claude" does. Introducing agent capabilities without confusing the base is a product design challenge as much as a technical one.

ℹ️

Logicity's Take

The 90% non-coding stat is the headline number here. It suggests Cowork is finding product-market fit with operations teams, not developers. For AI builders evaluating agent frameworks, this is a useful data point: general-purpose agents may grow faster outside engineering orgs. The mobile approval flow also sets a design pattern. Competitors like OpenAI's Operator and Google's Project Astra will likely follow with similar cross-device human-in-the-loop features. If you're building on agent infrastructure, plan for async approval workflows from day one.

Frequently Asked Questions

When will Claude Cowork be available on mobile and web?

Beta access is rolling out gradually over the coming weeks, starting with Max subscribers.

Can Claude Cowork access local files from a browser?

No. Local file access, browser control, and Computer Use features still require the desktop app.

What tasks is Claude Cowork used for most?

Over 90% of usage is non-coding work. Business operations and content creation are the two largest categories, making up roughly half of all usage.

Does Claude Cowork run when my device is off?

Yes. Claude continues processing tasks in the background even when laptops are closed or phones are turned off.

How does Anthropic handle human oversight in Cowork?

When Claude reaches a decision requiring human judgment, it pauses and requests approval via smartphone notification. Nothing is sent without user review.

Also Read
The worst data breaches of 2026 so far

With AI agents now running background tasks across devices, understanding current data breach patterns helps teams assess risk.

ℹ️

Need Help Implementing This?

Building AI agent workflows into your product? Logicity's consulting team helps engineering and product teams design human-in-the-loop architectures, evaluate agent frameworks, and ship faster. Get in touch at logicity.in/contact.

Source: The Decoder / Matthias Bastian

Advertisement
H

Huma Shazia

Senior AI & Tech Writer

Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.