Most AI agent demos are about writing code. But according to Anthropic's own usage data, more than 90% of Claude Cowork activity isn't software development at all — it's the unglamorous "work around the work": reconciling spend, drafting memos, chasing down contract terms. As of July 7, 2026, that work no longer has to stay tethered to your laptop. Cowork is expanding to mobile and web, with beta access rolling out over the next several weeks starting with Max plan users.

If you've been treating Cowork as a desktop-only tool, this changes the shape of what you can hand off.

Why This Matters

The core limitation of agentic AI so far hasn't been intelligence — it's been availability. A task that takes three hours can't run if closing your laptop kills the session. Cowork's expansion removes that constraint in three concrete ways:

1. Sessions persist across devices. Start a task at your desk, check progress from your phone, and grab the final output wherever you are.
2. Background execution with no device online. Scheduled tasks now run independently — you don't need a laptop lid open or a phone screen on.
3. Approval checkpoints still route to you. When Claude hits a decision only you can make, it pauses and asks — on your phone, mid-meeting if needed — rather than guessing or stalling.

Desktop still owns the deepest workflows, since it's the only surface with local file and browser access. But the web and mobile apps now cover a large share of what people actually delegate.

Five Examples of What This Looks Like in Practice

1. Overnight client prep. Schedule a 6 a.m. task: Claude works through email threads, call transcripts, and recent news, assembles a briefing doc, and drafts (but doesn't send) the follow-up email. You review it with coffee.

2. Quarter-end reconciliation. Point Claude at the spend data and ask for a variance memo. It runs the numbers and drafts the write-up while you're in back-to-back meetings.

3. Contract-to-tracker conversion. Hand Claude a folder of renewal contracts and ask for a structured tracker with risks flagged — the kind of task that used to eat an entire afternoon of manual review.

4. Live dashboard building on the move. One Ramp customer described starting a client-tracking dashboard on a laptop, then picking up the same session on a phone while waiting at baggage claim — no restart, no lost context.

5. Deck assembly from raw inputs. Feed Claude call transcripts and pipeline data and ask for tomorrow's client deck. It builds a draft; you shape the final version before it goes out.

A Quick Look at the Interaction Pattern

You're not writing API calls for this — Cowork tasks are natural-language handoffs. A typical scheduled instruction might look like this:

Getting Started

• Web: Start a Cowork session from the home screen at claude.ai.
• Mobile: Open Cowork from the sidebar in the Claude app (iOS/Android).
• Desktop: Still the fullest experience, with local file and browser access — download it at claude.com/download.

Anthropic is also doubling Cowork usage limits through August 5, 2026, specifically to encourage people to test bigger, multi-step tasks during the rollout.

The Takeaway

This release isn't a new capability so much as a removed constraint. The intelligence was already there; what was missing was the ability for that intelligence to keep working when you weren't looking at a screen. For anyone tracking how agentic AI moves from "impressive demo" to "actual daily habit," device-independent execution with human-gated decisions is the more interesting story than another benchmark score.

Resources

https://claude.com/blog/cowork-web-mobile/

Disclaimer

This article is not endorsed by Salesforce, Google, or any other company in any way. I shared my knowledge on this topic in this blog post. Please always refer to Official Documentation for the latest information.

0 Comments