Claude Cowork Comes to Phone and Web: an Agent That Works While Your Laptop Is Closed

On July 7 Anthropic brought Claude Cowork to phone and web. It used to be a desktop agent that lived on your laptop; now Claude Cowork shows up in the Claude app on iPhone, iPad, and Android, and on claude.ai. Anthropic's own line for it is blunt: "your work goes everywhere with you, and keeps going without you."

At Gless we build AI agents for client work every week, so we read releases like this with one question: what here actually changes how you work, versus just moving the interface to another screen. This is the case where a dull headline — "the agent is on your phone now" — hides something that matters: the asynchronous work model. You hand off a task at your desk, close the laptop, and pick up the result on your phone on the way to a meeting. Let's go through the facts: what was announced, how the handoff works, why decisions land on your phone, and what of this is usable in an ordinary business.

What Anthropic announced

Claude Cowork is now rolling out in beta on web and mobile — iOS, iPadOS, and Android. The announcement landed on July 7, 2026. Access opens gradually over several weeks, starting with the Max plan; other tiers follow. Once it reaches you, Cowork appears on the Claude home screen next to the regular chat.

One more detail for existing users: Anthropic extended its doubled Cowork usage limits through August 5. That isn't trivia — an agent that runs in the background and on a schedule burns through far more requests than an on-demand chat, so extra headroom during the beta takes some of the sting out.

The point of the announcement isn't a new model. It's the same Claude Cowork, unhooked from a single device. And that unhooking of "agent equals my laptop" is exactly what makes it interesting.

The rollout shape is worth noting too. Anthropic isn't flipping it on for everyone at once; it's shipping in waves, starting with the heaviest Max users. The logic is clear: a background agent creates a very different load profile than chat, and it's smarter to break that in on people who already hit their limits before opening it to mass-market plans. For a business that means one thing — if you want to try the format first, look at Max rather than waiting for the other tiers to catch up.

The async handoff: hand it off at your desk, pick it up on your phone

What changes is the work model. In a normal chat you sit and wait for the answer. In Cowork you specify an outcome, not the steps: "put together the brief for Monday's meeting." Claude builds a plan and works on its own until the task is done or it hits a decision only you can make.

The mobile version adds three things to that:

• Sessions and files sync across devices. Start at your desk, check progress from your phone, grab the finished document from any screen.
• Tasks run in the background even when no device is online. Laptop closed, phone in your pocket — Claude keeps going.
• Work can run on a schedule. Anthropic's own example: set Monday client prep for 6 a.m. — Claude works through email threads, call transcripts, and recent news and has the briefing ready by the time you open your laptop.

For a team that ships automation, this is a familiar picture. A good agent isn't a conversation partner that answers while you stare at the screen. It's a doer you hand a task to before getting on with your day. Cowork on the phone just takes that idea to its logical end: work stops being tied to whether your laptop is open.

Decisions land on your phone

Here sits the most underrated part of the announcement. Anthropic puts it this way: "when Claude reaches a call only you can make, it asks, and the question reaches your phone." You can redirect a draft mid-meeting, and Claude keeps going with the change.

This is human-in-the-loop done right. When we design agents for clients, the most common fear sounds like this: "won't it fire garbage at my customers while I'm not looking?" The answer lives in the decision points. The agent runs the routine on its own and stops exactly where the cost of a mistake is high: before an email goes to a client, before a budget gets spent, before something gets published. The decision goes to a human, the human confirms or redirects with one tap, and the agent moves on.

Mobile Cowork makes that pattern the default. The decision arrives as a push, instead of waiting for you to get back to the laptop. The difference between "an agent you're scared to trust" and "an agent that actually saves you hours" almost always comes down to how carefully those control points are placed.

In practice it looks ordinary. The agent prepares a campaign to your list, assembles the segments and copy itself, and stops at the "send" button to show you a preview. In the thirty seconds between two meetings you look it over, fix one line, and hit "ok." The grunt work is already done; the decision is all that's left for you. That's how autonomy stops being scary — the human doesn't drop out of the process, they just stop doing by hand what doesn't need a hand. We covered what an AI agent is, and how it differs from a chatbot, separately.

90% of Cowork's work isn't code

The most telling number in the release: over 90% of what people do in Cowork isn't software development. It's ordinary knowledge work. The two largest categories — business operations and content creation — together make up about half of all usage.

Behind those percentages are concrete tasks: reconcile the quarter's spending, write a variance memo, assemble a client deck from call transcripts. Not "write me some code," but "clear my routine."

For us that confirms what we see in practice every day. AI agents don't land in the IT department — they land in sales, finance, marketing, and support, where people drown in repetitive operations. If you want numbers and examples, we wrote about how AI agents are changing work using real usage data.

And yes, the same 90% raises the usual question: "if the agent does that much work, who does it replace?" Short answer — not engineers or specialists, but the routine part of their day. We laid out at length why AI doesn't replace engineers, it changes what you pay them for.

Cowork, Claude Code, and the desktop: what goes where

Cowork is easy to confuse with Claude Code, but they do different jobs. Claude Code is an agent for software development. Cowork is an agentic harness for general knowledge work: marketing, sales, legal, finance. Going mobile blurs the line between Anthropic's products a little, which is a story of its own — but for now the roles hold.

The desktop version isn't going anywhere either: you still need it where local context matters.

Simple rule: phone and web to assign tasks and make decisions from anywhere, desktop when the agent needs access to your machine and local environment.

What this means for your business

Strip away the Anthropic branding and you're left with an operating model worth trying on. Work as a task queue: you hand off an outcome, not an instruction; the agent plans and does it; it stops only at the forks where your decision is needed; you confirm from any screen. And the laptop can stay closed.

This isn't press-release science fiction, it's a concrete way to take load off a team. It works exactly as well as you split the work into "routine the agent handles alone" and "points where a human decides." The part that's easy to get wrong is usually what we take on: where to place the controls, which systems to grant access to, how to keep an autonomous agent from becoming a source of chaos. You can see how that's set up in our AI implementation services.

Claude Cowork on the phone is a signal of where this is heading: the agent stops being a browser tab and becomes a doer that works while you're busy with something else. The question isn't whether this format shows up in your company, but which processes you'll hand it first. If you'd like to scope something like this for your business, get in touch.