Claude Cowork comes to Android and web: AI agent now runs even with your computer turned off

The abbreviation “AI agent” has been bandied about so often in the last year that it makes my eye twitch. Nevertheless, one piece of news is worth noting: Anthropic has released its assistant Claude Cowork from desktop to web and, more importantly, to Android. And what’s more interesting than the phone button itself is where the work now takes place.

It works even when you sleep

The fundamental shift is that agent tasks now run on Anthropic’s servers, not on your hardware. In practice, this means you can schedule a routine — for example, have it sort your emails overnight, review industry news, and deliver a ready summary in the morning — and it will run even if your laptop is completely turned off. Moreover, you can start work on your computer, check it on your phone on the tram, and pick up the results anywhere. It sounds convenient; personally, however, with “set it and forget it” agents, I always find myself wanting to see what they’ve done.

Who actually uses it (and for what)

The numbers Anthropic published from an analysis of 1.2 million anonymous sessions are interesting. I would have expected that such a tool would be primarily used by programmers — but software development accounted for only 8.7% of activity. Over 90% of the work fell into ordinary office tasks: routine administrative tasks like expense reconciliation or compiling team reports took up 33.4%, writing texts and content another 16.4%. In other words, the “revolutionary AI agent” most often does what no one wants to do — paperwork.

The catch: the phone is more of a remote control

But before you put your computer in the closet, two sober notes. More demanding work with files and the browser still requires a running desktop application in the background — so the mobile acts more like a remote control than a full-fledged machine. And then there’s the price: it’s currently a beta only for the Max plan at $100 per month, which is approximately over 2,100 CZK. At least it’s nice that the agent doesn’t have complete free rein — before sending an email or making an important decision, it sends you a request for manual approval on your phone. Which is exactly the safeguard I want to have with an independently acting AI.

Would you let AI work in the background for you, or do you want to oversee every step?

Source: Anthropic (Claude.com), Notebookcheck