Most agentic AI coverage assumes you write code. Claude Cowork is different. It runs real routines on your computer while you sleep and helps you stay on top of things you would otherwise miss.
Claude Cowork runs scheduled routines on your computer without any engineering required. The first sign it is working is often money already in your account or a task already finished by the time you wake up.
Most coverage of agentic AI is written for people who write code. It talks about terminals, APIs, and scripts. Claude Cowork is aimed at a different person. It runs on your desktop, connects to the apps you already use, and does work on a schedule you set by describing what you want in plain English. No terminal required.
The scheduling capability is where it gets interesting. Here is what that looks like in practice.
The most useful thing about Cowork is that it does not need you in the room. You set up a routine once, and it runs on whatever schedule makes sense.
Southwest rebooking. Every night at 2am, a routine checks all my upcoming Southwest flights to see if any can be rebooked for the same exact flight and fare class at a lower price. Southwest keeps the difference as a credit. So far this has earned over $200 in flight credits without me thinking about it once.
Auto-refunds. Some retailers will refund you the difference if a price drops after you buy something. Finding those drops, reading the refund policy, and filling out the request is difficult, tedious, and annoying. It almost never gets done. A routine runs overnight and handles these when I have tokens left at the end of the day. It has recovered money I would have left on the table.
Wish list research. I keep a note with things I want to buy. Every Sunday morning, when I am rarely at my computer anyway, Cowork works through the list. For each item it finds the current price, checks for refurbished or scratch-and-dent versions, looks for rebates and bundles, and notes whether a newer version exists. I can do this research myself, and there is some joy in it when I do. But it has already found several items on sale below the price I was prepared to pay, and those I would have missed.
Around 8pm, after I put the kids to bed, a routine runs through all the conversations where I was not the last person to respond, the thread is at least a week old, and we have more than three months of message history. For each one, Cowork surfaces a natural way to restart the conversation.
By the time I come back from putting my kids to bed, I have a short list of people I could reach out to and something real to say. Maintaining relationships takes effort. This routine makes it easier at the exact moment of the day when energy for it is lowest.
Before I leave my computer for something that will take more than a few minutes, I run a routine that reads through my open applications, calendar, emails, and recent texts, then puts together a quickstart guide for when I come back.
Context-switching is expensive for people with ADHD. The cost of leaving a task mid-way is the time to reconstruct where you were. I think of this routine as a superhero hack: a one-page summary written by something that actually read your open windows makes re-entry manageable. This is one of those things that sounds minor until you use it regularly.
Different people communicate in different ways. I basically only communicate in one way. A message that works with one person can fall flat with another, and I rarely stop to think about which approach I am using.
When I have an email I need to send but not urgently, I can reference it in Cowork and ask it to look at my communication history with that person, figure out what tends to work, and suggest the right framing, channel, and tone to get a response. It is a small adjustment that saves real friction.
Most of these routines work because Cowork can control Chrome. It navigates pages, reads content, fills out forms, and completes flows on your behalf. That is what makes automated rebooking, refund requests, and price research possible without you doing any of the clicking.
It does get things wrong sometimes. A page layout it has not seen before, a multi-step form that behaves unexpectedly, a site that loads slowly. When that happens the routine usually needs a retry. The value of what it gets right is large enough that a retry now and then is an easy tradeoff.
If you are curious what Cowork actually looks like, Anthropic has a product overview and ran an intro webinar when it launched. It is available on all paid Claude plans through the desktop app.
If you want to try it, here is my referral link: claude.ai/referral. You get a discount on your first month and I get a credit.