Cowork — Beyond the Obvious
Real use cases beyond "organize my downloads folder" — and the reframe that connects all of them.
The honest starting point
I didn't know what to do with Cowork when I first opened it.
I'd seen the demo. I knew it could rename files and organize folders. That's fine, but it's not why I was interested. I was interested because it seemed like it could do something more substantial — I just couldn't picture what that looked like in practice.
So I asked. I opened a chat, described what I was already doing with Claude, and said: given all of this, what would you actually use Cowork for? What's the non-obvious version?
That conversation became the Obsidian note that became this page. Which is maybe the most on-brand outcome possible: if you don't know what something can do, ask the thing. It'll tell you.
What Cowork actually is
Cowork is a desktop app that gives Claude access to your local files, connected tools, and the ability to run tasks in the background while you do something else. The difference between Cowork and a regular Claude chat isn't what Claude can do — it's when. In a chat, you stay in the conversation. In Cowork, you hand something off and come back to an answer.
That shift changes what's worth asking.
Your files are already a dataset
The thing that connects most of the genuinely useful Cowork tasks isn't a feature — it's a reframe.
You already have the data. Resume versions sitting in a folder. A job tracker you've been updating for months. Daily debrief files you wrote and then stopped reading after day one. Slack channels full of conversations you didn't have time to follow.
None of that needs to be created. It already exists. What it's been missing is something that can sit down with all of it at once, read it without you having to be there, and tell you what it sees.
That's the non-obvious version of Cowork. Not "help me rename these files." More like: here's a folder of things I made — what's actually in here?
Use cases
The most useful instruction you can give Cowork before it touches anything: don't.
The reconnaissance pattern — read everything, change nothing, tell me what you see — is the right first move for any folder of files you've accumulated over time. You find out what's there before you decide what to do with it. And sometimes finding out what's there is the whole task.
Two examples of what this looks like in practice.
I had a folder of resume files — different versions, different targets, no clear naming convention. Instead of opening each one, I pointed Cowork at the folder and asked it to read every file and give me a brief summary: what role each one seemed targeted for, what was different about it compared to the others, and roughly when it might have been created. Don't change anything. Just read.
What came back was a clean comparison across all of them — which version had the most technical depth, which led with healthcare domain, which was trying to do too many things at once. Five minutes of reading instead of forty minutes of opening and closing files.
Same pattern, different file. Export your tracker as a CSV, drop it in, and ask: where in the funnel am I consistently stalling? How long between application and first response on roles that moved forward? What patterns show up in companies that engaged versus ghosted? And — this last line matters — is there anything that stands out that I might not be seeing because I'm too close to it?
That last instruction is important. Give it permission to surface things you didn't think to ask about. The pattern recognizer sometimes needs a pattern recognizer.
- Say "don't change anything" first, every time — pure reconnaissance
- Ask for comparison across files, not just individual summaries
- "Anything that stands out" gives it room to surface things you didn't think to ask about
- Works on any folder of versioned documents, not just resumes
I'd been running a daily debrief process for a couple of weeks — a structured end-of-day capture with five sections: what happened in session, job search and admin activity, Todoist completions, what's carrying over, and wins. Every day. Consistent format.
At some point I realized I had a small dataset sitting in Obsidian and no idea what was in it, because I'd been writing forward and never looking back.
So I opened Cowork, attached the debrief prompt as reference, and asked it not to run the prompt — just read it, understand the structure, and tell me: what kind of insights could we look for in a collection of these files?
This is what it came back with:
Momentum and friction patterns. The "carrying over" section is basically a running log of what keeps getting deferred. If the same item shows up across three or four consecutive debriefs, that's not a task problem — that's a resistance signal. Worth knowing which things you keep not doing.
Win pattern recognition. This one's interesting because you wrote "don't let me off the hook" about wins into the prompt, which means the wins section has real stuff in it. Over time you can ask: what type of things actually feel like wins to you? Is it outputs? Decisions? Conversations? Connection moments? That can tell you something about where you're actually drawing energy from.
Job search velocity. Applications, recruiter contacts, roles researched — that's a timeline you can actually chart. You'd be able to see weeks where momentum dropped and whether there's a pattern to why.
The "anything to remember" section is probably the most interesting. That's the unfiltered, preserved-voice stuff. Themes that keep surfacing there — frustrations, realizations, things that clicked — are probably telling you something your to-do list isn't.
That's not a summary of what I wrote. That's analysis of what the structure of what I wrote reveals. Those are different things.
After that conversation I pointed it at the actual files. The resistance signal observation turned out to be accurate in ways I hadn't noticed.
- Start by asking what's findable before you ask it to find things — that conversation shapes better questions
- "I'm too close to it" is doing real work in that prompt — it gives explicit permission to surface the non-obvious
- Your debrief format doesn't have to match this exactly — adapt the sections to whatever you actually capture
- Run this every few weeks, not daily — you need enough files for patterns to emerge
My co-conspirator Jess and I share a Slack workspace with a bunch of channels — ideas, project threads, links we're saving, ongoing conversations. It's useful when we're both in it. It's a lot to catch up on when you've been heads-down on something else for a few days.
The Slack connector is one of Anthropic's official connectors — it shows up in the connector menu in Cowork and takes a couple of minutes to set up. Once it's connected, there's a built-in channel summary skill. You point it at your workspace, tell it what you want — a digest across all channels, or a specific channel, or activity on a particular topic over the last week — and it reads and summarizes. It asks clarifying questions first so you can refine before it runs.
What makes this worth having: it's not just "summarize these messages." It's "here's what's been happening while you were doing other things" — which is a different question than anything you'd ask in a regular chat, because a regular chat doesn't have access to your Slack history.
- Catching up after a few days away from a shared workspace
- Getting a topic digest across channels ("what has come up about X in the last two weeks")
- Finding things you know were discussed but can't locate
Cowork can schedule recurring tasks — run this prompt every morning, do this thing every Monday. The interface for setting it up is straightforward, and there are suggested prompts to get you started.
Cowork runs locally. Your laptop has to be open and running for a scheduled task to fire. If you close your laptop at night and your task is scheduled for 7am, it won't run. If your machine is asleep, it won't run. I also ran into some timezone quirks — a task set for 7:30am was firing at 8pm until I sorted out the settings.
If you have a desktop that stays on, or a more always-on setup, this works considerably better. For everyone else: treat scheduled tasks as reliable nudges during hours you're already working, not true background automation.
Features in this space are moving fast, and the scheduling story is one worth watching.
If you need something that runs at 5am whether or not you're at your machine, a server-side architecture makes more sense. The Hunt's nightly job pipeline runs on Netlify's servers for exactly that reason — see how that's built.
- Things you'd run at the start of an active work session anyway
- End-of-day captures during hours when you're reliably at your machine
- Anything where "it ran sometime today" is fine and "it ran at exactly 7am" isn't critical
All of these started the same way: I wasn't sure what was possible, so I asked. The answers shaped better questions. The better questions found things I didn't know were there.
That's the pattern. It works on Cowork. It works on everything else on this site. If you don't know what to do with a tool, describe what you have and ask what's findable. You might be surprised what's already in your files.