One thing changed recently, and it's good news
If you've been confused about where Cowork fits, you weren't being slow. Anthropic changed it.
From 16 September 2026, Cowork stopped being a separate product and started being folded into the main Claude app. Anthropic's own product page now says it plainly: "Claude Cowork is now just Claude." The rollout began with Pro and Max plans, with others following, so depending on your plan you may still see the Cowork name for a while during the changeover.
What this means for you: nothing you'll learn here is wasted. The way of working that Cowork introduced (pointing Claude at a folder on your computer and handing over a whole job) is not going away. It's becoming a normal part of Claude. Only the separate name is disappearing.
What still hasn't merged: Claude Code. It remains its own separate thing, and it's still the most capable of the three.
So the honest way to think about it today is two tools and two gears: Claude (which has a quick gear and an agent gear), and Claude Code. This guide walks all three, because the three names are still what you'll see and hear.
The difference in one table
| Claude Chat | Cowork (agent mode) | Claude Code | |
|---|---|---|---|
| Best for | Quick questions, research, short writing | Real document work across your files | Big, repeating, mostly hands-off jobs |
| Where the work lives | In the conversation | In a folder on your computer | Across your computer and connected tools |
| How long it runs | Seconds | Minutes | Long stretches, largely unattended |
| How involved you are | You read every reply | You set the job, then review the result | You set it going and check back |
| Asks permission | Every step is yours | Often | Less often, by design |
| Feels like | A very fast researcher | A capable colleague | A colleague you've stopped supervising |
What each one is actually for
This is the one most people already know: you open Claude, ask something, and read the answer. It's built for tasks that start and finish inside a single conversation: a question you need answered, a bit of research, a first draft of an email or a social post.
It's easy to dismiss chat as the beginner option. It isn't. Skills and connectors work here too, which means Claude can reach your other tools and follow your own saved instructions about how you like things written. If you want a document produced a particular way, consistently, chat handles that well.
Use it when: the whole job fits in one conversation and you want to see each answer as it arrives.
This is the one worth understanding properly, because it's where the actual time savings live for most small businesses. It's built for real knowledge work: PDFs, spreadsheets, presentations and documents, the stuff your business already runs on.
Two things make it genuinely different from chat.
First, it works directly on the files and folders on your computer. You point it at a folder, and it reads what's in there, understands it, and produces new work in that same place. You're not copying text into a chat box and pasting results back out. It's working where your files already are.
Second, it handles big jobs on its own. Give it something that would take you an afternoon and it breaks the work into a task list and moves through it, checking things off as it goes. You can watch it work, but you don't need to sit there steering every step.
Put those together and it stops feeling like a chatbot and starts feeling like working with a capable colleague, one who reads the whole folder before starting.
Use it when: the information is already in files, and the job has more than a couple of steps.
The name puts people off, and it shouldn't. Yes, it was built for developers, and yes, there's a version that runs in a terminal window. But there's also a desktop app, and you instruct it in plain English exactly as you would anywhere else. The name describes where it came from, not who it's for.
The simplest way to think about it: it's agent mode on steroids. More powerful, more autonomous, runs for longer, and reaches more things without stopping to ask you first.
That last part is the real difference, and it's the reason it's worth treating with a bit of respect. It isn't harder to use. It just does more between check-ins, which is exactly what you want when you're building something that should happen repeatedly without you, and exactly why you want to be deliberate about which folder you've pointed it at.
Use it when: you want a job to keep happening without you having to start it each time: recurring reports, ongoing automations, AI agents that run on their own.
Which one do you need right now?
Two questions, about ten seconds. Answer for the actual task sitting in front of you.
Which one for the thing in front of you
Common small business tasks, and where each one belongs.
The five-second test: ask yourself where the work currently lives. In your head, or on a website? That's chat. In files and folders on your computer? That's agent mode. Does it need to keep happening without you starting it? That's Claude Code.
Three starter prompts, one for each
The fastest way to feel the difference is to run one of each, today, on something real.
I run a [type of business] in [location]. Our customers are usually [who they are, in a sentence]. Write me three short social posts about [what you want to promote this week]. Keep the tone [how you actually sound - e.g. warm and straightforward, not salesy]. No hashtags, no emojis.
Read every file in this folder and give me a one-page summary I could hand to someone who has never seen it. Tell me: what's working, what isn't, and the three things I should do next. Save it in this folder as a new document. If anything in the files contradicts anything else, flag it rather than picking one.
Every month I do the same thing by hand: [describe the job, step by step, as if explaining it to a new hire]. Set this up so it can be done again next month without me rebuilding it. Walk me through what you're doing in plain English as you go, and ask me before anything gets deleted or overwritten.
Your first week
Spend day one in chat only
Ask it the things you'd normally google. Get a feel for how specific you need to be.
Make a practice folder
Copy a handful of real documents into a new folder. Copies, not originals, so you can be fearless about it.
Point agent mode at that folder
Use the middle prompt above. Watch it build the task list. That moment is usually when it clicks.
Find your one repeating job
The thing you do every month that you resent. That's your Claude Code job, and it's the one worth the setup.
Write down what worked
Whatever prompt got a good result, save it. That's the start of a system rather than a lucky day.
- Do you know which of the three you'd open for your next task?
- Have you pointed Claude at a real folder at least once?
- Have you identified the one job you'd most like to stop doing by hand?