What is Claude Cowork, and does a one-person AI agency need it?
Claude Cowork lets Claude work on files in a folder you choose. What it does, how it differs from Claude Code, and whether a one-person AI agency needs it.
A florist sends you everything for her new website in one go: forty photos named IMG_4471 to IMG_4510, a price list typed into an email, three Word documents with the same "About us" in slightly different versions, and a voice memo about the Mother's Day range. Before you build a single page, someone has to sort that pile.
That kind of job is what Anthropic had in mind with Cowork.
Claude Cowork is a research preview Anthropic announced on 12 January 2026. In Cowork, you give Claude access to a folder on your computer, describe a task, and Claude reads, edits and creates files in that folder to get it done. Anthropic calls it "Claude Code for the rest of your work": the same kind of agent, aimed at documents and files rather than code. For a web designer, it's a helper for the paperwork around a project. For the build itself, Claude Code remains the tool made for that job.
What is Claude Cowork?
Cowork is a way of working with Claude where you hand over a task instead of having a conversation. Anthropic's announcement describes it like this: you choose a folder, Claude can then read, edit or create files in it, and once you've set a task it makes a plan and works through it, keeping you informed as it goes.
The examples Anthropic gave are ordinary office jobs: reorganising a downloads folder by sorting and renaming files, building a spreadsheet of expenses from a pile of screenshots, or writing a first draft of a report from scattered notes.
At launch, Cowork was available as a research preview to Claude Max subscribers in the Claude app for Mac, with a waitlist for other plans. Anthropic said it would improve it quickly and mentioned bringing it to Windows. "Research preview" means what it says: it's early, and it will change.
A few other details from the announcement matter for client work:
- It can use connectors. These are the links that let Claude reach information in other services. If you know what MCP means (the open standard many of those connections are built on), the idea is the same.
- It comes with some skills for creating documents, presentations and other files.
- It can work in a browser if you pair it with Claude in Chrome, Anthropic's browser extension.
- You can queue tasks rather than waiting for each one to finish.
What is the difference between Cowork and Claude Code?
Less than the names suggest. Anthropic says Cowork is built on the same foundations as Claude Code. Simon Willison, who writes closely about these tools, tried it on launch day and described it as Claude Code with a friendlier interface and a sandbox (a closed-off space the agent works inside) set up for you. In the desktop app, Cowork sits as its own tab beside Chat and Code.
The practical difference is who each one is for and where it works:
| Claude Code | Cowork | |
|---|---|---|
| Made for | Software and website work | Everyday tasks with files and documents |
| Where you use it | A terminal, an editor or the desktop app | The Claude desktop app |
| What it works on | A code project | A folder you choose |
| Good for web designers | Building and changing client sites | Sorting what the client sent, drafting documents |
If you already build client sites with Claude Code, Cowork doesn't replace any of that. Using Claude Code to build a website for a client covers what to decide before the build starts, and none of it changes.
Does a one-person AI agency need Cowork?
Probably not for building websites. Possibly for the work around them.
Here's where it could help in a typical small business project:
- Sorting what the client sent. Rename forty photos by what they show, move the price list into a clean text file, flag which "About us" version is newest.
- Drafting from notes. Turn your call notes into a first draft of the brief or the proposal, for you to rewrite.
- Making lists. A spreadsheet of every page, the photos each one needs and what's missing.
- Tidying your own admin. Receipts into an expenses list, old project folders into a consistent structure.
And where it doesn't fit:
- Building or editing the site. Claude Code is made for that, with tools for code, previews and version control.
- Anything that sends or publishes. A draft email is fine. Sending it, or changing a live site, should wait for you to read it.
- Sensitive client material. Contracts, financial documents and passwords shouldn't sit in a folder an agent is working in.
If you work alone and already pay for Claude, it's worth trying on one messy client folder. If you don't, your current setup won't be worse for skipping it.
What should I check before giving Cowork client files?
Anthropic's announcement is direct about the risks, and they're worth taking seriously with someone else's business information.
It only sees what you give it. You choose which folders and connectors Claude can see, and it can't read or edit anything else. So make a folder for the task and put copies in it, not your whole client drive.
It can delete and change files. Anthropic says Claude will ask before significant actions, but also that it can take destructive actions, such as deleting local files, if it's told to, and that it may misread an instruction. Give it copies, and be precise: "rename the photos, don't delete any".
Content can carry instructions. Anthropic warns about prompt injection: text hidden in a web page or document that tries to change what Claude does. It says it has defences but that securing an agent's real-world actions is still being worked on across the industry. For client work, that's another reason to keep Cowork on your own files and away from anything that sends, pays or publishes.
Keep the decisions yours. The same rule applies to every agent you use: it drafts, you decide. What is MCP and what it changes for agency work goes through what to check before connecting client information to an agent, and it applies here too.
How does Cowork fit with skills and plugins?
Cowork shipped with a starter set of skills (packaged instructions and files that help Claude with a type of task) for documents and presentations. If you've used Claude Code skills, it's the same idea: you teach the agent how you like something done once, instead of explaining it every session.
For a web designer, the useful skills are the ones that carry your standards: how you name files, how you write a brief, what a page list looks like. Those work best written down once and reused, whichever tool reads them.
Where does Volant fit?
Cowork, Claude Code and the tools that follow are all ways for an agent to do work. What a one-person agency also needs is a place for what surrounds that work: which client it's for, what was agreed, what's been approved and what's due next.
That's the split Volant is built on. Your agent does the thinking and the making. Volant keeps the records, the schedule and the approvals. If you're a developer building client sites with Claude Code, building client sites with an AI agent shows how the two fit together, and approving what your agent sends covers the approval step.
For now, Cowork is worth a try on one pile of client files. Give it a copy, give it a clear task, and read what it made before any of it reaches the client.