Cowork, for the rest of the team.

Everything so far in this book has leaned on Claude Code, the part of Claude that builds things on your computer. That's where the heavy lifting lives, and it's where you'll do your own best work. But most of the people in a business aren't building anything. They're closing the month, chasing paperwork, answering the same five questions, and turning a pile of notes into something a client can read.

For them, there's Cowork. Anthropic describes it as Claude Code's abilities brought to knowledge work beyond coding, with no terminal required. The idea is simple: instead of chatting with Claude one message at a time, you describe the finished thing you want, step away, and come back to it done. A spreadsheet with working formulas. A slide deck. A folder of receipts turned into an expense report. A week of scattered research turned into two clean pages.

Chat, Cowork and Code.#

The easiest way to understand the three is by how much you hand over. In Chat, you steer turn by turn: you ask, it answers, you ask again. Anthropic's own training puts it neatly: Chat is a conversation, and Cowork is a delegation. Code is delegation too, but for building software and systems, and it's the one this book has mostly been about.

One practical difference matters more than the rest. When Chat makes you a file, it hands it back as a download, and you save it somewhere. Cowork saves straight into the folder you gave it, next to everything else. That sounds small until you've asked for twelve client reports in a row.

You'll also see this shifting under your feet, so here's the honest state of it. Anthropic is folding Chat and Cowork into one conversation for Pro and Max accounts, where you just ask and Claude decides whether it's a quick answer or a task to go off and do. Team and Enterprise accounts keep a separate Cowork option for now. Whichever you see, the abilities in this chapter are the same. Only the button moves.

What to hand it first.#

Cowork needs a paid plan, and on a Mac or Windows computer the desktop app gives you the full version, because that's where it can reach your own folders and browser. You connect a folder, say what you want, and it works through the steps. Anthropic's own examples are refreshingly ordinary: sort a messy Downloads folder by type and date, turn a stack of receipts into an expense report, rename a batch of files, pull the key points out of a call transcript, clean up a spreadsheet, build a chart.

In a real business, the first wins tend to look like this. The office manager hands it the month's supplier invoices and gets back a spreadsheet that matches each one to a job. The owner drops in three competitors' price sheets and gets a one page comparison. A salesperson gives it a transcript from a call and a folder of past proposals, and gets a first draft proposal in the company's usual layout. None of these are clever. They're just the hours that disappear every week, handed to something that doesn't get bored.

Go through the invoices in this folder, match each one to a job in the jobs spreadsheet, and give me a new sheet with anything that doesn't match at the top. Don't change the originals.

Notice the shape of that request. It names the finished thing, points at the material, says what matters most, and draws one line it must not cross. That's the same four part ask from the best practices chapter, and it works just as well here.

One honest note on cost. Anthropic says plainly that Cowork uses your plan's limits faster than chat, because a task that reads forty files and builds a spreadsheet does far more work than answering a question. If your team leans on it daily, that's a reason to look at the plan choices in the next chapter.

Safe with your files.#

The first question every owner asks is whether it's safe to let this loose on real company files. The answer is a sensible set of fences, and one habit on your side.

The fences are these. Cowork can only read and write inside folders you've connected, so the smart move is a dedicated working folder rather than your whole drive. It asks for your explicit permission before permanently deleting any file, in every mode. When a session runs in the cloud, it runs in an isolated, temporary space on Anthropic's servers that can't reach your home or company network and is removed when the session ends. And you choose how often it checks in, just like the permission modes in Claude Code.

Now the habit. Anthropic is unusually direct that isolation limits where Claude's code runs, not what Claude reads or does. The real risk is content that tries to give Claude orders, like an email that says "transfer $1,000 to this account." Anthropic trains against this, scans untrusted content and screens actions, and still says the chance of an attack is not zero. So the rule from staying in charge applies here too: Cowork drafts, and a person sends, pays and deletes.

Jobs that run themselves.#

The point where Cowork starts to feel like a staff member is scheduled tasks. In any Cowork task you can type /schedule, or open Scheduled in the sidebar, and set a job to run hourly, daily, weekly, on weekdays, or only when you press go. These run in the cloud, so your computer doesn't need to be on.

Good first candidates are the reports nobody enjoys making: a Monday summary of last week's jobs from your connected spreadsheet, a morning digest of the inbox with anything urgent at the top, a Friday list of invoices that are thirty days overdue. Anthropic's own advice on scheduling is worth taking word for word in spirit: start simple, don't schedule anything that sends messages, makes purchases or touches sensitive files, read what it produces, and pause the ones you've stopped using.

Every weekday at 7am, read yesterday's new rows in the jobs sheet and write me five lines: what came in, what's stuck, and anything I need to decide today.

Cowork can also use the apps you've connected, like Gmail, Google Drive, Microsoft 365 or Slack, and a browser to open sites, read pages and fill in forms. Anthropic strongly advises against pointing the browser at sensitive things like bank or medical accounts, and that's advice to take. Connectors for reading, drafts for writing, and a person for anything that can't be taken back.

Skills: teach it once.#

Here's the problem Skills solve. Your business has a way of doing things. The monthly client report has a certain layout. Quotes follow a certain structure. Job files get named a certain way. Without help, every person on your team explains this to Claude from scratch, slightly differently, and gets slightly different results.

A Skill is Anthropic's answer: a folder of instructions, and sometimes templates or small scripts, that teaches Claude how to do one job your way. The clever part is that Claude doesn't read every Skill every time. At the start, it only sees each Skill's name and a one line description, a tiny cost. When a task matches, it opens the full instructions and follows them.

Claude ships with a few Skills already, for making Excel spreadsheets, Word documents, PowerPoint decks and PDFs, and it uses them on its own when a task calls for one. Your own Skills are where the value is, and you don't need to write one by hand. Tell Claude you want to create a Skill for, say, your monthly client report, show it a good example, and it will write the instructions and package them for you. On a Mac, Cowork can even watch you do the job once, by recording your screen and voice for a few minutes, and propose a Skill from that.

---
name: monthly-client-report
description: Build the monthly report for a client from their jobs sheet.
---

- One page. Our logo top left, client name and month top right.
- Start with three plain sentences on how the month went.
- Then a table: jobs done, jobs open, invoices paid, invoices due.
- Never include internal notes or margins.
- Save as a PDF named Client-Month-Year in their folder.

How is this different from the standing orders in CLAUDE.md, or from connectors? Anthropic draws the lines clearly. Your standing instructions apply to everything, all the time. A Skill is a procedure that only switches on for its one job. And a connector gives Claude access to a tool, while a Skill teaches it how to use that tool well. In Claude Code, the advice is the same: keep CLAUDE.md for short facts, and move anything that reads like a step by step procedure into a Skill.

Explain it once, well, and nobody on your team has to explain it again.

One way of working, shared.#

Skills are where a team plan earns its keep. On Team and Enterprise, an owner can add a Skill for the whole organization, and it's switched on for everyone straight away. Members can share their own Skills with specific colleagues or publish them to a company library, and owners can require a review before anything is published. Skills you turn on in your Claude account also carry over into Claude Code when you sign in with the same account, so the report Skill works the same whether it runs from Cowork or from your always on Mac.

The payoff is consistency, which is most of what a small business is fighting for. The new hire's first client report looks like the owner's. The quote your salesperson sends on Friday has the same structure as the one sent on Monday. And when you improve the Skill after a client complaint, everyone's next report gets better at once, without a meeting. Treat Skills like any other company file, too: only install ones from sources you trust, because a Skill is instructions Claude will follow.

If you take one thing from this chapter, take this. Claude Code is for the person who builds the systems. Cowork, with a handful of good Skills, is how everyone else in the building gets the same leverage, with their own work, in their own folders, without learning anything technical at all.