Getting Started with Claude Co-Work — Your First Day with an AI Coworker

What Co-Work Actually Is (And What It Isn't)

If you've used Claude before, you know the flow: you type a question, Claude responds, you go back and forth. It's a conversation. Co-Work is something fundamentally different. Instead of chatting with an AI, you're working alongside one.

Here's the practical difference. Regular Claude lives in a browser tab. You paste things in, it responds. Co-Work lives on your desktop. It can see your files. It can create new ones. It can open applications, navigate your browser, draft emails, build spreadsheets, and coordinate multiple steps to finish a task you'd normally spend an hour on.

Think of it less as "a smarter chatbot" and more as "a new hire who's extremely fast at research and document work." You tell it what you need, point it at the relevant files, and it goes to work. You review the output, give feedback, and iterate. Same dynamic as working with a human colleague — just faster.

What Co-Work is not: It's not autonomous AI that runs your business while you sleep. It's not replacing your judgment or your expertise. It still needs direction, still makes mistakes, and still requires you to review its work. The difference is that it can now do work, not just talk about work.

Anthropic launched Co-Work as a research preview in early 2026, and it's built specifically for knowledge workers — researchers, analysts, operations managers, legal professionals, finance teams, anyone who spends their day creating, organizing, or analyzing documents. You don't need to be technical to use it. That's the whole point.

Setting Up Co-Work: The 5-Minute Version

Getting started is deliberately simple. Anthropic designed this for people who aren't developers, and the setup reflects that.

Step 1: Get the Claude desktop app. Co-Work runs through the Claude desktop application, not the browser. Download it for your operating system (macOS or Windows) and sign in with your existing Claude account. If you don't have a paid plan yet, you'll need one — Co-Work is available on Pro ($20/month), Max, Team, and Enterprise tiers.

Step 2: Grant folder access. When you first launch Co-Work, it'll ask which folders on your computer Claude can read and write to. This is important: Claude can only see what you explicitly share. Start with one folder — maybe a "Projects" or "Documents" directory that contains the files you want help with. You can always add more later.

Step 3: Understand the permission model. Co-Work operates on an ask-first basis. Before it creates a file, opens an application, or takes any action beyond reading what you've shared, it tells you what it's about to do and waits for your approval. You're always in control. As you build trust, you can adjust these permissions, but the default is conservative — and that's a good thing.

Step 4: Start a conversation. Once your folder is connected, you can start giving Co-Work real tasks. Not "what is the capital of France" type questions — real work tasks. "Read through the files in my Q1 Reports folder and create a summary document." "Draft a response to this client email based on our project notes." That kind of thing.

The whole setup takes under five minutes. The learning curve is in figuring out what to hand off, not in figuring out the technology.

Your First Real Task

Let's walk through a concrete example so you can see what this actually looks like in practice.

The scenario: You have five meeting notes from the past week sitting in a folder. Each one is a rough text file — bullet points, action items, some scattered context. You need a single, clean summary document that captures the key decisions, open questions, and next steps across all five meetings.

Without Co-Work, this takes 30-45 minutes. You open each file, read through it, mentally synthesize the themes, start writing, go back to check details, reorganize, edit. It's not hard work, but it's time-consuming.

With Co-Work, here's what happens:

You type: "Read all five meeting notes in my Weekly Meetings folder. Create a summary document that includes: key decisions made, open questions that still need answers, and action items with who's responsible. Use the same format as the summary I created last month — it's called February-Summary.docx in the same folder."

Co-Work reads all five files. It identifies the reference document for formatting. It cross-references the notes, pulls out decisions vs. discussion vs. action items, and creates a new document in the same folder. The whole process takes about two minutes.

Is the output perfect? Probably not on the first try. You'll likely want to adjust phrasing, add context it didn't have, or reorganize a section. But you're editing a draft instead of staring at a blank page. That's the difference. You've gone from "create from scratch" to "review and refine." And for most knowledge work, that's where the real time savings live.

The key insight: your first task should be something that's time-consuming but not complex. Don't start with "reorganize my entire filing system" or "build me a quarterly strategy deck." Start with something contained — a summary, a draft, a data cleanup — where you can clearly evaluate the output and build your sense of what Co-Work handles well.

Dispatch: Your AI On Call

Dispatch is the feature that made me rethink what an AI tool can be. Here's the concept: you assign Claude a task from your phone (or desktop), and it works on it in the background. When it's done — or when it has questions — you get notified. You come back to finished work.

This flips the entire dynamic. With regular chat AI, you're sitting there watching it generate text in real time. With Dispatch, you fire off a request and go do something else. It's the difference between hovering over someone's desk and sending them a Slack message.

Real use cases where Dispatch shines:

• Morning commute task handoff: On your way to the office, you text Claude: "Read the three proposals in my Vendors folder and create a comparison table — pricing, timelines, key differentiators." By the time you sit down, it's done.
• Research while you're in meetings: "Find the five most recent industry reports on supply chain automation and summarize the key trends in a one-page brief." You go to your 2pm meeting. The brief is waiting when you get back.
• End-of-day prep: "Based on my calendar for tomorrow and the project files in my Active folder, create a prep document with context for each meeting." You send this at 5pm. Tomorrow morning, you're ready.

The mental model that works best is this: think of Dispatch like texting a coworker. You wouldn't micromanage every sentence they write. You'd tell them what you need, give them context, and trust them to produce a first draft. Then you'd review it. Same approach here.

One important note: Dispatch works within a continuous conversation. Claude remembers context from earlier in the thread, so you can build on previous tasks without re-explaining everything. Ask for a summary in the morning, then follow up in the afternoon with "Now turn that summary into a presentation outline." It knows what you're referring to.

What Co-Work Is Great At

After spending serious time with Co-Work, patterns emerge quickly. Some tasks feel like they were designed for this tool. Others, not so much. Here's where it genuinely delivers.

Research summaries. Give it a folder of articles, reports, or notes, and ask for a synthesis. It reads fast, identifies themes, and produces structured summaries that would take you hours to write manually. This is probably the single best first use case for most people.

Document creation with structure. Co-Work can create Excel spreadsheets with formulas, PowerPoint presentations with logical slide structures, formatted reports with tables of contents. It's not just generating text — it's creating functional documents. Tell it "Build a budget tracking spreadsheet with categories for marketing, operations, and personnel, with monthly columns and quarterly totals," and you get a working spreadsheet, not a description of one.

Data organization. Have a messy folder of files that need renaming, categorizing, or consolidating? Co-Work can read through them, understand what they contain, and organize them according to whatever system you describe. "Sort these invoices by vendor and create a master list with dates and amounts" is the kind of task that takes ten minutes to explain and two minutes for Co-Work to execute.

Email and communication drafting. Point it at a thread, your notes, or a brief, and ask for a draft response. It matches tone, includes relevant details, and saves you the cognitive load of context-switching between thinking and writing. You still review and edit — but the heavy lifting is done.

Multi-step workflows. This is where Co-Work's agent capabilities really show up. Using sub-agent coordination, it can break a complex task into smaller pieces and work through them systematically. "Read this contract, extract the key terms, compare them to our standard terms template, and flag any deviations in a summary document" — that's multiple distinct steps, and Co-Work handles them as a single request.

What Co-Work Can't Do (Yet)

Honesty matters here. Co-Work is impressive, but it's a research preview, and there are real limitations you should know about before you rely on it.

Cloud apps need Connectors. Out of the box, Co-Work works with your local files. If your work lives in Google Drive, Notion, Salesforce, or other cloud platforms, you'll need Connectors — integrations that organizations can set up to bridge Co-Work with those tools. Some are available now (Google Drive, Gmail), but the ecosystem is still growing. If your workflow is entirely cloud-based, you'll hit friction.

It can't make judgment calls. Co-Work can organize information and present options, but it shouldn't be making decisions that require human judgment. "Should we accept this vendor's proposal?" is not a Co-Work question. "Summarize the three proposals and highlight the tradeoffs" is. The distinction matters more than you'd think.

It needs clear instructions. Vague requests produce vague results. "Make this better" will get you something, but "Restructure this report to lead with the financial impact, move the methodology to an appendix, and add an executive summary" will get you something useful. The same prompting principles that work in regular Claude apply here — specificity, context, and clear success criteria.

Research preview means rough edges. You'll occasionally hit bugs, unexpected behavior, or tasks that don't complete cleanly. That's the reality of an early-stage product. Anthropic is iterating fast, but right now, you should treat Co-Work as a capable assistant that sometimes needs a second try, not a flawless system.

It can't verify facts independently. Like all large language models, Co-Work can generate plausible-sounding information that isn't accurate. When it creates documents based on your files, it's working with what you gave it. When it's drawing on its own training knowledge, the usual AI caveats apply: check the facts, especially for anything you're sharing externally.

Tips from the First Month

Here's what actually helped after spending serious time with Co-Work. Not theory — practical lessons from real use.

Start with low-stakes tasks. Your first few Co-Work projects should be things where a mistake doesn't matter. Internal summaries, draft documents, data cleanup. Build your intuition for what it handles well before handing it anything client-facing or high-stakes. This isn't about the AI's capability — it's about your calibration.

Review everything before it leaves your desk. Co-Work is a draft machine, not a finished-product machine. Treat every output as a first draft from a smart but imperfect colleague. Read it. Edit it. Add your judgment. The time savings come from not starting from scratch, not from skipping review.

Give context about your preferences. "Write a summary" gets you a generic summary. "Write a summary the way I'd write one — direct, no jargon, bullet points for action items, and always lead with the bottom line" gets you something that actually sounds like you. Co-Work can read your previous documents to learn your style, so point it at examples.

Use one folder to start, then expand. Don't grant access to your entire documents directory on day one. Pick one project folder, use Co-Work on it for a week, then add more as you get comfortable. This also helps you keep track of what Co-Work has created or modified.

Think in tasks, not conversations. The mental shift from "chatting with AI" to "delegating to AI" takes a minute. Instead of asking questions, give assignments. Instead of "What should I include in a project update?", try "Read the files in my Q1 Launch folder and write a project update email for the leadership team covering progress, blockers, and next steps." The more you think of it as delegation, the better your results.

Combine Dispatch with your real workflow. The biggest advantage isn't any single feature — it's integrating Co-Work into how you actually work. Before a meeting, Dispatch a prep brief. After a meeting, hand off the notes for cleanup. Between meetings, assign research tasks. It works best when it's woven into your day, not treated as a separate tool you occasionally visit.

Co-Work is still early. The product will change, features will improve, and the rough edges will smooth out. But even in its current state, it represents a genuine shift in how AI tools work. It's not a chatbot you visit. It's a coworker that's always available, infinitely patient, and getting better every week. The people who figure out how to use it well now will have a serious head start.