Here's something that caught my attention: nearly all of Anthropic's non-technical staff—designers, product managers, data scientists—use Claude Code daily. Greg Isenberg got a private walkthrough of Co-work and Claude Code, and I've pulled out the key takeaways below if you're short on time.
I got a private lesson on Claude Cowork & Claude Code
Three clients asked me the same question last week: "How do I get Claude Code without learning to code?"
I've spent 30 years watching technical tools stay locked behind terminal windows while business owners wait for "the easy version." Usually, the easy version never comes—or when it does, it's so dumbed down it's useless. Co-Work is different. It's the actual Claude Code engine, the same one engineers use, wrapped in an interface anyone can operate.
Here's what caught my attention: Anthropic didn't build Co-Work for customers first. They built it for themselves. Half their sales team uses Claude Code every week. Nearly all their non-technical staff—designers, product managers, data scientists—use it daily. They needed something their own people could use without a terminal. Co-Work is that something, and now it's available to everyone on a paid plan.
What's This New Way to Use AI Without a Terminal?
Co-Work is Claude with file access. You point it at a folder on your computer. It reads what's there, edits what you ask, and creates new files when needed. That's it. No command line. No configuration files. No Git repositories.
Under the hood, it runs the exact same Claude Agent SDK that powers Claude Code—the tool engineers use to build applications. According to Anthropic's engineering team, "We actually use that same exact SDK directly in Co-Work." You're not getting a watered-down version. You're getting the real engine with training wheels.
As of January 2026, Co-Work is available on Pro, Team, and Enterprise plans through the Claude Desktop app. One catch: macOS only for now. Windows support is listed as "coming soon."
Who's Actually Using This at Anthropic?
The adoption numbers inside Anthropic tell the story better than any feature list.
- 50% of Anthropic's sales/GTM team uses Claude Code every week
- Nearly 100% of non-technical staff (designers, PMs, data scientists) use it daily
- The tool was built because their own teams needed it
When half a company's sales team voluntarily adopts a coding tool, something fundamental has shifted. These aren't people who need to write software. They're people who need to get work done—and they found that an AI agent with file access does it faster than switching between apps.
How Does Pointing AI at a Folder Actually Work?
The mental model is simple: you're giving Claude a workspace.
- You designate a folder on your computer (your "workspace")
- Claude can read any file in that folder
- Claude can edit existing files or create new ones
- Claude asks permission before taking significant actions
- You can course-correct at any step
This permission system matters. Claude doesn't run wild through your filesystem. It tells you what it plans to do, waits for approval, then executes. If something looks wrong, you say stop. The agent loops you in rather than surprising you with changes.
For tasks that need browser access—drafting emails, working with Google Sheets, testing web applications—you can pair Co-Work with the Claude Chrome extension. This lets Claude see and interact with what's in your browser, not just your local files.
How Do You Set This Up in 10 Minutes?
- Download the Claude Desktop app (macOS only for now)
- Sign in with your Pro, Team, or Enterprise account
- Choose a folder for Co-Work to access—start with something specific like "Q1 Reports" rather than your entire Documents folder
- Install the Claude Chrome extension if you need browser-based tasks
- Give Co-Work a task and watch it work
That's the entire setup. No API keys. No environment variables. No YAML configuration. If you can install an app and choose a folder, you can run Co-Work.
What Can This Actually Do for Your Business?
The use cases I've seen work best are the tedious, multi-step tasks that eat up your afternoon. Here's a real example from Anthropic's own team:
I have a spreadsheet where we track all the team's work for the month. Instead of having to bug everyone on the team to fill out the status, I ask Co-Work to look at the spreadsheet, find any column that's not filled out, and message the engineer on Slack.
That's team management automation without writing a line of code. The AI reads the spreadsheet, identifies gaps, and handles the follow-up communication.
Other documented use cases include:
- Drafting emails based on context from multiple files in your workspace
- Analyzing spreadsheets and generating reports
- Editing documents with specific style guidelines
- Processing data from Google Sheets (via Chrome extension)
- Creating new files based on templates or patterns in existing files
If you're spending time on work that involves reading files, making decisions, and creating output—Co-Work can probably do most of it. The question is whether you trust the output enough to use it. (More on verification below.)
What Are These Reusable Shortcuts and Why Do They Matter?
Skills are prepackaged instructions that teach Claude how to work with specific file formats or tools. Think of them as recipes: step-by-step procedures for handling particular tasks.
Anthropic ships Co-Work with built-in skills for common formats like Excel. But the real power is custom skills—you can teach Claude to work with proprietary software or unusual file formats your business uses.
As one of the Anthropic engineers put it: "If you have some weird file format—if you work with AutoCAD or Salesforce or whatever—you just make a skill and then Claude can do it for you."
For most business owners starting out, you won't need custom skills. The built-in capabilities handle standard office workflows. But knowing skills exist means you have a path to automate niche processes as you get comfortable with the tool.
What Breaks? The Research Preview Reality
I'm going to be direct: Co-Work is early. Anthropic themselves compare it to "Claude Code a year ago where it was super buggy and kind of barely worked."
That warning from Anthropic's help documentation is important. If you're in healthcare, finance, or any regulated industry, Co-Work isn't ready for sensitive data. The audit trail doesn't exist. The compliance hooks aren't there.
Other limitations to expect:
- Bugs you'll need to work around (it's research preview)
- Missing features that feel obvious (they're coming)
- Occasional failures that require restarting tasks
- Performance varies—some tasks work perfectly, others struggle
The teams who get value from early tools are the ones who can tolerate rough edges. If you need polish and reliability on day one, wait 6 months.
Why Does Less Tweaking Lead to Better Results?
Claude Code—the engineer version—is massively configurable. Skills, custom agents, hooks, permissions systems, extensive settings. Engineers love that flexibility.
Co-Work intentionally starts simpler. The Anthropic team's advice: "I would probably not customize too much. If you have Co-Work installed and you install the Chrome extension, that's pretty much all you need."
This isn't dumbing things down. It's recognizing that most business tasks don't need complex configuration—they need a capable agent pointed at the right files. Start with the defaults. Add complexity only when you hit walls.
Three Tradeoffs Every Business Owner Should Weigh
- **macOS only (for now):** Windows support is coming, but no timeline. If your team runs Windows, you're waiting.
- **No enterprise compliance features:** Audit logs, data exports, and compliance APIs don't capture Co-Work activity. This is a dealbreaker for regulated industries.
- **Research preview quality:** Expect the rough edges of early software. Tasks will fail. You'll need to retry. Some things won't work at all yet.
The tradeoff calculation: can you tolerate early-stage software in exchange for automating tasks that currently eat hours of your week? For many business owners, the answer is yes. For those needing reliability and compliance, the answer is wait.
How Do You Know It's Actually Helping?
Before you invest time setting this up, here's how to verify it's worth keeping:
- **Time saved per task:** Track a specific workflow before and after. If a 30-minute task becomes 5 minutes, you have data.
- **Error rate comparison:** Is Claude's output as accurate as yours? Sample check 10% of what it produces.
- **Task completion rate:** How often does Co-Work finish what you ask vs. requiring manual intervention?
- **Verification habit:** Are you giving Claude ways to check its own work? (Per Anthropic's tip: "If Claude can verify its own output, the result is going to be way better.")
If you're building an AI strategy for your business, Co-Work fits into the "quick wins" category—tools that deliver value within days, not months. And if you want to understand where AI agents are headed more broadly, I covered the landscape in my piece on why AI agents are ready for real work.
Common Questions
Do I need to know how to code to use Co-Work?
No. That's the entire point. Co-Work wraps the same engine engineers use in an interface that requires zero technical knowledge. You point it at a folder, describe what you want done, and it handles the rest. If you can install an app and choose a folder, you can use Co-Work.
How is Co-Work different from regular Claude in the browser?
Regular Claude can only work with what you paste into the chat window. Co-Work has persistent access to a folder on your computer—it can read existing files, edit them, and create new ones. It's the difference between describing a document vs. handing someone the actual document to work on.
Is my data safe with Co-Work?
Co-Work stores conversation history locally on your computer, which means Anthropic doesn't retain your conversations the way they might with web-based Claude. However, this also means no audit logs or compliance features—so don't use it for regulated data. Your files stay on your machine; Claude reads and edits them in place.
What does Co-Work cost?
Co-Work is included with Pro, Team, and Enterprise Claude plans. There's no separate charge for the feature. Pro plans start at $20/month. If you're already paying for Claude, you have access to Co-Work.
When will Windows support be available?
Anthropic says "Windows coming soon" but hasn't given a specific date. As of January 2026, Co-Work is macOS only. If your team runs Windows, you'll need to wait or use macOS machines for now.
