Anthropic dropped Cowork on Monday, and honestly, it’s one of the more interesting agent plays I’ve seen in a while. The idea is simple: give Claude access to a folder on your Mac, and it can read, edit, or create files in there without you needing to touch a terminal. No coding required.
The company says the whole thing was built in about a week and a half, mostly using Claude Code itself. That’s a recursive flex that’s hard to ignore — AI building AI, and doing it fast.
Cowork is available as a research preview for Claude Max subscribers (that’s the $100–$200/month tier) through the macOS desktop app. It’s essentially the non-developer version of Claude Code, the terminal-based tool that developers have been using since late 2024 to automate programming tasks.
But here’s the kicker: Anthropic noticed that developers were using Claude Code for way more than coding. People were doing vacation research, building slide decks, cleaning up email, canceling subscriptions, recovering wedding photos, monitoring plant growth, even controlling ovens. As Boris Cherny, an engineer at Anthropic, put it on X: “These use cases are diverse and surprising — the reason is that the underlying Claude Agent is the best agent, and Opus 4.5 is the best model.”
So Anthropic stripped out the command-line complexity and built a consumer-friendly interface around the same agentic loop. Instead of typing commands, you just point Claude at a folder and give it a task. The AI formulates a plan, executes steps in parallel, checks its own work, and asks for clarification if it gets stuck. You can queue multiple tasks and let it run in the background — Anthropic describes the experience as “much less like a back-and-forth and much more like leaving messages for a coworker.”
The examples they give are practical: reorganizing a cluttered downloads folder by sorting and renaming files, generating a spreadsheet from receipt screenshots, or drafting a report from scattered notes. These are the kinds of tasks that knowledge workers actually spend time on, and they’re exactly the sort of thing that current AI chat interfaces handle poorly because they can’t touch your local files.
The architecture relies on what’s called an “agentic loop.” When you assign a task, Claude doesn’t just spit out text — it plans, executes, checks, and iterates. That’s a fundamentally different interaction model from the back-and-forth chat pattern we’ve gotten used to. It feels more like delegating to a junior colleague than having a conversation.
Under the hood, Cowork runs on the same Claude Agent SDK as Claude Code, so it shares the same underlying capabilities. Anthropic explicitly says it “can take on many of the same tasks that Claude Code can handle, but in a more approachable form for non-coding tasks.”
Now, I have some reservations. The folder-based sandbox approach is smart for security, but it’s also limiting. You can’t have Claude work across multiple folders or access system-level files without explicit permission. That’s fine for contained tasks, but it means Cowork isn’t going to replace full-blown automation tools like Keyboard Maestro or Hazel for power users.
Also, the price is steep. Claude Max at $100–$200 per month is aimed at professionals who already see high value from AI tools. For casual users, that’s a hard sell. But if you’re already paying for Copilot or ChatGPT Pro, the comparison becomes more interesting.
The broader context here is that Anthropic is betting that enterprise value lies in practical, hands-off productivity — not just chat. Microsoft has Copilot, Google has Gemini, but neither has a folder-based agent that can actually manipulate your files without you watching over it. Cowork feels like a direct shot at that market.
I’m curious to see how well it handles messy real-world data. Receipt screenshots from different angles, folders with inconsistent naming conventions, notes written in bullet points mixed with full sentences — these are the kinds of edge cases that trip up most AI tools. If Cowork can handle them reliably, it could be genuinely useful.
The fact that it was built in under two weeks using Claude Code is impressive, but it also makes me wonder about polish. Research previews are notoriously rough around the edges. I’d expect bugs, weird behaviors, and the occasional refusal to do something obvious. That’s the nature of the beast.
Still, the direction is right. We’ve been stuck in chat interfaces for too long. Agents that can actually do work in your files — without hand-holding — are where the real productivity gains are. Anthropic seems to understand that, even if the execution is still in preview.
Comments (0)
Login Log in to comment.
Be the first to comment!