Inching closer to OpenClaw.
That’s what I keep thinking every time a major AI lab ships a new collaboration feature. Anthropic just released Claude Cowork, and I was saying to a friend today that, for as big a leap forward as it is, it still feels handcuffed compared to Claude Code. And OpenClaw is still going to be closer to Claude Code on the go than Claude Cowork on the go.
Let me explain what I mean.
The Gap Between “Assist” and “Do”
Claude Cowork is impressive. It’s a significant step forward in making AI collaboration accessible to people who aren’t living in a terminal all day. It’s polished, it’s thoughtful, and it brings some of the multi-step reasoning that makes Claude Code so powerful into a more approachable environment.
But here’s the thing — it’s still operating within a sandbox. A nice sandbox, sure. But a sandbox. You can collaborate within the boundaries Anthropic has defined for you.
Claude Code is different. When I’m working in Claude Code, the AI has access to my terminal, my file system, my project structure. It can run commands. It can read logs. It can iterate across multiple files in a codebase and then TEST what it built. That’s not assistance — that’s a genuine working partner.
The gap between those two experiences is bigger than most people realize.
Why OpenClaw Matters Here
I’ve been running OpenClaw for a few months now, and it sits in an interesting spot in this conversation. OpenClaw gives you the Claude Code experience — a full computing environment, tool access, real autonomy — but makes it accessible without needing to be a developer.
Think about it this way: Claude Cowork is Anthropic bringing AI collaboration to the masses within their walls. OpenClaw is what happens when you take that same AI and give it the keys to YOUR world — your email, your calendar, your CRM, your project management, your fleet of servers. No sandbox. Real tools. Real actions.
When I’m on my phone and I need something done, I’m not opening a collaboration workspace. I’m messaging my AI assistant and it’s actually DOING the thing — checking my email, updating my calendar, running health checks on my infrastructure, drafting responses. That’s the “on the go” experience that matters.
The Convergence Is Coming
I think every major lab will eventually ship something that looks a lot like OpenClaw. The trajectory is pretty clear — move from chatbot to assistant to agent to full digital worker. Cowork is a step on that journey.
But we’re not there yet with the official products. And in the meantime, the people who are already operating with full agent autonomy — who have AI that can actually interact with their real systems — are building at a different pace than everyone else.
This isn’t a criticism of Anthropic. They’re being appropriately careful about giving AI real-world agency at scale. That’s responsible. But for those of us who’ve been building this capability ourselves, going back to a sandboxed environment feels like putting training wheels back on a mountain bike.
The Takeaway
If you’re comfortable with the terminal, Claude Code is still the best AI coding experience available. Full stop.
If you’re not a developer but want genuine AI agency — not just chat, but actual tool use across your real systems — OpenClaw and platforms like it are where the action is.
And if you’re watching Cowork and thinking “this is amazing” — you’re right. It is. But it’s also a preview of something much bigger that’s coming. The question isn’t whether AI will operate with full autonomy on your behalf. It’s how soon, and who gets there first.
My bet is on the open ecosystem. It usually wins.
