Claude just killed OpenClaw
TL;DR
Anthropic just shipped a Claude version of the “claw” experience — Wes argues Claude Co-work plus Dispatch now covers “90%” of what most people wanted from OpenClaw, echoing Ethan Mollick’s line that it delivers the utility without the same security paranoia.
OpenClaw was a real phenomenon, not a niche dev toy — created by Peter Steinberger, it became the fastest-growing open-source project of its moment, reportedly caused Mac Mini shortages, and got so big that Andrej Karpathy now talks as if everyone will have their own “claw.”
Anthropic’s edge is polish, safety, and onboarding — Claude’s version is hosted, Mac desktop-based, sandboxed, and wired into connectors like Slack and Google Workspace, so beginners don’t have to touch CLI setup, config files, or raw API keys.
The big strategic difference is open vs. closed — OpenClaw still matters because it’s free, open source, cross-platform, and model-agnostic, which means you can swap in whatever LLM fits the task, from Anthropic to Grok 4.20 for real-time search-heavy work.
Anthropic is shipping at absurd speed — Wes cites Pavle Hrin’s tally of 74 Claude releases in 52 days, with Boris Cherny’s team behind major building blocks like MCP, Skills, and Claude Code, which helps explain why features from the wild keep showing up in Claude so quickly.
Claude’s computer use is finally useful, but still imperfect — in Wes’s test, it successfully opened Photopea, created multiple “Hello World” thumbnail drafts, and exported one, yet it also took over his whole screen and needed nudges, reinforcing that it’s powerful now but not fully polished.
The Breakdown
The question: did Anthropic just kill OpenClaw?
Wes opens with the blunt version: Anthropic shipped features that make Claude feel “almost identical” to OpenClaw, the thing people were obsessing over under names like Moldbot, Clawbot, and OpenClaw. He uses Ethan Mollick’s summary as the cleanest framing: this is 90% of what many people wanted from OpenClaw, just with way less chance of your personal data ending up somewhere cursed.
Why OpenClaw blew up so hard
He quickly rewinds the saga: OpenClaw is an open-source AI agent that runs on your own hardware, builds context about your life and work, and can be messaged through Telegram or WhatsApp like a persistent digital operator. Wes says once you’ve lived with that kind of agent memory, going back to a normal chatbot feels like losing “a good friend” and returning to a faceless assistant.
From viral project to a whole new category
Wes reminds you how massive the thing got: Peter Steinberger built it, it went viral in early 2026, caused literal Mac Mini shortages, and even got broad mainstream coverage from places like Fortune and Forbes. After Anthropic sent a friendly cease-and-desist over the original naming, rebrands followed — Moltbot, then OpenClaw after crypto scammers made things weird — while Steinberger himself went to OpenAI, even though OpenClaw remained its own project.
Anthropic’s shipping spree and the rise of Dispatch
Then comes the key pivot: Anthropic has been shipping nonstop. Wes cites Pavle Hrin’s infographic claiming 74 Claude releases in 52 days, and points to Boris Cherny’s team as the group behind MCP, Skills, and Claude Code — three of the biggest pieces of the current agent stack. The new pieces that matter here are Co-work and Dispatch: Claude can now use your computer, and you can assign tasks from your phone and let it finish them on your desktop.
Computer use goes from comedy to credible
Wes spends a minute on Anthropic’s own computer-use lore, including the old demo where Claude accidentally stopped the screen recording and, in another case, wandered off to browse Yellowstone photos. That’s what makes this launch feel notable: computer use used to be “comically bad,” but now Claude can open apps, navigate websites, fill spreadsheets, and act more like a real desktop user instead of a backend script.
His live test: making a thumbnail in Photopea
He walks through the setup on Mac: enabling computer use in the Claude desktop app, granting accessibility and screen recording permissions, and choosing which apps Claude can or cannot touch. Then he gives it a real task — open Photopea in an incognito Chrome window and make a YouTube thumbnail that says “hello world” with a smiley face — and watches it muddle through multiple drafts, briefly hijack the full screen, and eventually export a PNG.
Why this absolutely hurts OpenClaw — but doesn’t replace it
Wes’s verdict is nuanced: for average users, Claude’s version is safer, easier, more beginner-friendly, and heavily subsidized through Anthropic’s consumer plans, which can be dramatically cheaper than paying equivalent API token costs directly. But OpenClaw still has the serious advantages: it’s open source, works on Linux/Mac/Windows/headless servers, supports any LLM, and gives you much deeper long-term memory and control.
The real moat: ownership, memory, and personal data
His strongest defense of OpenClaw is that it can become deeply personalized in a way Claude likely won’t, at least not easily. He gives the example of uploading his own bloodwork and medical PDFs so an agent can answer questions based on his actual health data — something he sees as incredibly valuable, but also exactly the kind of gray-area liability Anthropic will be cautious about because of privacy laws and regulatory scrutiny.