Situation report: Meta has confirmed that its hit AI assistant Muse draws heavily on OpenClaw, the open source personal agent that took off earlier this year. According to TechCrunch AI, Nat Friedman, head of product at Meta Superintelligence Labs (MSL), said on X that Muse was “definitely heavily inspired as a product by OpenClaw.” He also said the team built it “from scratch.”
This matters for the open source agent ecosystem. The biggest consumer AI launch of the year openly borrows from a community project, down to its file structure.
🎯 What Happened
It started with a viral X post. Ansh Nanda, co-founder of an AI app, claimed that “Muse is LITERALLY OpenClaw for normies.” He shared a conversation in which Muse itself said the similarity between its system files and OpenClaw’s wasn’t “a coincidence” but “a match.”
The AI crowd on X quickly piled on with more findings:
- Same workspace files. Muse’s agent workspace uses the same file names as OpenClaw’s.
- A SOUL.md file. Muse ships with the plain text Markdown file that sets an agent’s personality, tone, values, boundaries, and expertise. It’s an OpenClaw signature.
- Nearly identical content. One user found that the files didn’t just share names. Their contents were almost entirely the same.
When someone asked Friedman directly why Muse copied the exact file names and had “nearly identical content” in SOUL.md, he didn’t push back. He replied that “we thought that Peter [Steinberger, OpenClaw’s creator] got those things exactly right.”
🧭 Friedman’s Explanation
Friedman, the former GitHub CEO who joined Meta last year alongside Chief AI Officer Alexandr Wang, was open about where the idea came from. After he tried OpenClaw in January, he said he “bought hundreds of Mac minis for the MSL team,” and a lot of the team “fell in love with using OpenClaw (and other personal agents).”
He called Steinberger “a genius” and said his harness “was pioneering from the jump.” A harness is the scaffolding around a model: the files, tools, and instructions that turn a chatbot into an agent that actually does things for you.
The stated goal: “build something like OpenClaw that we could make safe and secure and easy to use and scale to billions of people.”
A Meta spokesperson pointed TechCrunch AI back to Friedman’s statement and had nothing to add.
⚖️ Assessment
What stands out is how unbothered Meta is about all this. It isn’t denying anything. It’s owning it.
That fits a pattern. Meta has a long record of spotting a winning product and copying its best features. The best-known case is the Stories format it lifted from Snapchat. The difference this time is that OpenClaw is open source, so borrowing its design is permitted by design. OpenClaw was also successful enough that OpenAI hired Steinberger earlier this year, so the playbook has already been validated twice over.
The strategic read:
- Distribution beats invention. OpenClaw proved personal agents work. Meta is proving they can reach the mass market. Muse recently hit No. 1 on the U.S. App Store, and data suggests it’s outpacing ChatGPT’s launch when you compare platforms and market availability directly.
- SOUL.md is becoming a de facto standard. When the world’s largest social company adopts an open source config format almost word for word, other agent builders will likely follow. Plain text personality files could become as common as README files.
- Open source creators carry the risk. Steinberger’s design now powers a Big Tech product at a scale his project couldn’t reach alone. That’s either validation or a warning, depending on where you sit.
📋 What to Watch
- Community reaction. Expect a heated debate over whether Big Tech owes anything beyond praise to the open source projects it builds on.
- Competitive response. OpenAI now has OpenClaw’s creator in-house. A consumer agent from that direction wouldn’t be a surprise.
- Safety claims. Meta says Muse makes the OpenClaw model “safe and secure.” Security researchers will test that claim fast, especially since the underlying design is now public knowledge.
- For builders. If you’re working on agents, study the OpenClaw harness. Meta just told you it’s the template worth copying.
The personal agent race just moved from hobbyist Mac minis to billions of phones. The full story, with Friedman’s complete comments, is at TechCrunch AI.