Meta confirms Muse borrows OpenClaw's design but says the model was built from scratch

Meta confirms the resemblance was intentional

Early users of Meta's Muse assistant had been guessing that its strong performance came from something familiar: OpenClaw, the breakout open-source AI agent, repackaged for a mainstream audience. According to TechCrunch, Meta has now conceded that the comparison is fair — at least at the product level.

Nat Friedman, head of product at Meta's Superintelligence Labs (MSL) and former GitHub CEO, wrote on X that Muse was "definitely heavily inspired as a product by OpenClaw." At the same time, he maintained that Muse itself was "built from scratch."

The evidence that started the thread

Friedman was replying to a post that had gone viral among AI watchers. Ansh Nanda, an AI app co-founder, argued that Muse was "LITERALLY OpenClaw for normies." Nanda had shared a conversation with Muse in which the assistant said the overlap between its own system files and OpenClaw's was not a "coincidence" but a "match."

Others on X soon added their own findings, as TechCrunch reports. Muse ships with a SOUL.md file — a plain-text configuration file written in Markdown that defines an AI agent's personality, tone, communication style, values, behavioral boundaries and expertise — the same convention OpenClaw uses. One observer noted that the files did not merely share names across the two assistants; their content was almost entirely the same.

Friedman's explanation

Confronted with the matching file names and near-identical SOUL.md content, Friedman did not dispute the observations. Asked why Muse had copied the exact file names that make up OpenClaw's agent workspace, he replied that Meta believed Peter Steinberger, OpenClaw's creator, "got those things exactly right."

Friedman added context of his own: after trying OpenClaw in January, he bought hundreds of Mac minis for the MSL team, and many on the team became devoted users. He called Steinberger a genius whose "harness was pioneering from the jump," and said the aim for Muse was to build something like OpenClaw that Meta could make safe, secure and easy enough to use at a scale of billions of people. A Meta spokesperson reached by TechCrunch pointed to Friedman's statement and said the company had nothing further to add.

A familiar playbook

Because OpenClaw is an open-source project, Meta looking to it for inspiration is not unusual — particularly given its success, which reportedly prompted OpenAI to hire Steinberger earlier this year. Even so, TechCrunch notes that the episode fits Meta's well-documented practice of identifying promising products and copying their best features, most famously with Snapchat's Stories format.

The commercial stakes are considerable. According to TechCrunch, Muse recently reached No. 1 on the U.S. App Store, and available data suggests it is outpacing ChatGPT's launch once differences in platforms and market availability are compared directly.

Why it matters

Friedman's admission draws a careful line between the model and everything wrapped around it. Meta's position is that Muse's underlying system is original, while its product design, file layout and persona configuration deliberately mirror OpenClaw's — including content that users found nearly identical. Whether that distinction matters to consumers is an open question, though the app's chart performance suggests it has not hurt adoption so far.

The episode also shows how open-source agent design is becoming a shared vocabulary. Conventions like a Markdown-based SOUL.md file were popularized by an independent project, and Meta treating them as the correct default effectively validates them as a de facto standard — while raising familiar questions about how much credit flows back to the original creators when their patterns are absorbed into products built for billions.

Finally, it demonstrates a new form of scrutiny. Users diagnosed Muse by inspecting its system files and asking the agent about itself, and the resulting thread forced a public clarification from one of Meta's most senior AI product figures. As AI assistants proliferate, companies should expect their internals to be probed this way — and to be asked, plainly, what is original and what is borrowed.