Nat Friedman says Meta built Muse from scratch
Peter Steinberger, whose harness started the argument, backed Friedman's account and said kudos.
Ansh Nanda posted that Meta's Muse is "LITERALLY Openclaw for normies" and added "It's actually just openclaw." Nat Friedman answered directly, saying Meta built Muse from scratch but that it is "definitely heavily inspired as a product by openclaw."
@steipete is a genius and his harness was pioneering from the jump
Muse is LITERALLY Openclaw for normies.
The detail in Friedman's reply is the part worth keeping. He says he used OpenClaw in January, then bought hundreds of Mac minis for the MSL team, and that lots of them fell in love with using OpenClaw and other personal agents. He credits Peter Steinberger by handle, calling him a genius and the harness pioneering from the jump, and says a lot of people were inspired by it. The stated goal for Muse was to build something like OpenClaw that Meta could make safe and secure and easy to use and scale to billions of people.
Steinberger backed the account
Steinberger, whose harness is the thing under discussion, did not litigate it. He wrote that the "Meta uses OpenClaw" story going around is wrong, that Meta built its own agent while being inspired, and that Nat and his team did a great job. Guillermo Rauch's response to the exchange was four words, "Game recognizing game."
They built their own agent, being inspired. Nat and his team did a great job. Kudos!
So the copying accusation collapsed within a day, with the accused and the supposedly copied party agreeing on the same version of events. Nobody in the sources disputes that Muse is Meta's own code.
What this actually tells you
For working engineers, the interesting claim is not about provenance, it is about product shape. A personal agent harness, built by one person and used by a small number of people who liked it, set the template that a large lab then went off and rebuilt with its own constraints around safety, security and scale. The hundreds of Mac minis are the tell. Friedman's team did not evaluate OpenClaw in a doc, they ran it on real hardware until enough of them wanted the thing daily, and the product followed from that.
That is a different path from the usual one, where the lab ships a harness and the community bends it. Here the direction ran the other way. If you have been treating personal agent setups as hobby tooling that the big products will eventually subsume, the sequence here suggests the hobby tooling is where the shape gets discovered first, and the subsuming is the second step, not the first.
What Muse actually does differently from OpenClaw, and where those safety and scale constraints bite, is not something the sources cover.