Inside Meta's months-long grind to turn Muse into a hit
Meta’s Muse AI agent followed a difficult period for the company, which had faced doubts about its position in the AI race. Its launch has helped improve employee morale and coincided with a rise of more than 20% in Meta’s stock price.
The project began in September 2025 under Nat Friedman, who leads product and applied research in Meta’s superintelligence group. The agent was initially called Hatch, and Meta also recruited AI talent through acquisitions, including the leaders of startup Dreamer.
Development remained challenging close to launch. Two weeks beforehand, Muse still struggled to produce fluent English, according to someone familiar with the work. The team was relieved when it began performing well on coherence tests.
Thousands of Meta employees tested Muse over several months and suggested ways it could be used. The team also worked to make its responses more considerate and to help it use internet tools effectively.
The launch offers evidence that Meta is still competing in AI after being viewed as behind a year earlier. The article does not describe a specific next milestone for Muse.
Key points
- Nat Friedman’s team worked on Muse from September 2025, when it was codenamed Hatch.
- Thousands of Meta employees tested the agent and helped shape its capabilities.
- Muse’s launch has lifted morale and coincided with a more than 20% rise in Meta shares.