Article
Meta How to create a hit product Muse? Fewer than 200 people developed it, and it launched only after passing all 100 tests
Dongcha Beating AI news brief:Meta Chief AI Officer Alexandr Wang discussed Muse's development process in a recent interview. This personal AI Agent, which can handle tasks for users, took 7 months to go from prototype to launch, with fewer than 200 people on the core model and product team.
Earlier this year, Anthropic was far ahead in the coding Agent space, and many in the industry believed that programming was the most important application area for advanced AI. However, Wang and former GitHub CEO, now Meta AI product lead Nat Friedman, were more optimistic about personal Agents. Wang believes that Meta's applications are used by approximately 3500000000 people every day, and that building AI products for ordinary users is the company's strength.
Meta As early as 2025, Meta proposed the direction of "personal superintelligence," and OpenClaw gave this concept a more concrete product form. This year February, after deeply experiencing OpenClaw, Wang and Friedman each submitted a memorandum to the board. Friedman wrote, one of which was titled "Trust Is a Must," emphasizing that personal Agents must earn users' trust. The team then produced a Muse prototype in just or weeks, including early versions of the mascot Jolly and the dedicated hardware Muse Charm. However, the product was highly unstable at the time: it was occasionally impressive, but most of the time it could not even complete basic tasks.
Over the following months, the team repeatedly refined the model. They identified 100 essential capabilities, designed tests for each, and set launch thresholds. Some model versions had already passed approximately 80 tests, but approximately 20 remained below standard and required further training. The evaluation was not passed until the specially trained Muse Spark 1.3 version met all requirements.
Meta Senior leadership views Muse as one of the year's most important products, yet never significantly expanded the R&D team. Wang explained that products at large companies can easily become a collection of ideas from many people, with every product manager wanting to add something, ultimately resulting in an unfocused mishmash.Muse From the outset, the aim was to maintain a unified product experience, with Friedman leading the review process.
The AI research team managed by Wang follows a similarly lean approach. He disclosed that he currently has more than 200 direct reports.TBD Researchers on the 200 team largely report directly to him, without multiple layers of people-management roles. Each group still has a technical lead responsible for determining technical direction. Wang acknowledged that he may not be the best manager for these