Meta's Muse AI agent has become a fresh example in the debate over whether frontier model performance still determines commercial success. In a new Stratechery article, founder Ben Thompson said the AI sector is now facing five forms of overhang: model capability, product, pricing, capital, and safety. He argued that raw model leadership no longer guarantees a durable business moat.
According to Sensor Tower data reported by Bloomberg, Muse logged more than 902,000 downloads in its first six days, compared with 773,000 for the Meta AI app over the same period after its launch. By September 21, Muse had also reached the No. 1 position on the US free-app charts on both Apple's App Store and Google Play, displacing ChatGPT.
Thompson argued that Muse Spark 1.3 is not the most advanced model, but is still good enough to support a highly sticky personal AI agent. He said this reflects a broader shift as model and application layers move toward modular separation, with the next competitive battleground likely to be applications rather than models alone.
He also said Microsoft has moved its Copilot Cowork offering away from being tied to Claude and toward a multi-model architecture. Copilot Cowork offers GPT and Claude models, while Microsoft has said its architecture can accommodate open-weight models. Thompson used that example to argue that once AI performance becomes "good enough," competition shifts toward speed, convenience, customization, and data handling rather than pure model quality.
Thompson said this dynamic is especially important for personal AI agents, which he described as far stickier than chatbots because users place more of their information and daily routines inside them. He added that Anthropic and OpenAI should devote more resources to product development, while their high pricing reflects supply shortages rather than lasting cost advantages. He said that if progress slows, compute now used for training, reinforcement learning, and research and development could be redirected to inference, putting downward pressure on prices.
Article edited by Ysi Chen
