Meta's Muse personal agent has below 1 million daily active users, but it is already reshaping where AI money flows. Ramp data shows frontier model spending dropped from 53% to 45% of total AI dollars between August and September 2026. Ramp chief economist Ara Kharazian attributes this to competition from other US models, not China or open source.
The author predicted Muse would fail on two counts: privacy friction limiting integrations like Gmail, and its reliance on non-frontier models leaving it exposed. The first concern may still stand. The second was wrong. Meta built a functional, competitive AI app without GPT-4-class intelligence, and that fact has direct consequences for OpenAI and Anthropic's revenue trajectories.
The full piece is worth reading for its analysis of model routers, 'tokenmaxxing' fatigue among enterprise buyers, and the structural shift toward cost-optimized AI stacks. The author's willingness to document his own misprediction makes the underlying argument harder to dismiss.
[READ ORIGINAL →]