Meta's Muse and the ad business hiding inside it

Meta's AI personal agent Muse launched 8 September and hit 730,000 US downloads by day five, according to Sensor Tower - more than ChatGPT, Claude, Polymarket or Kalshi pulled over the same stretch. Digiday's Seb Joseph and Krystal Scanlon argue the more popular Muse gets, the harder it is to rule out advertising inside it: Meta built its $243 billion ads business on human attention, and an agent that executes tasks rather than scrolling cannot be reached with a banner. The plausible ad slot is a placement inside the stripped-down data an agent actually reads, not a page - Time already converts its site to markdown and sells one "balled" ad inside those files for a bot to read. What nobody has settled is what an agent does with an ad once it reads one. Farhad Divecha, group CEO of AccuraCast, puts the stakes plainly: "Our world could suddenly become very small... They're either in with Muse (or other personal agents) or not." Meta insists Muse conversations stay separate from its ad systems and is building subscriptions plus a merchant-paid transaction fee instead - but its 2026 AI capex of $135 to $145 billion already consumed 98% of operating cashflow in Q2, all paid for by the ad business Muse is supposed to sit apart from.

The crux

Whether the personal agent becomes the next advertising interface, and whether Meta can keep a data-rich product walled off from the ads business that is funding it, is the defining commercial question for every publisher whose content an agent might one day read, summarise or recommend.

Sell-side read

Publishers should treat agent-readable content as the next yield surface - the ones who structure inventory so buyer and personal agents can discover and transact against it (Time's markdown model, AdCP/CoMP-style standards) position for premium access before a gatekeeper owns the layer.

Watch

Whether Muse keeps its no-ads promise through the next two earnings cycles as AI capex keeps consuming the ad business's cash.