Ads are the business. Signal is the constraint.

Ad revenue is $59.4B of $60.8B. Every other thing Meta does, the glasses, the models, the $130 billion of capex, is funded by that line. So the honest question for any partner is not what we would like to build together. It is what moves that line.

And the line is moving in a specific way. Impressions grew 14 percent and price per ad grew 12 percent. Impressions are inventory, and inventory growth has a ceiling: there are only so many surfaces and so many hours. Price per ad is relevance, and relevance is a function of signal. It is the half of the growth that can keep compounding, and it is also the half whose input is getting harder to source every year.

There are two ways to answer a signal shortage. Infer harder, which is what the whole industry is doing and which regulators, operating systems and people are all pushing back on at once. Or be given the signal. A declared intent from somebody who chose to publish it is more accurate than any inference, current rather than modelled, and it is the only kind of signal that gets easier to obtain as privacy law tightens instead of harder.

That is the layer we build, and it is the reason this is a partnership rather than a pitch. Meta has the demand, the auction, the ranking and the reach. What it cannot manufacture is a person's willingness to say what they actually want, to a company they are wary of. That has to be earned by somebody standing on the person's side of the line.

Read those six lines together and a single requirement runs under all of them. An agent that manages somebody's finances, health, household and relationships needs that person to hand over the four most sensitive data sets they have. An ads model that predicts the whole sequence of actions around a purchase gets dramatically better with declared context than with inferred context. Glasses that see what the wearer sees need permission from the wearer and social licence from everyone else in the room.

Every one of those is the same requirement, and it is the one thing capex cannot buy. Compute can be bought, models can be trained, distribution already exists. A person's willingness to say what is actually going on in their life, to a company they are wary of, has to be earned, and it has to be earned by somebody standing on their side of the line.

There is a real tension here and it is better said out loud. Meta's personal agent will run on Meta's infrastructure under Meta's incentives. Ours runs on the person's. Those are different products and we are not pretending otherwise. But they are not competitors for the same person, because a large number of Americans will never hand that data to a platform, at any level of capability, and no amount of model quality changes their mind. Serving them requires somebody whose entire business is being on their side. That is the segment we open, and it is a segment Meta structurally cannot reach alone.

Which is why the honest framing is not a vendor pitch. It is: you are building the agent, and we are building the reason a cautious person lets one near their life.

Muse Spark 1.1 plans, calls tools, drives interfaces and delegates to subagents. That is not an assistant. That is an agent acting on somebody's behalf, and an agent acting on somebody's behalf needs their real context: their calendar, their money, their health, their family. The announcement is a model release and says nothing about where that context lives, what is retained, or what a person can take back. We are not reading anything into the silence. We are pointing out that the question is still open, and that it is the question that decides whether anyone hands the context over at all.

🤫 Private Agent One already answers it, and answers it in a way that costs Meta nothing to try. Bring your own API and bring your own compute are native capabilities, not integrations we would have to build: the person supplies the key, the quota is theirs, and the agent runs on hardware they own. The context never has to leave. The model gets called, not fed.

Meta is not a partner, an investor, or a customer, and no one at Meta has seen this before it was published. Workstream 00 needs nothing from Meta beyond an API that is already in public preview, which is why it is the one marked ready. Workstreams 01 through 04 are proposals and are marked as proposals. The consented signal layer they depend on is partly built: the consent protocol, the per-field grants and the receipts are real; the auction, the advertisers and the payouts are not.