Meta's Muse hit 2.8 million downloads in 12 days. Here is why the map didn't change color.

Meta launched Muse, a personal AI agent, on 8 September. Twelve days later the measurement firms Apptopia and Appfigures put it at 2.8 million installs, with 1.8 million of those on iOS in the US and Canada against 1.3 million for ChatGPT at the same age, and roughly 642,000 US daily active users against ChatGPT's 231,000 at the same point in its life. Muse sends emails, books travel and completes transactions, and sells at free, $20 and $100 monthly tiers. Meta's shares rose more than 20%, adding about $200 billion of market value, and Jefferies published a figure of $10.8 billion of annualized revenue if Muse reaches a billion users with 3% of them paying.

That is a genuinely big story, and this map's response to it was to add a minor signal that colors nothing. It is worth explaining why, because the reasoning is the product.

Four different kinds of number, and only one of them is evidence here

Downloads are not users. An install is a person tapping a button once. The relevant question for the AI economy is whether they are still using it in March, and nothing available today answers that.

Daily active users are closer, but these are estimates. Apptopia and Appfigures model app activity from the outside; they say so themselves. Meta has disclosed nothing. An outside estimate from a reputable firm, reported by reputable press, is good evidence of roughly what is happening — and it is not a company disclosure, which is the tier this map treats as solid.

The $10.8 billion is a projection built on two assumptions. A billion users and 3% paid conversion. Neither has happened. Our rubric caps analyst projections at minor magnitude for exactly this reason, and it was written before this event, which is the only reason a rule is worth anything.

The $200 billion of market value is price, not evidence. It tells you what investors expect. This map never colors anything on price movement, because doing so would make it a momentum tracker wearing a research costume.

What would actually move it

The signal is filed against a new row in the thesis tracker — consumer and enterprise agents — which we added this week because the evidence finally exists to track it. The row's threshold is deliberately specific: sustained use after the launch window, paid conversion, transaction volume, and unit economics that work, evidenced by companies rather than by estimates.

So the next Muse data point that changes something on this map looks like one of these:

• Meta discloses a subscriber number, a usage figure or agent revenue in a quarterly report.
• Meta or a payments partner discloses transaction volume actually executed by the agent.
• A credible figure appears for tokens consumed per task, which is the number that connects a consumer app to the data centers on the rest of this map.
• Or the opposite: retention falls off a cliff after the launch spike, which would be evidence against the row.

Why the optimistic case here is still the interesting one

Nothing above is a reason to be dismissive. If consumer agents work, they are the first thing in this build-out that connects ordinary people's monthly spending to the trillion-dollar infrastructure question — the one that asks who ultimately pays for all this. An agent that books travel consumes far more inference than a chatbot that answers a question, and it can charge $20 a month for doing so.

That is why this is filed at all rather than ignored. A minor signal is not a dismissal; it is a placeholder with a date on it, so that when the real number arrives there is something to compare it against.

The general rule

A site that colored a company green every time a product launched well would be right often and useless always. The discipline that makes the green rings mean something in three months is the discipline of not awarding one this week.