Meta's Muse, Amazon and the intent economy: two days into the agent era

I start this written record of SIGNAL on day two of what everyone called the agent era. The day before, OpenAI launched dots at its DevDay and Meta's agent, Muse, was already climbing the charts. Twenty-four hours later the story had teeth. The three things that come with every platform shift showed up at once: growth, money, and someone getting hurt.

I take them in that order, because the order matters. Growth attracts money. Money attracts gatekeepers. And gatekeepers always end up talking about safety.

Growth: Muse overtakes ChatGPT in ten days

The numbers first. Muse passed 3.4 million downloads, according to Sensor Tower data reported by TechCrunch. The app had been number one on the US App Store free chart for nearly two weeks. It took ten days to overtake ChatGPT. Sensor Tower, cited by Bloomberg, counted 902,000 downloads in its first six days, and its Instagram integration had just gone live.

It is a strange sentence to write. The fastest growing app on the planet at that moment is an agent that does your shopping while you sleep.

I stop on this point because it says something about what people want from AI. For three years, consumer AI was conversational. You asked, you got an answer, then you did the work yourself. Muse flips the burden. The user states an intent and the agent executes. Perceived value moves from the answer to the action. Muse's growth shows the public was ready for that shift well before most companies planned for it.

Money: the toll on intent changes hands

Wall Street noticed. Bank of America published a note, by analyst Wamsi Mohan according to 9to5Mac, framing Muse as a new risk to Apple's services revenue. The analyst's sentence deserves to be read in full: whoever owns the agent collects the routing economics that used to flow to operating systems, search and app stores.

Let me translate. Every time you searched, clicked and bought through Apple, someone in the middle took a cut. The search engine sold your query to advertisers. The app store took its commission. The operating system controlled the entry point. That levy on the path between an intent and a purchase is what I call the intent economy. It has funded a large part of consumer tech for twenty years.

Now the agent picks the merchant. The agent becomes the storefront.

Muse is already wired into real commerce rails. 9to5Mac mentions integrations with Shopify and Shop Pay, Expedia and PayPal. The agent does more than advise. It holds your payment methods, which makes it a new kind of gatekeeper. And a gatekeeper that bothers other gatekeepers.

Amazon blocks the agent that buys everything

Amazon seems to have decided it does not like this. According to reports this week, Amazon blocked Muse from its platform, citing concerns about agentic AI.

The whole story fits in that one headline. The company that built its empire on being the place where you buy everything is now afraid of an AI that buys everything for you. Amazon owns the customer relationship, its internal search data, the ad inventory on its own results pages. An outside agent that arrives, compares and buys without looking at sponsored products strips out part of Amazon's margin. Blocking it is a rational defensive move. Mostly, it reveals how much value is at stake.

The injury: a flaw in an agent reaches further

There is a security angle too, and nothing about it is theoretical. An independent researcher found a vulnerability in Muse and reported it through Meta's bug bounty program. Meta rated it severity 2, the third level on a five-point scale, for an issue that could have had significant impact. No evidence it was exploited.

A word on bug bounties. A company pays outside researchers who find flaws and report them privately before going public. It is a sign of maturity, and the system working here is good news. The message stands anyway. An agent holds your email, your files, your payments. A flaw in a chatbot exposes a conversation. A flaw in an agent exposes everything the agent is allowed to do. The attack surface grows with the permissions.

Is this real, or launch-week noise?

I ask honestly, because launch weeks manufacture noise. The downloads are real. The analyst note is real. Amazon's block is real. The vulnerability is real.

What is not real yet is revenue. Meta stock had its best month in thirteen years on the back of this product, according to Kobaran, and the app has no reported revenue. This is launch accounting. Business accounting comes later, and it is stricter.

So I hold two ideas at once. The market is right to see a platform shift. It is ahead of the economic proof.

Three prices, three theories of power

The name of the game this week is ownership of intent. I look at pricing, because it tells you exactly where each company thinks the power sits.

• OpenAI includes dots in existing subscriptions (Pro and Business Premium), at no extra cost, on its best model. The agent is a feature of the subscription.
• Meta makes Muse free to start, then charges 20 dollars a month for a weekly token allowance (500 million tokens per week), and 100 dollars for the top tier (3 billion). The agent is sold like a phone plan.
• xAI shipped Grok Bot first, in August, and bundles it inside Cursor and SuperGrok subscriptions with no standalone price. The agent rides inside someone else's product.

Google, for its part, reserves Gemini Spark for Google AI Pro and Ultra subscribers.

Three companies, three theories of where the money goes. OpenAI bets the agent strengthens a subscription you already pay for. Meta bets usage itself can be sold, like mobile data. xAI bets on distribution. Saganote's analysis captures the philosophical difference well: Muse is built around the person, Grok Bot around the job, like a teammate, and dots around an ongoing responsibility inside ChatGPT.

For a brand strategist this is a textbook case. A pricing model is a positioning statement. It says what you believe you are really selling.

The detail that says a lot

OpenAI named its agent dots. As it happens, the domain dot.com was acquired by xAI in July, and it redirects straight to Grok's download page. OpenAI has not commented.

A brand own goal, or the pettiest troll in tech history. Either way it tells you how personal this race has become. It also restates a rule I apply on every naming project: check the domains before you announce the name.

Why a business leader should care

Most leaders I know are not buying or selling AI agents this week. They should still pay close attention, for two reasons.

The first click is no longer human

The customer journey just changed its first click, and that click is not human. If the agent picks the merchant, being findable by agents becomes as important as being findable by people.

Search engine optimization had a twenty-year reign. It shaped websites, content, marketing budgets, whole professions. Optimization for agents is arriving now, and nobody knows its rules yet. I can describe what any serious agent will need to find in a brand: a clearly described offer, prices that match from one source to the next, structured information a machine can read, verifiable proof, a reputation that holds when cross-checked. A vague brand gets misread by a person. It gets skipped by an agent.

Trust becomes the constraint

The day before, the news was that OpenAI had shelved a model over deceptive behavior. Today it is a vulnerability in Muse and Amazon blocking agents. The pattern is clear. The more an agent can do on your behalf, the more it has to prove it will not do it wrong.

Brands that want agents to carry their products will have to be verifiable, sensible and safe. Exciting brands lose this round. Reliable brands win it. I find the reversal healthy. For years, attention rewarded noise. An economy where machines coldly compare promises will reward consistency.

What changes, concretely

For businesses, the storefront moves from the app to the agent. The website, the product page, the brand page all remain necessary, and they become data sources as much as experiences.

For consumers, convenience now comes with an audit of their credentials. Handing a card to an agent is an act of trust that requires knowing exactly what it can do.

For Apple, Google and Amazon, the platform owners, the question is blunt. Keep every hardware sale while losing the intent economy, or fight their own users by blocking the agents they chose.

Scoreboard after two days

1. Muse reaches 3.4 million downloads, the fastest growing consumer AI app since ChatGPT.
2. Wall Street sees a threat to Apple's services revenue.
3. Amazon blocks it, a researcher finds a flaw, and the trust battle is officially open.

My reading fits in one sentence. The agent era will be decided less by how powerful the models are than by whom people trust with their intent, and at what price.

The noise filtered. The signal kept.