A visitor who never scrolls, never clicks an ad, and never falls for a countdown timer has arrived. Meta built it.
Picture a shopper who ignores your banner. It dismisses the cookie pop-up instantly. It skips the “you might also like” carousel and heads straight for checkout.
That shopper now has a name: Muse. Meta’s personal AI agent is less a new chatbot than a new kind of web user.
What Muse does
Meta launched Muse on Tuesday, September 8. The agent targets adults in the US who want help with schedules, shopping and long-term goals. Users message it in a dedicated app or in WhatsApp.
Meta says Muse opens a browser, fills in forms, and negotiates for the user, according to the Associated Press.
It reaches into email, calendar, payments, health, shopping and the smart home. Meta modeled it on the open-source agent OpenClaw. Reuters reported those details at launch.
The packaging deserves a mention. Meta wrapped the agent in a cream-colored, doll-like design that makes a data-hungry product feel like a toy.
The vision behind it
Muse didn’t appear from nowhere. Mark Zuckerberg laid out the idea in a 6,500-word essay a month earlier.
He wrote that everyone will have a highly capable personal agent. It will work around the clock on relationships, health, career, finances, and home life, per the same AP report.
That promise explains the product. Meta doesn’t want to sell a smarter search box. It wants to sit between each person and every task they do online.
From chatbots to errand-runners
A chatbot hands you an answer. An agent finishes the errand.
Developers already felt that jump. Coding agents moved from pasted snippets to shipping working applications.
Muse brings the same shift to daily life. It books, buys, and completes paperwork.
At Connect on September 23, Meta pushed further. TechCrunch reported that Muse will arrive on Meta’s AI glasses. On the desktop, it will operate any app, much like agents from OpenAI and Anthropic.
Meta also built a new model, Muse Realtime Avatar, to power a face for the agent.
The three tolls Muse never pays
The web charges visitors in three currencies. I call them tolls.
- The attention toll. Display ads pay for a glance. An agent never glances.
- The friction toll. Pop-ups, consent walls, and “create an account” nags slow people down. An agent clears them in a blink.
- The persuasion toll. Fake scarcity, urgency timers, and upsells work on tired humans. They do nothing to software.
Each toll funds part of the open web. Shrink the number of human visitors and each toll earns less.
Who loses first
Not every site faces the same risk. The damage depends on what a page sells.
Ad-funded publishers sit closest to the edge. Their revenue depends on human eyes, and agents bring none.
Comparison and affiliate sites come next. An agent can read ten reviews in seconds and skip the click that pays the commission.
Stores hold up better. An agent that completes a purchase still pays the merchant.
Brands with strong names also hold up. People tell agents to buy a specific product, and the agent obeys.
This gap matters because it shows where the money moves. It shifts from catching attention to winning the agent’s shortlist.
Publishers face a hard choice
Sites can block agents and risk vanishing from the answers people trust. They can welcome agents and watch old revenue models weaken.
Neither path feels safe. Blocking costs visibility. Welcoming costs ad income.
Expect a middle road to emerge. Publishers will license content to agent makers, charge for access or demand a credit link. The terms don’t exist yet, and the first deals will set the pattern.
Wall Street already bet on it
Investors didn’t wait. Forbes reported that Muse passed 2.5 million downloads in about two weeks. Meta shares jumped 11% in one session.
The same story flags a catch. Contractors handle some calls through a “human concierge” feature during trials, which raises doubts about scaling profitably.
An agent that quietly leans on people behind the curtain is still a demo.
How Meta plans to earn money
Pricing shows the plan. A basic version costs nothing. Paid tiers run $20 and $100 a month for heavier use.
That structure mirrors a familiar playbook. Free access builds the habit, and the heavy users pay.
Meta also owns the audience. Billions of people already use its apps, so Muse starts with a distribution edge no startup can match.
The engine under the hood
Muse runs on Muse Spark, Meta’s agent-focused model. Meta shipped version 1.3 on September 2, and Dutch Startup reported that it needs roughly 20% fewer tool calls and 25% fewer tokens for the same tasks, based on Meta’s internal tests.
The price stayed flat. Capital & Compute tracked $1.25 per million input tokens and $4.25 per million output tokens, unchanged since July.
A discounted endpoint costs far less. In exchange, Meta uses the prompts and completions developers send.
Efficiency matters here. An agent that runs all day burns far more tokens than a chatbot that answers one question. Cheaper runs make that model affordable.
Meta isn’t alone. OpenAI just introduced always-on dots agents that keep working after you close the chat window.
The trust problem
Handing an agent your payments and inbox carries real risk. Meta’s own record shows why.
In March, TechCrunch reported that an internal AI agent posted advice without permission. Company and user data then reached unauthorized engineers for two hours. Meta rated it a Sev 1.
Reuters added another number. Major technical and security incidents inside Meta rose 40% year over year, which the company tied to an AI coding surge and agent problems. Time spent firefighting rose 70%.
Meta’s response includes real safeguards. The company says it delayed Muse in April to tighten security. Muse runs on a dedicated, secure virtual machine that holds both the agent and the user’s data.
Users can opt out of having their interactions train Meta’s models. An encrypted version arrives later this year.
Those steps help. They don’t answer a harder question: who pays when an agent makes a bad purchase?
What publishers and brands should do now
Treat agents as a new audience. Five moves cost little and pay off fast.
- Put price, availability, and policies in plain text. Agents can’t act on what a pop-up hides.
- Keep key facts out of interstitials. A wall that blocks people may block agents too.
- Write for citation, not only clicks. Clear answers earn mentions when agents summarize a topic.
- Retire stale pages. Outdated documents mislead the assistants that read them.
- Run your own test. Send an agent through your checkout and note where it stalls.
None of this promises traffic. It only keeps your pages usable by the newest visitor.
What to watch next
Three signals will show whether this shift is real or hype.
Retention. Muse drew 2.5 million downloads fast. Daily use after the first month will matter more.
Cost. Investors want proof that agents earn more than they cost to run. The concierge detail makes that proof harder to find.
Rules. Sites will write terms for agent visitors. Regulators may follow. Whoever sets the first rules shapes the market.
Quick answers
Q. How much does Meta Muse cost?Meta Muse has a free basic tier. Meta also offers paid plans priced at $20 and $100 per month for users who need heavier or more advanced usage.
Q. Who can use Meta Muse?At launch, Meta Muse is available to adults in the United States. Availability may expand to more countries and user groups over time.
Q. Can Meta Muse access other apps?Yes. Meta says Muse can work across services such as email, calendars, payments, shopping, health tools, and smart-home apps, depending on the permissions a user grants.
Q. Can Meta Muse replace Google Search?Not completely. Meta Muse adds an agent-based way to use the web by finding information and completing tasks for users. Search engines remain important for discovery, research, and direct website access.
The bottom line
Muse doesn’t wreck the internet. It walks past the toll booths that built it.
The sites that adapt first will serve agents cleanly and still reward people who visit. The rest may discover their best customer never saw the homepage.
