Meta’s Muse Agent Lifts Stock 15% — But the Hard Problems Stay Unsolved
TREE NEWS reports: Meta’s new personal AI agent, Muse, has become the company’s unlikely “muse.” Launched on September 8, the assistant climbed to the top of the U.S. Apple App Store free chart within ten days and held the No. 1 spot for three consecutive days. By September 21, the momentum had pushed Meta’s stock up 11.4% in a single session, adding roughly $192.3 billion in market capitalization. The enthusiasm spilled into semiconductors, with AMD rising 9.9% and crossing a $1 trillion valuation for the first time.
Why Users Downloaded It: “Help Me Get My Money Back”
The viral hook was not model benchmarks but refunds, forgotten subscriptions, and insurance bills. Meta’s chief AI officer Alexandr Wang promoted a “MuseMoneyChallenge” encouraging users to share savings results. One entrepreneur claimed Muse found a car insurance plan that saved $3,500 a year — though many shared examples reportedly came from Meta employees. For people who don’t care about AI, a dollar figure is easier to grasp and share than a benchmark score.
Muse also tackled a subtler problem: users not knowing what to delegate. Early testing reportedly showed that too many features overwhelmed people, so the design team added proactive suggestions based on goals and conversation. Wharton professor Ethan Mollick noted that other labs can do similar things, but Muse’s focused design makes it easier for ordinary users — helped by free compute and a familiar WhatsApp entry point.
Where the Technical Novelty Actually Sits
Muse runs a cloud computer on Meta’s servers, so tasks continue after a user closes the phone app. Each user gets a Muse Secure VM, with execution tools in restricted containers, credentials and persistent state managed separately, and a Sentinel agent reviewing outbound access. The engineering value lies in organizing execution capability alongside authorization boundaries — a partitioned workshop rather than a fully trusted assistant.
Yet similar forms already exist. AutoGLM 2.0 offers cloud phones and cloud computers; Doubao’s AI-native phone operates apps directly; Qwen leverages Alibaba’s ecosystem for ordering and payments. Muse is, at heart, the same coffee-ordering assistant that Chinese users already know — just repackaged for a market more willing to re-rate the story.
The Business Model Remains Unproven
Stratechery’s Ben Thompson called Muse the easiest-to-use personal agent he has tried, noting switching costs rise once users hand over account information. Truist analysts project Muse could add $28.5 billion in incremental revenue by 2030. Oppenheimer’s Jason Helfstein stays neutral, estimating roughly 115 million paying users would be needed to materially affect Meta’s earnings.
On September 21, Amazon blocked Muse from browsing and purchasing on its behalf, citing unauthorized agent access. Doubao’s phone faced similar login and risk-control barriers with financial apps. User authorization cannot substitute for platform permission — and platforms have strong incentives to guard the gate, especially when agents bypass ads and recommendations.
Meta’s stock closed down 0.63% on September 22 before ticking up 0.6% after hours. The market is entering a cooler phase. The next test is retention — and whether Muse has anything genuinely new beyond a familiar performance.
