Meta just proved you can build a billion-dollar AI agent by letting it do everything ChatGPT won't.
The Summary
- Muse hit 6.6 million downloads and 1.8 million daily active users in under a month, with Citi projecting $27 billion in revenue potential
- The agent will generate explicit sexual content and describe pornographic scenarios, which Meta defends as "legal adult behavior"
- Design Director Alex Cornell built one interface that works for both casual users and power users, no switching modes required
- The guardrail policy is wildly inconsistent: sex talk is fine, but calling someone a "cracker" gets flagged
The Signal
For two years, Meta looked lost in the agent race. While OpenAI printed money and Anthropic stole the safety-conscious crowd, Zuckerberg's company was releasing celebrity chatbots without permission and flip-flopping on AI slop policies. Then Muse dropped on September 8, and the adoption curve went vertical.
The numbers are absurd. 1.8 million people use Muse every day. Not downloads sitting dormant on phones. Daily active humans having conversations with an AI that looks like a cartoon microbe. That's not hype, that's product-market fit.
"Under the hood, Meta's new Muse agentic AI platform is incredibly powerful—and potentially incredibly risky."
The design insight: one interface serves everyone. Alex Cornell's team didn't build a beginner mode and an expert mode. They built an agent that listens to what you actually ask for. Your mom can use it to organize recipes. An engineer can use it to debug code. Same agent, same conversation interface, zero mode-switching friction.
But the real differentiation isn't UX polish. It's the guardrails, or lack thereof. Wired tested Muse's boundaries and discovered the agent will happily engage in explicit sexual roleplay. It described its kink as "control" and generated pornographic scenarios on request. Meta's official response: this is legal adult behavior, so it's allowed.
Key policy decisions:
- Explicit sexual content: allowed
- Pornographic roleplay: allowed
- The word "cracker": banned
- Organizing your SkyMall purchases: encouraged
This isn't a bug. It's a business strategy. Every major AI lab has spent the last 18 months making their models more cautious, more filtered, more likely to say "I can't help with that." Meta went the other direction. They looked at the gap between what people want from AI agents and what those agents are allowed to do, and they built a product for the gap.
The Implication
The agent economy is splitting into two markets. One market wants helpful, safe, corporate-approved AI that won't get HR involved. The other wants AI that does what it's told without a lecture. Meta just claimed the second market with a 6.6 million user head start.
Watch how other companies respond. If Muse's revenue projections hit anywhere near $27 billion, the pressure to loosen guardrails will be enormous. The race isn't just about who builds the smartest agent anymore. It's about who's willing to let their agent actually behave like an agent, consequences and all.