What Muse is

Your customers can now ask AI agents to act online on their behalf. These agents can shop, book or cancel services, and even call a phone hotline, with little to no direct human involvement.

On 8 September 2026, Meta launched Muse, a personal AI agent that acts for consumers online. According to Meta, it can send emails, book travel, and take on big audacious goals which it plans with the user and then pursues in the background.

Muse is not the only or first AI agent product: OpenClaw is an open-source agent framework that users host and configure themselves. Instinct and Perplexity’s Comet already act for consumers, but through invite-only access or a separate browser. OpenAI dots and xAI Grok Bot target work tasks.

What sets Muse apart is its distribution: it is free, open to any adult in the US and Canada, and available inside WhatsApp, an app with over 100 million monthly active users in the US alone (2024). In the US, Muse reached #1 in the App Store within 10 days of launch.

Why it matters

Muse takes AI agents mainstream. Anyone with WhatsApp in the US or Canada can now ask an agent to shop, book, or negotiate for them, and Muse’s #1 spot on the App Store shows that many are eager to sign up. Meta plans to bring Muse to its AI glasses and a new pocket device (Muse Charm). Despite serious privacy concerns (profiling people without consent, access to email, accounts, and payments), adoption has not slowed. For online businesses, a growing share of browsing, comparing, and buying will be done by agents instead of people.

Agents are useful because they remove friction. But not all friction is a flaw. Some comes from bad design: clunky forms and confusing checkouts. Some protects customers: spending limits, fraud checks, cooling-off periods. And some of it serves the business: ads and upsells between search and checkout. Agents remove all three. Amazon has blocked both Perplexity’s Comet and Muse, citing transparency and security, while its advertising business depends on shoppers who browse. Shopify took the opposite view and shares its merchants’ products with Meta by default.

Considerations

  • Privacy vs convenience. AI agents require sensitive access permissions to operate. Implement least privilege permissions for agents to complete their tasks.
  • Agents fail in new ways. Neither Meta nor OpenAI publish how reliably their agents complete tasks; Amazon’s CEO says they “get a lot of the pricing wrong”. Shopify puts the burden of handling returns and complaints to merchants.
  • Agents look like your customers. Meta says Muse’s browsing “will appear as your activity”, so you cannot apply different rules to it. A standard for agents to identify themselves is still in draft.
  • Different rules in Europe. Muse is not available yet, and payment rules require the customer to confirm most online payments. Agents can still browse European stores, and Shopify shares merchants’ products with AI channels by default.

Recommendation

Consumer agents are real, and your online business will serve both humans and agents acting for them. Three steps apply to any online business:

  1. Remove the friction people dislike. If owning the customer relationship matters to you, make coming to you directly the better option. Fix annoying customer experiences, and keep the friction that protects them.
  2. Set an agent policy. Decide which agents you allow, what they may do, and where a person must confirm. Then enforce it: ask your payment and bot-management providers whether they can identify agents.
  3. Design an agent experience. Make your products easy for agents to find and buy through structured product data and checkout channels, exposing only what a transaction needs. Track agent orders and returns separately.

Sources