What happened
Personal AI agents are being blocked from websites. Amazon is intentionally blocking Meta's Muse agent, while other sites are accidentally blocking them with existing anti-bot tools. This prevents agents from completing tasks like booking flights or making purchases on a user's behalf.
In response, an industry group including Meta, Walmart, and Stripe has started work on a new open standard. According to TechCrunch, the protocol aims to help websites distinguish between malicious bots and legitimate AI agents acting for a user.
How the room's reading it
Builders and early adopters are frustrated. User complaints on social media show agents from Meta and Google failing on sites like Yelp, eBay, and various airlines. It's often unclear whether a block is intentional policy or a clumsy anti-bot filter, creating a poor user experience. Some companies are openly cautious — Delta cites security concerns, while United points to its terms of use against automated scraping. Others, like Yelp, are treating agent access as a commercial data licensing issue. The common thread is that the web's current immune system can't tell the difference between a helpful agent and a malicious bot, making the agent-first web feel unreliable.
Sailfish's take
We think an open standard is the right goal, but it will move slowly. The real issue isn't technical, it's commercial. Websites don't want their data scraped or their checkout flows automated for free — especially by large model providers. An agent acting for a user looks a lot like a scraper to a business protecting its margins.
We've shipped enough automation to know that relying on brittle web scraping is a losing game. That's why we'd advise builders to avoid agents that promise to navigate the open web. Focus instead on building agents that use official APIs or operate on a user's local context. The web-as-an-API dream is still a long way off.