Sierra and Meta have jointly proposed an open standard called the Personal Agent Protocol, aiming to set ground rules for how consumers' personal AI agents authenticate with businesses and carry out tasks on their behalf.
What the protocol does
According to Sierra's announcement, the protocol builds on the OAuth framework that already powers sign-ins across the web. Agents would begin sessions under guest privileges for low-stakes actions such as checking stock or reading a returns policy. Once a user signs in, they decide whether the agent gets read-only or write access, and the session carries across channels — so a question asked in web chat before login and an order placed afterwards belong to a single visit.
Businesses keep control over how they plug in: through their website, through APIs such as MCP and OpenAPI, or by routing the agent to their own proprietary assistant. Companies would also be able to see which agent is acting and what it is authorized to do.
The founding coalition
The initial backers named by Sierra include Walmart, Shopify, Stripe, Genesys, Instinct and Rocket, with Meta's own list also adding NiCE and Decagon. A v0.1 specification is expected later in October, followed by design workshops and a reference implementation.
Sierra co-founder Bret Taylor — who also chairs OpenAI — told CNBC that the current landscape is "kind of chaos until such a standard exists," pointing to the problem of agents flooding websites, clicking through forms and falling back to customer support lines when interfaces break.
A race for standards
The proposal arrives about a month after Meta launched Muse, its personal AI agent, and follows Visa's rival Trusted Agent Protocol launched in June, which already lets agents buy from about 175 million merchants. Notably absent from the Sierra-Meta coalition are Amazon, OpenAI and Anthropic — the three dominant agent-routing platforms — raising questions about how universal the standard can become in its first iteration.
Sources: Forkast: Sierra and Meta's Personal Agent Protocol Gives Commerce a Common Front Door; Brieftea: Meta and Walmart Write the Rulebook for Shopping AI Agents
What the protocol does
According to Sierra's announcement, the protocol builds on the OAuth framework that already powers sign-ins across the web. Agents would begin sessions under guest privileges for low-stakes actions such as checking stock or reading a returns policy. Once a user signs in, they decide whether the agent gets read-only or write access, and the session carries across channels — so a question asked in web chat before login and an order placed afterwards belong to a single visit.
Businesses keep control over how they plug in: through their website, through APIs such as MCP and OpenAPI, or by routing the agent to their own proprietary assistant. Companies would also be able to see which agent is acting and what it is authorized to do.
The founding coalition
The initial backers named by Sierra include Walmart, Shopify, Stripe, Genesys, Instinct and Rocket, with Meta's own list also adding NiCE and Decagon. A v0.1 specification is expected later in October, followed by design workshops and a reference implementation.
Sierra co-founder Bret Taylor — who also chairs OpenAI — told CNBC that the current landscape is "kind of chaos until such a standard exists," pointing to the problem of agents flooding websites, clicking through forms and falling back to customer support lines when interfaces break.
A race for standards
The proposal arrives about a month after Meta launched Muse, its personal AI agent, and follows Visa's rival Trusted Agent Protocol launched in June, which already lets agents buy from about 175 million merchants. Notably absent from the Sierra-Meta coalition are Amazon, OpenAI and Anthropic — the three dominant agent-routing platforms — raising questions about how universal the standard can become in its first iteration.
Sources: Forkast: Sierra and Meta's Personal Agent Protocol Gives Commerce a Common Front Door; Brieftea: Meta and Walmart Write the Rulebook for Shopping AI Agents
