Enterprise AI company Sierra and Meta jointly announced the Personal Agent Protocol on October 6 — an open standard for how personal AI agents interact with businesses, open for anyone to implement. Walmart, Shopify, Stripe, customer-service software maker Genesys, and two other companies, Rocket and Instinct, helped draft it, and the founders say they welcome more partners.
The protocol's v0.1 specification is expected later this month, to be followed by design workshops and a reference implementation. The announcement is signed by Sierra co-founders Bret Taylor and Clay Bavor.
What problem it's trying to solve
Right now, when an AI agent visits an e-commerce site or a customer-service system, the business has no real way to tell whether it's talking to a human or a machine — let alone who the agent is acting for or how far it's authorized to go. The protocol aims to set common practices for three things: identity verification, consumer authorization, and giving businesses visibility into what an agent does across their website, APIs and customer-service agents.
On the technical side, the protocol is built on top of OAuth. An agent can first enter as a guest to check inventory or review a return policy; once the user logs in, the business decides whether to grant it read-only or read-write access. Sessions can carry across channels, so questions asked before login and order changes made after login count as the same visit.
There are three ways in: browsing a business's website directly; calling its APIs through standards like MCP or OpenAPI; or talking to the business's own customer-service agent. The announcement says future extensions will cover finer-grained, action-level permissions, push notifications for order and flight status, and payment extensions.
"Personal AI is creating a new front door to the enterprise. Brands need a trusted way to know who an agent represents."
That's from Genesys CEO Tony Bates in the announcement — in short, personal AI is becoming a new front door into the enterprise, and brands need a trustworthy way to know whom an agent represents.
How it lines up against existing protocols
Over the past year or so, several agent-related protocols have emerged, each covering a different piece of the puzzle:
| Protocol | Led by | What it governs |
|---|---|---|
| MCP | Anthropic | How agents call external tools and data |
| A2A | How agents coordinate with other agents | |
| Agentic Commerce Protocol | OpenAI, Stripe | Agents completing checkout inside a chat |
| Personal Agent Protocol | Sierra, Meta | Under whose identity, and with what permissions, an agent enters a business |
The new protocol lists MCP as one usable access path and positions itself as more of an identity-and-authorization layer that can sit alongside the others. Stripe's position is distinctive: it's involved in both OpenAI's checkout protocol and this new one, suggesting the payments company wants to hold onto the payment step no matter which agent doorway gets used.
Who hasn't signed on
The founding list covers retail, e-commerce, payments and customer service, but Meta is the only model maker. OpenAI, Google and Anthropic each have their own personal-assistant products and have each led protocols of their own; none has publicly said whether or when it might adopt this one.
For Meta, the upside is fairly direct: with a large user base for its Meta AI assistant, having a set of credentials that merchants are willing to recognize helps when its own agent places an order at, say, Walmart. As for governance — whether the protocol will eventually be handed to a neutral foundation — the announcement doesn't say; that will have to wait for the v0.1 spec.
Sources: Sierra official blog, CocoLoop, CNBC, The Next Web; the founding member list, technical design and v0.1 timeline follow Sierra's official announcement.