Meta and Sierra are collaborating with Shopify, Stripe, and Genesys on the Personal Agent Protocol. This open standard defines how personal AI agents communicate with companies. The consumer chooses what access an agent receives, while the company determines what that agent is allowed to do. Version 0.1 of the specification will be released later this month.
Personal AI agents schedule appointments, book flights, and compare car insurance policies. But this process is often still cumbersome. Most agents use websites and apps just as a human would: loading pages, filling out forms, and clicking through. If that doesn’t work, they call customer service or open a web chat. That takes time, and sometimes the task simply fails.
The Personal Agent Protocol (PAP) aims to provide a direct connection to address this. Meta and Sierra are developing the standard in collaboration with Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart. Anyone can implement the protocol.
According to the initiators, such a connection must work for all parties. Consumers want speed and reliability. Brands want to know when an agent is acting on a customer’s behalf and to decide what that agent is permitted to do. And agent builders want a single, consistent way to connect with businesses.
OAuth as the foundation
How does it work in practice? An agent starts on a company’s website, discovers what the company offers, and finds how to contact it. The agent then starts a session on the user’s behalf. This can initially be done as a guest, which is sufficient to check inventory or view the return policy. If the agent needs access to a customer account, the customer logs in on the company’s page or uses login credentials already known to the agent. The customer chooses between read-only or write access.
The session runs on OAuth and continues across channels. A question asked before logging in and an order change made afterward are therefore part of the same visit. The company then chooses the route: the standard website, APIs based on standards such as MCP and OpenAPI, or a proprietary agent for tasks requiring a conversation.
Genesys CEO Tony Bates sees personal AI as “a new front door to the enterprise.” According to him, brands need a reliable way to know who an agent represents and what that agent is authorized to do. “Personal Agent Protocol is an important step toward creating the shared foundation this new era of customer experience will require.”
The partners plan to publish version 0.1 of the specification in October. They will follow this with design workshops with interested parties and a reference implementation for developers. The roadmap includes more detailed permissions per action, push notifications (for example, for a delayed flight or a shipped order), and payment extensions that allow an agent to complete a purchase without sharing credit card information.
Tip: Time for a new operating model? Genesys is revamping its agentic orchestration layer