Personal Agent Protocol: what it is and who's behind it

Personal Agent Protocol (PAP) is an open standard, announced October 6 by Sierra and Meta, that sets the rules for how a customer's personal AI agent identifies itself to a business, logs in, and gets permission to do things. It is not a product you can install yet: the first draft spec, v0.1, is due later in October.

The announcement came from Sierra cofounder Bret Taylor, who posted on X that Meta and Sierra are building it with Genesys, Instinct, Rocket, Shopify, Stripe and Walmart, and that it is "open for anyone to implement." @stripe confirmed it is a founding partner, and Meta's David Singleton, a former Stripe CTO, backed it from the Meta side.

The problem it is trying to solve

Personal agents like Meta's Muse already try to book, buy and ask questions for people. Today they do it the slow way: they click through websites, fill in forms, and sometimes call the support line. Sierra's announcement post calls that slow and error-prone, and it is.

It also leaves businesses blind. When an agent shows up at your checkout or your booking page, you can't tell who it works for or whether the customer actually agreed to what it is about to do. PAP is an attempt to fix that.

How Personal Agent Protocol works

Based on Sierra's description, a visit goes like this:

• The agent reads the company's website to learn what it offers and how to reach it.
• Without logging in, it can do read-only things as a guest, like checking stock or a return policy.
• When the task needs the customer's account, the customer approves access on the company's own page. This runs on OAuth, the same standard behind "Sign in with Google."
• The customer decides whether the agent can only look or can also act, say by placing an order.
• The guest questions and the logged-in actions count as one visit, even if the agent switches channels partway through.

The business picks which door agents use: the regular website, an API built on standards like MCP or OpenAPI, or the company's own AI agent for anything conversational.

Payments are not in the first version. Sierra lists them as a later extension, along with push notifications and finer-grained permissions for specific actions.

A rival spec showed up the same day

Less than an hour after the PAP announcement, @DecagonAI open-sourced PACT, short for Personal Agent Consent & Trust Protocol, which it built with Instinct. Decagon says PACT lets a business "verify who a personal agent is acting for" and what the customer authorized, and it sits on A2A and OAuth. Decagon also said it is joining the PAP working group. Instinct appears on both announcements.

For now it reads like a handful of companies negotiating in public, with some of them sitting on both sides. According to The Next Web, Stripe and Shopify have also signed on to Visa's Trusted Agent Protocol, so the big payment players are keeping their options open.

The notable absences are Amazon, OpenAI and Anthropic, none of which are on the founding list. Sriram Krishnan compared it to the early days of OAuth and asked how retailers like Amazon and the airlines will cope when they "can't own the final user experience."

Our read

Krishnan's OAuth comparison holds up. OAuth won because one boring, shared login standard beat a dozen custom ones, and agent access needs the same thing. Walmart, Shopify and Stripe matter more here than Meta does. They control the checkouts that will have to accept or reject agents, and a standard they agree on is likely to stick.

The caution is that this is a press release about a spec that doesn't exist yet. Until v0.1 is out and someone ships a reference implementation, nobody, including us, can say how hard it will be to support.

What it means for a small business

You don't need to do anything this month. If you run on Shopify or take payments through Stripe, support will most likely arrive as a setting from them, not as a project for you.

What you can do now is make your business easy for an agent to read. Put your hours, prices, policies and booking rules on your site in plain words, and say what a customer can do without calling. An agent checking your return policy as a guest can only find what you've published.

Agents will have the easiest time with businesses whose processes are already written down and consistent, which is what New Face Design's free process audit looks at. In about 20 minutes we map how customers book, buy and ask questions today, and which of those steps an agent could handle without a person stepping in.