Specoria
Menu
TR Free test
← All insights
Agentic commerce8 min read

Personal Agent Protocol: what stores should know

Meta, Sierra, Walmart, Shopify and Stripe back an open standard for how personal AI agents sign in to a store and what they may do there. How to prepare.

On 6 October 2026, Sierra and Meta announced Personal Agent Protocol (PAP), an open standard for how personal AI agents deal with businesses. The partners named in the announcement are Genesys, Instinct, Rocket, Shopify, Stripe and Walmart. The specification itself is not out yet: Sierra says version 0.1 will be published later in October, followed by design workshops and a reference implementation.

The timing is not an accident. Meta's shopping agent Muse launched in September, and according to Bloomberg, Amazon began blocking it from its retail site on 20 September, saying the agent did not identify itself while browsing. According to CNBC, Bret Taylor, Sierra's co-founder, said the situation is "kind of chaos" until such a standard exists. PAP is an attempt to replace that chaos with a front door.

For a store owner, the useful questions are simple: what problem does this solve, what does it not solve yet, and is there anything to do before the specification is published?

The problem PAP is trying to solve

Today, most personal agents use a store the way a person does. They load pages, read them and click through forms. Sierra's announcement says it plainly: when that fails, an agent may fall back to calling the support line or using the web chat, which is slow and often does not finish the job.

The deeper problem is that the store cannot tell what it is dealing with. A request from an agent in the shopper's browser looks like a person. A request from a server-side agent may look like a scraper. In neither case does the store know which customer the agent represents or what that customer actually allowed it to do. That uncertainty is why stores either block agents broadly or let everything through.

We wrote earlier about how agents identify themselves to a store: user agent strings, published IP ranges and cryptographic signatures such as Web Bot Auth. Those methods answer one question: is this really the agent it claims to be? PAP goes after two different questions: which customer is this agent acting for, and what has that customer authorised it to do?

How it works, based on what has been published

Sierra's announcement, written by Bret Taylor and Clay Bavor, describes the design at a high level. The principle is that consumers decide what access to give their agents, and companies decide what those agents can do.

  • Discovery starts on your website. A personal agent finds out from the company's site what it offers and how to reach it.
  • Guest sessions for public questions. Without signing in, an agent can do things like check product availability or ask about the returns policy.
  • Sign-in for account tasks. For anything tied to an account, the customer signs in on the company's own page or uses credentials already set up with their agent.
  • Read or write, chosen by the customer. The customer decides whether the agent gets read-only access (checking an order, say) or write access (changing an order or placing a new one).
  • Built on OAuth. The session uses OAuth, an established authorisation standard. In press coverage, Taylor compared the experience to signing in to a website with a Google or Facebook account.
  • One session across channels. A question asked before sign-in and an order change made after it belong to the same visit.

The company also chooses the route. An agent can work through the regular website, through APIs built on standards such as MCP and OpenAPI, or by talking to the company's own customer service agent for conversational tasks like a warranty claim.

What it does not cover yet

It is worth being precise about the gaps, because early coverage tends to blur them.

  • No specification yet. As of 9 October, the spec has not been published, and no licence or independent governing body has been announced. Everything above comes from the announcement, not from a document you can implement.
  • Payments are a future extension. Sierra lists "letting a personal agent complete a purchase without sharing credit card information" as planned work, not part of the current design. Checkout and payment still run through the existing routes.
  • Finer permissions and notifications come later. Limits on specific actions and push notifications (an order shipped, a flight delayed) are also listed as future extensions.
  • It is not the only effort. According to The Next Web, Stripe and Shopify are also part of Visa's Trusted Agent Protocol, which covers agent identity and payment intent from the card network side. The two are not described as replacements for each other, but a store may end up dealing with more than one.
  • Adoption is voluntary. Amazon is not among the partners, and neither are the card networks. Whether this becomes a common front door depends on who joins.

Why it matters for a store anyway

Look at what Shopify chose to emphasise. Mani Fazeli, its VP of Product, framed PAP around agents answering product questions and handling purchases, returns and exchanges. Stripe's Kevin Miller made a similar point: "A customer relationship doesn't start or end at checkout."

That is the part most agent readiness work skips. Store owners tend to think about agents at the moment of purchase: can the agent find the product, read the price, get through checkout. PAP is mostly about what happens around the purchase: is this in stock, what is your return window, where is my order, I want to exchange this for a larger size. Those are the tasks that currently end up in your inbox, call centre or chat widget, and they are the tasks a personal agent will try to do on the customer's behalf.

If those answers only exist inside a PDF, a chat transcript or a support agent's head, no protocol will help an agent find them.

What you can do before the spec arrives

There is nothing to install today. There is, however, groundwork that pays off whether PAP takes off or not.

  • Make the guest questions answerable on the page. Stock status, delivery times and the return window should be visible on product and policy pages as text, not hidden behind a chat bubble or an image. These are exactly the guest session examples Sierra gives.
  • Give order status a path that does not need a phone call. If customers can only check an order by emailing you, an agent will hit the same wall. A sign-in-based order page, or at least an order lookup form, is the minimum.
  • Write down your return and exchange flow. Who can start a return, within how many days, with which steps. An agent with write access will only be as useful as the steps you have defined.
  • Separate reading from changing in your own systems. PAP's split between read-only and write access mirrors a sensible internal rule: looking up an order should carry less risk than changing it. If your platform or helpdesk lets you scope access, start there.
  • Review your bot rules with "who" in mind. Blocking every automated visitor will also block the agents of your own signed-in customers. Keep protection on, but ask your CDN and platform how they plan to treat identified agents.
  • Ask your platform. On Shopify, PAP support, if it comes, will most likely arrive through the platform, as WebMCP did. On ikas, Ticimax, IdeaSoft, T-Soft or WooCommerce, as of this writing we have found no public statements about PAP. It is a fair question for their roadmap, especially for order status and returns.

The bottom line

PAP is an announcement, not yet a standard. But it shows where large retailers, platforms and payment companies think the friction is: not only at checkout, but in the dozens of small account and service tasks that agents currently handle by clicking around or giving up. Stores that already answer those questions clearly on their own pages will be ready for whichever protocol wins, and will work better for today's screen-reading agents in the meantime.

Specoria checks whether your public pages answer the questions agents ask, from stock and delivery to returns, and sends real agents through shopping tasks to show where they stop. Start with a free readiness test.

Sources