Sierra and Meta publish the Personal Agent Protocol.
Personal Agent Protocol
An open standard by Sierra and MetaAnnounced on 6 October 2026 · in preview
The customer decides what access to give their agent
The business sets what agents may do with that access
No sign-inAs a guestThe agent finds the business on its website and looks at what it offers, without knowing who the customer is.Open to any agent, within the limits the business sets
Customer signs inReading the accountThe customer signs in on the business's own page or with an account they already have there, and gives the agent read-only access.Granted by the customer, through OAuth
Customer allows moreActing on the accountWith write access, the agent can act for the customer, only through the actions the business has chosen to allow.Granted by the customer, bounded by the business
Three ways in
The websitePages and forms, as today
An APIUsing standards such as MCP or OpenAPI
The business's own agentFor tasks handled in conversation
On 6 October 2026 Sierra announced the Personal Agent Protocol, an open standard it is developing with Meta for the way a person's own AI agent deals with a business: how the agent finds the business, how the customer signs in and gives it access, and what the agent is then allowed to do. The protocol is in preview, and the first version of the specification is due later in October. This page sets out what was announced, how the protocol is meant to work, who is behind it and what comes next, from Sierra’s announcement.
By The AI Adoption · Published on 7 October 2026 · Source: Sierra, 6 October 2026
What was announced
A personal agent is an AI assistant that works for one person and does things on their behalf, such as checking an order, comparing what a shop offers or making a change to an account. Sierra and Meta are developing the Personal Agent Protocol as an open standard for how these agents deal with businesses. Sierra describes three things the protocol has to handle together: authentication, so the business knows which customer the agent represents; the customer’s control over what the agent may do; and the business’s visibility over what agents do on its systems.
Sierra’s announcement says the protocol is open for anyone to implement. It is being developed with a group of founding partners that includes Shopify, Stripe and Walmart, listed further down this page.
The problem it addresses
Today a personal agent that wants to do something on a business’s website mostly works the way a person does: it loads the pages and fills in the forms. Sierra describes this as slow and prone to errors. A direct connection between the agent and the business could do the same task securely in seconds, but only if the customer, the business and the company that built the agent can trust each other, and the announcement sets out what each of them needs before that happens.
What each side needs, as the announcement lists it
The customer
Speed, reliability, and an agent they can trust to act in their interest.
The business
Visibility over what agents do on its systems, and control over what they are allowed to do.
The agent builder
Efficient and consistent access to the systems of the companies its agents deal with.
How it works
The agent first finds the business on its website. It then opens a session built on OAuth, an existing standard for signing in, and that session carries over if the agent moves from one channel to another. As a guest, without signing in, the agent can look at what the business offers. When the customer signs in, on the business’s own page or with credentials they already have there, the agent can reach the customer’s account, and the customer chooses whether it may only read the account or also act on it.
The business keeps the other half of the decision. Sierra’s announcement puts it this way: customers decide what access to give their personal agents, and companies set the parameters for what those agents can do. The business also chooses how agents reach it: through its website, through an API built on standards such as MCP and OpenAPI, or through its own agent for tasks that are better handled in a conversation.
Deciding which actions an AI may take, and which ones wait for a person, is the same question a business answers for the AI its own team uses. We explain how we set those access levels and approvals in adopting AI securely, and how a task can be handed to AI while a person keeps the approval in handing a repetitive task to AI.
Who is behind it
Sierra leads the announcement and develops the protocol with Meta. The announcement names six other companies as founding partners.
Founding partners named in the announcement
SierraDevelops the protocol
MetaDevelops the protocol
GenesysFounding partner
InstinctFounding partner
RocketFounding partner
ShopifyFounding partner
StripeFounding partner
WalmartFounding partner
What comes next
Sierra says it will publish the first version of the specification later in October, run design workshops with the companies that want to take part, and release a reference implementation that developers can start from. It also lists the extensions it plans to work on after that: more detailed permissions, push notifications, and secure payments.
From preview to the planned extensions
6 October 2026Announced, in previewOpen for anyone to implement
Later in October 2026First version of the specificationPublished by Sierra
ThenDesign workshops and a reference implementationWith the companies that want to take part
After thatPlanned extensionsMore detailed permissions, push notifications, secure payments
What the announcement does not cover
The specification is not published yet, so the announcement describes the protocol’s principles and not its technical detail. It does not say which agents will support it first, when the founding partners will use it in their own products, or what it will cost a business to take part. It names no country, so nothing in it says when or how the protocol will apply to businesses in the United Kingdom, Australia or elsewhere.
The announcement is also written for companies that have their own developers or work with a platform among the founding partners. It does not say how a business without a technical team would take part, beyond the fact that the standard is open for anyone to implement.
How we start
We set up the AI a business uses, with the access and the approvals that fit each role. If you are working out how to put AI to work in your business and want to talk it through, the first step with us is a free AI pre-audit: 45 minutes on a call about how you and your team use AI today and which work to hand over first. You leave with a clear next step, with or without us.
The AI AdoptionFree AI pre-audit