TL;DR
- Meta and Sierra previewed Personal Agent Protocol on 6 October 2026, an open standard for how personal AI agents interact with businesses.
- Meta's post names Genesys, NiCE, Decagon, Rocket, Shopify, Stripe and Walmart as partners; Meta says anyone will be able to implement the protocol.
- Muse applies a 'one honest person' test and an 'if every agent did this' test, and asks explicit approval before sign-in, reservations or purchases.
- Meta is building ways for businesses to steer agents to preferred site flows, connectors such as order tracking or real-time inventory sources, and their own business agents.
- Meta's post gave no specification, launch date, pricing or country availability; Sierra published a draft specification on 9 October.
Meta announced on 6 October 2026 that it is previewing Personal Agent Protocol, an open standard that defines how personal AI agents such as its Muse assistant interact with businesses. Meta (through Muse and Meta Business Agent) is developing it with Sierra and industry partners Genesys, NiCE, Decagon, Rocket, Shopify, Stripe and Walmart. The protocol is meant to handle authentication, empower consumers and give companies visibility into what personal agents do on their websites, APIs or company agents, and Meta says it will be open for anyone to implement. In the same post Meta set out the rules Muse follows when it browses business websites today, and argued that businesses blocking personal agents are turning away the customers behind them.
What Meta announced on 6 October 2026
The announcement appeared on the Meta for Business news page under the title "A New Way for Businesses and Personal Agents to Work Together". It does two things. First, it previews Personal Agent Protocol, which Meta describes as an open standard "that defines how personal agents interact with businesses". Second, it publishes the standards Meta built Muse to follow on the open web, where Muse acts as an extension of the person who sent the request. Meta says it thinks all personal agents should be built this way.
Meta frames the protocol as work done jointly with Sierra and partners "across retail, payments, and customer experience". The named partners are Genesys, NiCE, Decagon, Rocket, Shopify, Stripe and Walmart. Sierra's own announcement also names Instinct, and on 9 October Sierra announced 35 additional design partners. Meta also says it has designed the core parts of the protocol and that a working group of industry partners will build them out together and get customer experiences live quickly. Meta's post gave no launch date, specification or version number. Sierra's own announcement the same day said a v0.1 specification would follow later in October, and on 9 October Sierra published a draft of the protocol, nicknamed Poppy, at personalagentprotocol.org.
How Muse works with businesses today
According to Meta, Muse gives every user their own computer and browser in the cloud, with internet access, where Muse can run code and use the web under the user's direction. Meta says Muse already buys things, makes appointments and resolves support questions, either through a connector or directly on a business's website. The examples Meta lists are practical: booking a family activity, finding and buying gifts, scheduling a dentist appointment, and sorting out a support issue without the user waiting on hold.
Meta's argument to businesses is that a person who asks Muse to book with them or reorder from them has already decided to be their customer. When the agent can finish the task, Meta says, the business gets a completed sale and the person gets their time back.
What came before this announcement
This is the third Muse development in about a month. Relevant Audience reported Muse's launch on 9 September 2026 as a personal agent that books, fills in forms and buys. On 30 September 2026 it covered Meta adding small-business skills and connectors to Muse, so owners could link Instagram professional account analytics, Facebook Pages, Meta ad accounts and third-party tools such as Shopify and Stripe. Those updates were about what Muse can do for a user or for a business owner. The 6 October post is different: it is about the rules of engagement between a consumer's agent and a business it visits, and about a shared standard that other agent makers and platforms are meant to adopt.
Why some businesses block personal agents, in Meta's words
Meta openly acknowledges that some businesses have responded to personal agents by blocking them on their websites, and says it understands why. The concerns it lists are load spikes, abuse and liability, whether the business can keep the customer relationship, and whether it can still guide and influence the user's journey with its brand.
Meta's counter-argument is that Muse acts when a person asks it to, so a blocked agent means a blocked customer. The post puts it bluntly: "Turning away a personal agent means turning away the customer behind it." Meta adds that as more people hand errands to agents, more current and future customers will arrive this way. That is a forecast from Meta, not a measured figure; the post gives no data on how much traffic or revenue currently comes through personal agents.
The two tests Muse applies before acting
Meta says that when browsing, Muse behaves like a personal assistant acting for a single person while respecting the legitimate interests of site owners and the public. Before acting it applies two tests.
- The "one honest person" test. Would a reasonable, honest person doing this by hand do it this way and at this scale? Meta's examples: a person could refresh a page 500 times an hour but would not; could create five accounts to redeem a one-per-household coupon but should not; could write a fake five-star review of their own restaurant but should not.
- The "if every agent did this" test. If every Muse user made the same request, would the system still function? Meta's example is tee times: one person hoarding them is annoying, every agent hoarding them breaks the market.
When a request crosses the line, Meta says Muse does what it can, skips what it cannot and explains the difference to the user. For consequential actions, Muse is designed to continue only with explicit user approval, and Meta names signing in, completing a reservation and making a purchase as examples of actions that need it.
For a site owner, these tests matter because they describe the behaviour Meta is committing its own agent to. They are self-imposed policies stated in a blog post. The announcement does not describe any enforcement mechanism a business could rely on, or how other agent makers would be held to the same standard.
What Personal Agent Protocol is meant to cover
Meta describes a shift from today's browser-based approach towards direct agent-to-agent communication. The table below summarises what the announcement says about each approach.
| Approach | What Meta says it involves |
|---|---|
| Business guidance to agents | A business tells personal agents how to use its website, for example which flow to use for bookings, so Muse follows the preferred path instead of trial and error. |
| Connectors and skills | Direct access to a business's APIs for read and write actions, such as a connector for order tracking or a structured source for real-time inventory. Muse already uses connectors today; Meta says businesses will be able to point agents to them. |
| Agent to agent | The business runs its own agent that answers Muse directly, instead of Muse interpreting a support article; Meta cites lower latency and fewer errors. |
| Open questions to be specified | How personal agents securely convey user identity, how businesses steer agents to efficient website use or optimised APIs and skills, and how personal agents talk to business agents for the user. |
Meta says the browser is "a good way to use rails that already exist" but calls it slower and more brittle than necessary, because it means loading full pages, clicking through flows designed for human eyes and parsing visual layouts. It compares the goal to email, which it says took off because everyone agreed on addressing, transport and delivery, and which no one owns and anyone can build on.
What Sierra's own posts add
Sierra published its own announcement on 6 October 2026, written by Bret Taylor and Clay Bavor, and a follow-up on 9 October 2026. Neither is linked here, but both add detail that Meta's post leaves out. Sierra's 6 October announcement says the principle behind the protocol is that consumers decide what access to give their personal agents, and companies set parameters for what those agents can do.
According to Sierra's 6 October announcement, the protocol starts on the company's website, where a personal agent discovers what the company offers and how to reach it. The agent then opens a session for its user. It can start as a guest, which Sierra says may be enough to check product availability or ask about a returns policy. When a task needs the customer's account, the customer signs in on the company's page or uses credentials already set up with their agent, and the customer decides whether the agent gets read-only or write access. The session is built on OAuth and carries across channels, so a question asked before sign-in and an order change made afterward count as one visit.
Sierra's 9 October post describes the draft, nicknamed Poppy, as five building blocks: discovery, sessions, sign-in, one session across every channel, and getting the job done. For discovery, a company publishes one file at a standard address on its website, /.well-known/poppy.json, telling agents how to interact, how customers sign in and what the agent can do. Sierra says the protocol is built on OAuth, JSON Web Tokens, HTTPS, OpenAPI and MCP, and that companies choose which channels to offer: their regular web pages, APIs built on standards such as OpenAPI and MCP, or their own agent for tasks that need conversation, such as a warranty claim. The 9 October post also lists 35 more design partners, including Visa, Mastercard, PayPal, OpenAI, Target, United Airlines and Cloudflare.
Sierra's 6 October announcement names possible future additions: more detailed permissions, push notifications so a company can tell an agent when a flight is delayed or an order ships, and payments extensions so an agent could complete a purchase without sharing credit card details. These are described as possibilities, not features with dates. Sierra's 9 October post says that over the next month it will host design workshops and publish a reference implementation.
What Meta's post did not say
Several things a business would want to know are absent from the post, and it is worth being clear about them rather than filling the gaps.
- Meta's post links no specification or developer site; the draft specification was published separately by Sierra on 9 October at personalagentprotocol.org.
- Neither Meta nor Sierra has given a date for when businesses or personal agents can use the protocol in production.
- Meta's post does not say what each named partner will contribute or whether Shopify, Stripe or Walmart will support the protocol in their own products.
- There is no country availability. Relevant Audience's 30 September coverage reported Meta describing Muse as available in the US and Canada; this post does not update that.
- There is no pricing, and no word on whether businesses will need a Meta product such as Meta Business Agent to take part.
What site owners and ecommerce teams can check now
None of the protocol features are live, so there is nothing to switch on yet. The announcement does, however, point at decisions most online stores and booking sites already face as AI agents start visiting. These are practical checks, not steps Meta has asked for.
- Know your current bot policy. Find out whether your CDN, firewall or bot-management settings already block automated browsers, and whether that was a deliberate choice for agents acting on behalf of real customers.
- Look at checkout and booking flows as an agent would. Meta's point that browser automation is brittle on flows "designed for human eyes" applies to any agent. Clear form labels, predictable steps and structured product data help both people and software.
- Watch your analytics for agent sessions. If agents complete purchases, those orders may look different in analytics from human sessions. Teams should decide how they want to identify and report them before volume grows.
- Follow your platform vendors. Since Shopify and Stripe are named partners, store owners on those platforms should watch for vendor announcements rather than building custom agent integrations now.
There is also an AI-visibility angle. Meta describes agents reading "structured sources" for inventory and following a business's preferred paths. That overlaps with work brands already do so AI systems can read and cite them accurately, which is the focus of generative engine optimisation. Clean structured data and clear site architecture are useful whichever agent standard wins.
What this means for Thai marketers and businesses
For Thai brands this is an early signal rather than an immediate task. Muse has not been described as available in Thailand, and the protocol is still a draft. But the partner list matters. Merchants whose stores or payments already run on Shopify or Stripe should follow those vendors' announcements, and businesses that sell or answer customers through Facebook, Instagram or Messenger are already inside the Meta ecosystem where Muse and Meta Business Agent sit. If personal agents become a normal way to shop or book, Thai merchants will face the same block-or-welcome choice Meta describes.
Teams that run ecommerce marketing should treat agent traffic as a measurement and conversion question, not only a security one. Brands investing in Facebook ads should also note that Meta is building Muse and Meta Business Agent on both sides of this exchange, which suggests agent-led shopping will become part of the wider Meta commerce picture over time. That last point is analysis; Meta has not announced any link between the protocol and its ad products.
FAQ
What is Personal Agent Protocol?
Personal Agent Protocol is an open standard that Meta and Sierra previewed on 6 October 2026 to define how personal AI agents interact with businesses. Meta says it will cover authentication, empowering consumers and business visibility into what agents do on websites, APIs or company agents, and that anyone will be able to implement it.
Who is building Personal Agent Protocol?
Meta (through Muse and Meta Business Agent) and Sierra are developing it with Genesys, NiCE, Decagon, Rocket, Shopify, Stripe and Walmart. Meta says a working group of these partners will build out the core parts it has designed.
Is Personal Agent Protocol available to use now?
No. Meta described it as a preview. Sierra published a draft specification on 9 October for feedback, but no production release date has been given.
Should my business stop blocking AI agents?
That is a business decision the announcement cannot make for you. Meta argues that blocking a personal agent turns away the customer behind it, but it also lists real concerns such as load spikes, abuse and liability. A sensible first step is to review what your bot settings currently block and why.
Is Muse available in Thailand?
The 6 October announcement does not state any country availability. Meta's earlier small-business update, covered by Relevant Audience on 30 September 2026, described Muse as available in the US and Canada.
Personal agents are still early, but the questions they raise about checkout design, structured data and measurement are already practical ones. If you want help preparing your store or booking flows for both human and AI-assisted customers, talk to Relevant Audience about ecommerce marketing in Thailand.
