Sierra, the customer service agent company founded by Bret Taylor and Clay Bavor, published a draft of the Personal Agent Protocol, known as Poppy, on Thursday, October 9, 2026. It is a proposed open standard governing how personal artificial intelligence agents, the assistants that act on someone’s behalf, deal with companies when they need to buy something, handle a return or resolve an issue.
The protocol had been announced two days earlier, on Tuesday, during the Sierra Summit conference, together with Meta and a first group of companies made up of Genesys, Instinct, NiCE, Rocket, Shopify, Stripe and Walmart. According to Taylor and Bavor, the response was so overwhelming that they decided to publish the draft right away and announce 35 additional design partners that will take part in its development, contributing the perspective of their industries.
The list of new partners is broad and varied: Adyen, Atomic, Bank of America, BBVA, Chime, Cigna, Cloudflare, Comcast, DIRECTV, ElevenLabs, FOX, Gap Inc., GEICO, Hertz, Insurify, Klaviyo, Liberty Mutual, Mastercard, Nordstrom, Notion, Okta, OneSignal, OpenAI, PayPal, Plaid, SiriusXM, Synchrony, Target, United Airlines, Venmo, Visa, Wells Fargo, Zapier, Zendesk and 1Password. It includes banks, insurers, airlines, retailers, media companies and payments and identity firms.
The problem Poppy is trying to solve is concrete. Today, when someone asks their personal agent to make a purchase or resolve a problem, the agent often has to sign in as if it were the customer and navigate pages designed for people. The protocol’s documentation describes it as an all-or-nothing situation: either the agent passes as the user, with their password, or the company blocks it and it can do nothing. The user cannot limit what the agent does, and the company cannot tell it apart from a person.
Poppy rests on a simple principle: the customer decides what access to give their agent, and the company sets the parameters for what that agent can do. For example, an airline could let any agent search flights, allow signed-in agents to see the user’s trips and require the user’s approval before buying a ticket. The user, in turn, could let the agent view their trips but not change them.
The draft defines five building blocks. The first is discovery: each company publishes a file at a standard address on its website telling agents how to interact with it, how customers can sign in and what the agent can do. The second is sessions: the agent can start as a guest, which is enough to check whether something is in stock or what the returns policy is, but it always identifies itself, so the company knows when it is dealing with an agent.
The third block is sign-in. When a task needs the customer’s account, the person signs in on the company’s page through OAuth, the common authorization standard, and the agent receives a session token that grants only the access the customer approved. The fourth is a single session across every channel: the same token works for the company’s APIs, its website and conversations with its own agent. The fifth is getting the job done through whichever channel the company prefers.
Sierra stresses that the protocol is built on existing standards: OAuth, JSON Web Tokens, HTTPS, OpenAPI and MCP, the model context protocol that has spread among agents. The idea is that teams can implement it with libraries and tools they already know. The documentation adds a privacy principle: until the person signs in, the company knows them only by an identifier with no personal information, stable over time but different at every company.
The company also presents it as complementary to other initiatives. Protocols such as the Universal Commerce Protocol (UCP) focus on shopping, from finding products to checking out and managing orders. Poppy, by contrast, aims to cover how a personal agent identifies itself, obtains the customer’s permission and works with a company on any kind of task, “from buying a sweater to applying for a mortgage,” according to Sierra.
The most eye-catching example came at the conference. Alex McGillis, vice president of product at Rocket, showed live how Muse, Meta’s personal agent, handled a mortgage pre-approval. Muse went to Rocket’s website, found the company’s agent through Poppy and asked McGillis for permission to connect. He signed in to share the necessary credit and income details, and the two agents worked through the price, down payment and interest rate until, within moments, a pre-approval letter was ready.
The list of participants is perhaps the most significant part of the announcement. The presence of OpenAI, whose ChatGPT is one of the most widely used personal agents, and of Meta suggests that the big assistant developers see value in a common language. That of Visa, Mastercard, PayPal and several banks indicates that the financial sector wants a say in the rules from the start, rather than reacting to agent traffic that is already reaching its systems.
There are important caveats. This is a draft, version 0.1, and real adoption will depend on companies implementing it and on agent developers respecting it. Sierra also has an obvious commercial interest, since it sells customer service agents to major brands. Over the next month, the company will host design workshops and publish a reference implementation, and it has invited as many companies as possible to read the draft and help shape it.
If Poppy takes hold, it could address one of the bottlenecks of the agent economy: mutual distrust between the assistants acting for users and the companies receiving them. According to the documentation, as agents take on more tasks, companies spend more effort detecting them and agents spend more effort evading detection. A shared standard would let both sides work together under clear rules, with the user as the owner of the permissions.
