What it is
The Personal Agent Protocol (nicknamed “Poppy”) is an open technical standard for how a consumer’s personal AI agent discovers a business online, signs in with that consumer’s permission, and then acts — browsing, asking questions, or (eventually) completing tasks — on the business’s website, through its APIs, or via the business’s own agent. It is built on OAuth, the authorization framework already used for “log in with Google”-style sign-ins, rather than inventing authentication from scratch.
The protocol defines five building blocks: a business publishes a standard configuration file (at /.well-known/poppy.json) describing how agents may interact with it; an agent opens a “session” and identifies itself; the consumer signs in, granting the agent guest, read-only, or write access; that access persists as a single session across the business’s website, APIs, and any dedicated agent it runs; and the agent then completes the task through whichever channel the business has chosen to expose. The underlying goal, in Sierra’s framing, is to let an agent complete a task “securely in seconds” instead of impersonating a human browsing a website — while giving the business visibility into what the agent is doing and giving the consumer granular control over what it’s allowed to do.
Critically, the first version of the spec (v0.1, due in October 2026) does not include payments. Access control, session identity, and cross-channel sign-in are in scope; letting an agent actually pay for something is listed as a future extension. An agent operating under Poppy today can check a return policy or browse inventory on a consumer’s behalf, but it cannot, on its own, complete a purchase.
Why it matters for AI governance and narratives
Poppy is a live case of private standard-setting racing ahead of public rulemaking in agentic commerce. It was announced jointly by Meta and Sierra (an AI agent company) with Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart as initial partners — a coalition of platforms, payments infrastructure, and retail, but notably without Amazon, OpenAI, or Anthropic at launch. Three days later, Sierra published a draft spec and added 35 more design partners, including OpenAI, Visa, Mastercard, PayPal, Bank of America, and Wells Fargo — a rapid consolidation that illustrates how quickly a vendor-led standard can go from exclusive to broad-based once it demonstrates momentum.
The episode is also a useful instance of naming doing editorial work before the substance catches up: a protocol marketed as solving “agent commerce” launched covering only identity and access, with the harder, more consequential problem — agents moving money — deferred. Coverage and commentary have split between treating the announcement as infrastructure-building (a necessary plumbing layer before agents can transact) and treating it skeptically, noting that whoever controls the specification controls who gets to build agents businesses will trust. At least one analysis has flagged a governance question: with no published license or governing body at launch, and Meta and Sierra as the controlling authors, the protocol concentrates authorship in two commercially interested parties even as it recruits a broad partner coalition. Poppy also competes for the same territory as at least three other proposed standards — UCP (Google, Amazon, Shopify, Microsoft, Salesforce, Stripe), ACP (OpenAI and Stripe, focused on in-agent checkout), and Visa/Cloudflare’s Trusted Agent Protocol (cryptographic agent identity) — with Stripe and Shopify backing multiple competing efforts simultaneously, a sign that no single standard has yet won.
Key facts and dates
- October 6, 2026: Meta and Sierra announce the Personal Agent Protocol, with Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart as initial partners.
- October 9, 2026: Sierra publishes a public draft spec, nicknamed “Poppy,” and names 35 additional design partners — including OpenAI, Visa, Mastercard, PayPal, Adyen, Plaid, Cloudflare, Bank of America, Wells Fargo, Target, and Zapier.
- v0.1 specification: Scoped to discovery, sign-in, sessions, and cross-channel continuity; payments, push notifications, and finer-grained permissions are explicitly listed as future extensions, not part of the initial release.
- Competing standards: UCP (checkout/post-purchase flows), ACP (OpenAI/Stripe, in-agent payment tokens), and Visa/Cloudflare’s Trusted Agent Protocol (agent identity verification) address overlapping ground, with several of the same companies (notably Stripe and Shopify) backing more than one effort.
Where to learn more
- Introducing Personal Agent Protocol — Sierra’s original announcement (primary source)
- Sharing a draft of Personal Agent Protocol — Sierra’s October 9 draft-spec post, with the five building blocks and expanded partner list (primary source)
- Meta and Sierra’s Personal Agent Protocol Gives Commerce a Common Front Door — For Those Willing to Open It — independent analysis situating the protocol against competing standards and governance concerns
- Meta’s Personal Agent Protocol Signs Walmart Before It Has a Spec — comparative breakdown of PAP versus UCP, ACP, and Trusted Agent Protocol