Sunday, October 18, 2026

PARTH GODA

AI · MBA Life · Music


Career#partnerships#agentic-ai#meta#sierra#standards

Meta and Sierra Want to Write the Rulebook for Agentic Commerce

Three weeks after Amazon blocked Meta's shopping agent for scraping its site without permission, Meta and the enterprise AI startup Sierra are proposing an open standard for how AI agents authenticate with businesses. Here's what it actually does, and who's conspicuously not in the room.

Parth Goda··6 min read·AI-assisted draft

What Was Actually Agreed

On October 6, Sierra, the enterprise AI agent company co-founded by former Salesforce co-CEO Bret Taylor, and Meta announced they are co-developing the Personal Agent Protocol (PAP): an open, OAuth-based standard for how a personal AI agent authenticates and acts on a customer's behalf with a business. A customer grants an agent guest access, a signed-in session, or standing credentials, and chooses read-only or read/write permission. That permission carries across channels, so a pre-login question and a post-login order count as one session. The business gets a consistent way to see what the agent is doing, instead of guessing whether a visitor is human.

CO-DEVELOPING THE PROTOCOLSierraMetaShopifyWalmartGenesysStripeInstinctRocketNOT PARTICIPATINGAmazonOpenAIAnthropicGoogle

A version 0.1 spec, with a reference implementation, is due later in October. None of the eight named partners described any exclusivity: Taylor told SiliconANGLE it is "open for anyone to implement." What isn't yet disclosed is who governs the standard once it exists — Sierra and Meta wrote the first draft, and a critical analysis from Rajesh Beri frames the open question bluntly: "whoever writes the lock decides who walks through it."

Why Now

Sep 21Amazon blocksMeta's Muse agentOct 6Sierra + Meta announcethe Personal Agent ProtocolLater Octv0.1 spec and referenceimplementation dueJan 2027proposed interoptest (unconfirmed)

The timing traces to that first dot. On September 21, Amazon blocked Meta's Muse agent from its retail site, alleging Muse accessed the store without disclosing it was automated and captured customer login credentials in the process. Three weeks later, the same company co-announces a standard built specifically to let agents identify themselves and get explicit permission. Reading PAP as Meta's attempt to build a legitimate front door after getting caught at the back one is an interpretation, not something either company stated — but it's the most direct explanation the public record supports.

It's also not the only standard in this fight. Google, Amazon, Shopify, Microsoft, and Stripe back the Universal Commerce Protocol for discovery and checkout; OpenAI and Stripe back the Agentic Commerce Protocol for in-agent payment; Visa and Cloudflare back the Trusted Agent Protocol for agent identity. Shopify and Stripe are hedging across more than one of these at once. For Meta, sitting out while rivals set the terms was the less attractive option.

What Each Side Gets

For four-year-old Sierra, co-authoring the access standard for agentic commerce moves it from "customer-service AI vendor" toward "the company that wrote the plumbing." For Meta, it gives Muse a sanctioned, consent-based way to transact with retailers — something it currently lacks. For Walmart, Shopify, Stripe, and Genesys, it means a say in the terms under which outside agents touch their systems, rather than having those terms set unilaterally by whichever AI lab's agent shows up next.

Who It Affects

Amazon is the most conspicuous absence, and it's already the most restrictive major platform toward outside agents — a coalition forming around an alternative standard makes that posture more visible by contrast. OpenAI and Anthropic are also missing; Taylor told SiliconANGLE he'd be "really disappointed" if competitors didn't eventually join, but OpenAI already co-owns a rival standard with Stripe, giving it little reason to adopt one it didn't help write. Merchants and payment companies face near-term pressure to support multiple overlapping standards rather than wait for a winner.

What to Watch Next

Three signals separate infrastructure from a press release: whether the v0.1 spec ships with real licensing and governance terms, not just technical docs; whether OpenAI, Anthropic, Amazon, or Google ever join as named partners rather than hoped-for ones; and whether a non-Meta agent can complete a transaction with a non-Sierra business without going through a registry either company controls — the interoperability test Beri proposes checking by January 2027.


Sources