Meta and Sierra introduced the Personal Agent Protocol, an open standard for how AI agents like Meta’s Muse and Instinct’s personal assistant sign into and act on company websites, with Walmart, Shopify, and Stripe among its key launch partners. Sierra, an enterprise AI startup that builds customer service agents for brands, was co-founded by former Salesforce co-CEO Bret Taylor, who also chairs OpenAI’s board.
Taylor, who is leading the initiative, compared it to logging into websites with your Google or Facebook account, technology he worked on as Facebook’s CTO. Both of those login technologies run on OAuth, an open standard that lets you sign into a site without handing it your password, which is the same foundation the new Personal Agent Protocol is built on.
Taylor said the standard will let businesses know when they’re dealing with a bot and see what personal agents are actually doing on their sites.
“For a lot of companies there’s a risk: you don’t want just a random bot that isn’t acting on behalf of a person to have access to this service. It is kind of chaos until such a standard exists.”
Here’s what we know about the Personal Agent Protocol so far:
- Agents can act as a guest for low-stakes tasks like checking stock or reading a return policy, or sign in to the customer’s account for account tasks like changing an order, processing a return, or booking an appointment.
- Customers choose whether an agent gets read-only or write access.
- Businesses choose which actions agents can take.
- Businesses also choose the entry point, which can be either their website, APIs, or their own customer-service agent.
- The protocol keeps the agent in one continuous session as it moves from guest to signed-in customer, and from a company’s website to its APIs or its own support agent, so that the entire visit is tied to a real customer rather than a string of anonymous bot requests.
- Businesses get full visibility into what agents are doing on their sites, same as they can now with human visitors. The lack of transparency around bot traffic is businesses’ main gripe with bots today.
- There are no payments built into the protocol yet, but the ability for agents to buy without handling card numbers is listed as a future extension, along with more granular permissions and push notifications.
Who’s on board and who’s competing?
- OpenAI and Anthropic haven’t signed on, but Taylor told CNBC he expects both to join.
- Google and Amazon aren’t mentioned, and it might be a challenge to get Amazon onboard given how anti-outside-bot they are. Though maybe a protocol can change that.
- It’s one of several competing agent standards. For example, Visa’s Trusted Agent Protocol (with Microsoft, Stripe, Shopify, and Worldpay) launched last October, so Stripe and Shopify are hedging across both, same as they did with OpenAI’s ACP and Google’s UCP in the early days.
One big question: Is this new protocol even needed?
It feels more like a feature than a protocol. Meaning, there are already numerous agentic protocols that handle most of what PAP does, minus a couple of items that could be added.
- It’s built on OAuth, which already handles “let this app act on my account with limited permissions.”
- MCP already supports OAuth sign-in, and PAP can run over MCP.
- Visa’s Trusted Agent Protocol and Cloudflare’s Web Bot Auth already tackle “is this a legitimate agent?” Though who decides what’s “legitimate” will definitely be an issue, given the incentive the Big Tech companies building these protocols have to keep “illegitimate” smaller competitors off their turf.
- UCP and ACP already handle checkout, and payments aren’t even in PAP v0.1.
- The only feature that appears to be missing from the current lineup is defining how a personal agent moves from a guest to a signed-in customer across a company’s properties, but that seems like a layer that could be added to UCP and ACP, as opposed to facilitated through a separate protocol.
Everybody wants a seat at the table creating the playbook that best aligns with their objectives, but the end result is that we wind up with a lot of playbooks. My biggest priority is that we arrive at a playbook that creates an even playing field for startups, independent developers, and Big Tech alike. Otherwise, we’re constructing a wall around agentic commerce that small developers won’t be able to afford to enter.
What are your thoughts on PAP? Join the conversation on my LinkedIn post.
