AI agents are supposed to be the next big leap in how we use the internet. Instead of clicking through websites ourselves, we will ask a personal assistant to check stock, compare prices, change a flight, or reorder groceries. There is just one problem: many websites do not know how to handle these agents — and a growing number are simply blocking them.
On October 6, 2026, Meta and customer-service AI company Sierra proposed a fix. Their new Personal Agent Protocol is an open standard that defines how personal AI agents identify themselves to businesses and what those businesses allow them to do. Here is what it is, how it works, and why its success depends heavily on companies that have not signed up yet.
The Problem: Agents Keep Getting Locked Out
As AI agents have become more capable, users have increasingly complained that their assistants hit walls when trying to complete tasks online. TechCrunch documented cases at major sites including Amazon, Delta, United, Zillow, Yelp, and eBay.
Not all of these blocks are intentional. Some companies, such as Amazon, have deliberately chosen to restrict outside agents. Others appear to be accidental side effects of anti-bot systems. In particular, an update to Cloudflare’s crawler protections in September began catching legitimate AI agents along with unwanted scrapers.
From a business perspective, the hesitation is understandable. An unknown bot could be a scraper, a fraudster, or a genuine customer’s assistant — and most websites have no reliable way to tell the difference. Sierra’s argument is that rejecting a personal agent effectively means rejecting the customer who sent it.
How the Personal Agent Protocol Works
The protocol is built on OAuth, the same widely used standard that powers “Sign in with Google” or “Log in with Facebook” buttons across the web. Instead of a person logging in, an AI agent authenticates on that person’s behalf and declares what it is there to do.
Access is tiered:
- Guest mode: An agent can perform simple, low-risk tasks without logging in, such as checking product availability or reading a returns policy.
- Signed-in mode: Once the customer logs in, they decide whether the agent receives read-only access or permission to take actions, such as placing an order.
- Business controls: The company sets which actions are allowed in the first place.
Sessions are also designed to carry across channels. That means a question an agent asks before login and an order it places afterward can be treated as part of the same visit, rather than as two unrelated interactions.
Businesses keep flexibility over how agents connect. According to the announcement from Sierra co-founders Bret Taylor and Clay Bavor, a company can let agents work through its regular website, through its APIs using standards such as MCP and OpenAPI, or through a conversational agent of its own.
Who Has Signed On
Six founding partners were named at launch:
- Walmart
- Shopify
- Stripe
- Rocket
- Genesys
- Instinct
Walmart and Shopify give the effort significant retail reach, while Stripe adds credibility on the payments side. A version 0.1 specification was scheduled for release later in October 2026. Features such as payments, push notifications, and more detailed permission controls are flagged as future additions rather than part of the first version.
Who Is Missing — and Why It Matters
The most notable thing about the launch may be the companies that are not on the list. Amazon, OpenAI, and Anthropic — three of the platforms best positioned to control how agent traffic flows — are not founding partners.
OpenAI’s absence stands out for a specific reason: Bret Taylor, who co-founded Sierra and is leading this initiative, also serves as chair of OpenAI’s board. Observers have pointed to that overlap as an awkward gap in a standard that depends on broad adoption.
Without the largest agent builders and the largest online retailer, the protocol risks becoming one option among several rather than the default way agents and businesses connect.
A Crowded Field of Agent Standards
Meta and Sierra are not alone. In June 2026, Visa launched its own Trusted Agent Protocol, which already allows agents to buy from roughly 175 million merchants. Interestingly, Stripe and Shopify are participating in both efforts, suggesting that commerce platforms are hedging their bets until a clear winner emerges.
The key difference is focus. Visa’s approach centers on transactions, while the Personal Agent Protocol starts with identity and permissions — establishing who an agent represents before worrying about how it pays.
Will Consumers Trust It?
Even a well-designed standard faces a human obstacle. Reporting from Forkast cited data showing that only about 3% of US adults trust AI agents to make purchases for them. That suggests the technology may be ahead of public comfort.
Clear permission controls, such as the read-only versus write-access choice, could help close that gap. Being able to see and limit what an agent is allowed to do is likely to matter as much as what the agent can technically achieve.
What It Means for Online Businesses
For store owners and digital marketers, the message is that AI agents are becoming a real customer channel. Practical steps worth considering include:
- Checking whether your bot protection is accidentally blocking legitimate agents.
- Making key information — stock, shipping, return policies — easy for machines to read.
- Following the major agent standards so you are ready when your platform adds support.
Frequently Asked Questions
Is the Personal Agent Protocol free to use?
It is presented as an open standard, meaning the specification is publicly available for businesses and developers to adopt.
Can agents make payments with it?
Not in the first version. Payments are listed as a planned future extension.
Does it work with ChatGPT or Claude?
OpenAI and Anthropic were not founding partners, so support from their assistants has not been announced.
The Bottom Line
The Personal Agent Protocol tackles a real and growing problem: businesses need a trustworthy way to know which AI agent they are dealing with and what it is allowed to do. With Walmart, Shopify, and Stripe on board, it has a credible start. Whether it becomes the internet’s front door for AI agents will depend on whether the biggest missing players decide to walk through it.