We built a personal agent and sent it to a real store using the Personal Agent Protocol (PAP), Sierra and Meta’s new open standard. We gave it 3 goals: authenticate, browse, and check out. Along the way, we learned that a signed-in agent can shop as the customer, with their size, preferences, and membership details already in hand. Merchants can use that to build a custom experience for each customer. The gaps showed up at checkout: the agent got the shopper to the cart, but connecting PAP to commerce still takes work.
We built a prototype personal agent and sent it to shop a Shopify store set up for PAP. The shopper’s agent enters through PAP, as both a guest and then signed in. A Fabric agent sits in the middle and connects it to Google and Shopify’s Universal Commerce Protocol (UCP), which handles the catalog, cart, and checkout.
PAP gets an agent into your store, but it doesn’t get you found. It covers who the agent is and what it’s allowed to do for you, either as a guest or signed into your account. Signing in works a lot like “Sign in with Google.” The agent signs in on the store’s own page, so it doesn’t need your password. Payments aren’t part of PAP yet; they’re listed as a future extension. Launch partners include Genesys, Shopify, Stripe, Walmart, and others, and while a draft spec is live, it isn’t 1.0 yet.
Goal 1: Authenticate as a guest and sign in
As a guest, our agent could do everything a shopper can, but it didn’t know who it was shopping for. Once signed in, it shopped with the shopper’s authority and what the store already knew: their shoe size, membership tier, favorite colors, and favorite activity (trail running). For a merchant, that changes who’s on your site. A signed-in agent is no longer an anonymous visitor. It’s your customer, with their history attached.
Goal 2: Browse and build a cart
Signed in, the agent started from what the store knew: “Want to look for more trail running shoes? I know you’re a size 9 and prefer blue.” The shopper said yes, and it suggested a few models. When the color they wanted was out of stock in their size, the agent offered other options and added the one they picked to the cart.
PAP got the agent in, but what it could offer came from the store’s catalog. Accurate size and stock data is what kept the agent from hitting a dead end.
Goal 3: Check out
When the shopper asked to check out, the agent created a checkout session, but the shopper still had to open the link and enter their payment info. That’s one gap, but the bigger one is on the merchant side. Shopify provides UCP out of the box for all its merchants, but PAP still takes manual setup, and most platforms don’t support it yet. If you’re on Shopify, PAP is the piece you need to add.
When an agent signs in through PAP, you know it’s acting for a real customer. Everything else is harder to read. One merchant we work with saw a sudden spike in visits. All of it came from desktop Chrome, none from phones, and none of it turned into a purchase. It was bots, and most teams would never have caught it. Until more agents sign in, watch behavior: who’s visiting, what they’re looking for, and what they’re asking.
Nobody agrees on how fast this will move. Watson Weekly rounded up the latest signals in early November. At Retail Club’s AI Festival, a show of hands from more than 30 retailers put agentic commerce at under 5% of sales by the end of 2027. Shopware co-founder Stefan Hamann expects about 30% by 2030, with B2B moving first. Either way, the foundation work is the same: of the 4 “agentic” projects at Agentic Island, only 1 was actually close to agentic. The rest were replatforming jobs. Merchants are also starting to gate agents on purpose. Walmart, a PAP launch partner, has more than 500 applications for its Agent Gateway.
Most of the work that gets you into that top 3 has nothing to do with PAP. Our agent searched the catalog directly, but most agents still shop the way people do. They land on your site, click around, and use your search bar. Start here:
- Make sure agents can get in. Cloudflare rules, firewalls, and bot protection often block agents before they see a single product.
- Make sure your search works. Agents use your search bar, so if a person struggles with it, an agent will too.
- Put your product facts on the product page. Returns, warranty, delivery times, and any guarantees should sit next to the price and buy button, not in a PDF or a separate help center.
- Start measuring agent traffic. Even a rough split between agents and humans is better than nothing.
- Check that UCP is on if you’re on Shopify. Shopify enables it by default, but you can confirm by adding /.well-known/ucp to the end of your store’s URL.
- Ask ChatGPT and Claude about your category the way a customer would, and see if you come up.
Free tools like the Agentic Commerce Alliance scanner check whether your site meets the technical basics for agents. They’re useful, but passing one doesn’t mean an agent will pick you.
Fabric’s AI catalog assessment goes deep and checks what decides whether an agent surfaces your brand:
- Can your catalog answer the questions agents actually ask, including what a product isn’t for?
- Can agents get into your site, and does your search work when they do?
- Where are the gaps and inconsistencies that limit discovery?
Want an assessment for your catalog? Contact us to learn more.
