Imagine telling your assistant, “Find me a waterproof jacket under a certain price and have it here by Friday,” and it just handles everything. Search, compare, checkout, delivery. That’s the idea behind agentic commerce, and it explains why people searching for google ucp and similar terms are curious about a new kind of shared protocol for AI shopping. Here’s how this might work, and what stands in the way.
Why shopping needs a protocol
Right now, an AI agent that wants to buy something has to pretend to be a person. It reads web pages, finds buttons, fills in forms, and hopes the layout doesn’t change. This is fragile. A redesigned checkout page can break the whole flow, and sites often block automated browsing outright.
A protocol offers a better path. It’s a shared set of rules describing how an agent asks for product details, how a merchant answers, how the cart is built, and how payment is authorized. Instead of scraping, the agent speaks a language the store understands.
Think of how email works. Different providers can exchange messages because they follow common standards. Commerce between agents and stores needs something similar.
What a commerce protocol has to cover
- Discovery. Structured product data: price, availability, size, shipping options, return policy.
- Negotiation of details. Variants, discounts, bundles, and delivery windows.
- Cart and checkout. Creating an order without human clicking.
- Payment authorization. Proving the user approved the purchase, within limits they set.
- Order tracking and returns. Updates and issue resolution after the sale.
Each of these steps needs a standard format so any agent can talk to any compliant merchant.
The trust problem
Letting software spend your money is a big step. Several questions come up immediately.
Who authorized this? The system needs a way to prove the purchase is something you actually approved. Spending limits, allowed merchant lists, and confirmation prompts for larger amounts are natural safeguards.
Is the agent honest? If an agent is paid by certain retailers, does it steer you toward them? Clear disclosure of sponsored results matters as much here as in ordinary search.
Can a merchant trust the agent? Stores worry about fraud and bot abuse. They need ways to verify that an agent represents a real customer.
Who is liable when it goes wrong? If an agent buys the wrong size, or a scam site tricks it, who covers the loss? Existing payment dispute rules weren’t written with this in mind.
What it could change for merchants
For retailers, agent-friendly commerce is both an opportunity and a threat. On the upside, clean product feeds make it easier to be found by a new class of shoppers. On the downside, if agents compare options purely on price and specs, brand storytelling and impulse merchandising lose power. Competition shifts toward data quality, accurate inventory, fast shipping, and fair return policies.
Small merchants could benefit if the protocol is open and cheap to adopt. If it’s expensive or favors giant platforms, the gap could widen.
What it could change for shoppers
Repetitive purchases are the obvious win: restocking household items, reordering printer ink, booking a standard service. Complex purchases, where taste matters, are harder. You may still want to browse for a sofa yourself and let an agent handle the boring parts like price checks and form-filling.
There’s a risk of over-delegation. If you stop looking, you stop noticing mistakes, so good designs keep a clear summary and an easy undo window.
Privacy and data
Agents will handle sensitive information: addresses, payment methods, preferences, and purchase history. Protocols should limit what each party sees. A merchant doesn’t need your full browsing history to fulfill an order, and an agent platform doesn’t need to store card numbers it can tokenize instead.
Open versus closed
The big strategic question is whether these protocols become open standards that many companies adopt, or competing walled gardens. History suggests both happen at once for a while, then one or two standards win. Developers and merchants should watch for documentation quality, governance, and whether independent implementations exist.
What to watch
- Adoption by large payment providers and e-commerce platforms.
- Clear consumer protections for agent-made purchases.
- Security reviews, especially against prompt injection, where malicious content tries to trick an agent into buying something.
- Real usage beyond demos.
Agent shopping is still early. The protocol layer is the unglamorous part, but it’s what decides whether the idea works safely or turns into a mess of broken checkouts and disputed charges.
Featured image: A Furhat Robot Head.jpg by Guojing92, CC0, via Wikimedia Commons.
