Shopping agents are already reading store catalogs and recommending products. Your store has a stance on them whether you picked one or not. This week, pick it on purpose and write it down.
Product data goes out. Customer data stays home.
What your peers chose
On Oct 21, BCG published a survey of 50 of the top 300 global retailers and 100 top brands. Per BCG, 42% structure their data so any agent can discover their products. 32% give access only to chosen agents through direct feeds. 6% block agents. 20% are still deciding.
BCG also says 68% expect a significant or transformational effect within two years, and only 18% say their marketplaces are ready for agents.
Keep the source in mind. These are large retailers and brands surveyed by a consultancy that sells advice on this exact question. A 12 person DTC brand may land somewhere else. The split is still useful, because it shows the question has three real answers.
The two poles are easy to see. Yahoo Finance reported on a Bank of America note on Oct 20 that frames Walmart as open, with its Sparky assistant and ties to several AI platforms including Meta's Muse. Amazon blocks Muse from purchasing and bars other companies' automated shopping tools. The note also flags open questions about what agent checkout does to advertising.
A day later, both sides got more concrete. Walmart announced Agent Profiles, launching within the next month: each customer's AI agent gets its own profile, with a spending limit or an approval on every purchase, and Walmart published terms for agents that shop its site. BigCommerce went further and turned agent shopping on by default for eligible stores. Agents can search, compare and fill a cart, checkout stays on your storefront, and you can switch it off in a new settings menu. If you sell on BigCommerce, your store already has a stance. Check the setting before you write the policy.
The policy, in five decisions
Plan on an afternoon. Get the founder, whoever owns the store, and whoever owns marketing in one room.
1. Pick a stance
Three options, from BCG's buckets:
- Open. Any agent can read your catalog.
- Selective. Only agents you choose get a feed or a login path.
- Block. You keep agents out and sell through your own site and apps.
BCG's rule of thumb: brands with strong direct appeal can restrict more, and brands that depend on reach need to share more. Be honest about which one you are. If most new buyers find you through search, marketplaces or ads, blocking agents cuts off one more way to be found.
Write the stance as one sentence at the top of the page.
2. Draw the data line
BCG's line is clear. Product, inventory, pricing, offer and fulfillment data should be reachable by agents. Customer identities, preferences, loyalty histories and behavioral data stay with the retailer.
Put two columns on the page. Left column, what agents can read. Right column, what never leaves your systems. Then check your actual feeds and APIs against the left column. Old feeds tend to carry more than anyone remembers.
3. Make the open half readable
If you picked open or selective, an agent can only recommend what it can parse. Clean product data comes first: titles, variants, prices, stock, shipping times.
On Shopify, Vercel Shop is one concrete reference from Oct 16. Vercel says the same product URL can serve a visual page or machine readable product info, with structured data and commerce tools agents can find at the merchant's own domain.
Then test it. We wrote up how to send a shopping agent through your store and see what it gets wrong.
4. For selective, authenticate instead of guessing
Selective only works if you can tell which agent is which. On Oct 20, Visa, BILL and Billtrust announced a live pilot where a registered agent pays approved supplier invoices. The agent uses tokenized credentials, its payments are tied to instructions the buyer already approved, and there are no shared portal logins. Visa says this authentication "can satisfy the site protections historically provided by CAPTCHA and other bot blocking mechanisms."
That pilot is B2B invoice payment, so it doesn't mean your consumer checkout can drop its bot checks. Take the pattern: a named agent, its own credentials, limits tied to what a person approved. Write down which agents you allow and how each one proves who it is. The same rule applies to your own agents, which is why we give each Always On Agent its own seat.
5. Price the middleman and measure the traffic
BCG expects agent routed demand to stop being free at some point, and points to online travel agencies charging roughly 15% to 25% per booking as the comparison. Add an agent channel line to your margin model now, even if it reads zero today.
Decide what loyalty perks an agent order gets. Per BCG, only 14% of surveyed retailers plan to give external agents full loyalty benefits, and 58% plan a lower tier.
Then tag agent referrers in your analytics. BCG says agent referrals were under 1% of visits at leading European brands in late 2025, could reach 5% by the end of 2026, and the true traffic effect can be up to four times the referral count. Last click will undercount it. Watch assisted visits too.
What this doesn't tell you
It doesn't tell you the right stance for your store. The BCG numbers describe big retailers. The BofA view comes to us secondhand, through Yahoo Finance, with no figures attached. And the Visa pilot is one B2B flow at a scale Visa hasn't disclosed.
It does give you a peer benchmark, a data line you can defend, and a working pattern for letting known agents in.
This week
Write one page. Your stance in one sentence. The two column data line. The list of agents you allow and how each one is identified. The agent channel line in your margins. The referrer tags you'll watch. Put a date on it and review it next quarter.
If you want that policy wired into how your store and your own agents actually run, Start your build.
Linked sources
- BCG: How Retailers Can Reinvent the Digital Flywheel for Agentic Commerce (2026-10-21)
- Yahoo Finance (Hamza Shaban), reporting a Bank of America note led by Christopher Nardone (2026-10-20)
- Visa: Visa, BILL and Billtrust Launch First Live Pilot for Trusted AI Agent Payments (2026-10-20)
- Vercel: Vercel and Shopify make agents first class shoppers (2026-10-16)
- Walmart: Making It Possible for Agents to Shop on Behalf of Customers (2026-10-21)
- Commerce.com: Commerce and Cloudflare bring AI shopping assistants to BigCommerce (2026-10-21)
- Related article: Send a shopping agent through your store
- Related article: Give each Always On Agent its own seat, not your login
Get our Articles and posts by email.
Up to one email a day with new Articles and posts from 22atlas.com and sean-weldon.com, plus occasional news about 22Atlas services.
Also on the desk
Technique
Send a shopping agent through your store before Muse's new model does
Meta's next Muse model is trained to try another route when a website stops it. Before it reaches your product pages, run real errands through an agent yourself and fix what it trips on, in six steps.
Technique
Give each Always On Agent its own seat, not your login
Google's new Gemini coworker agent gets its own account, its own audit trail and a spend cap that pauses it. You can set up your agents the same way today on the tools you already pay for. Five moves.