The Call
Meta's shopping agent can now complete a purchase on every Shopify store on the internet. Amazon asked to be excluded.
The second fact is the one to read closely.
Muse runs agentic checkout through Shop Pay across all Shopify merchants. A shopper asks, the agent searches, compares, sometimes negotiates, and buys. You didn't opt in and you can't fully see it.
Amazon pulled itself off the list, citing rules against automated shopping tools. Sponsored Products only earn money when a human scrolls a results page. An agent doesn't scroll. It picks. Amazon has more to lose here than anyone, and it just told you what it thinks agent traffic does to ad revenue.
You now have a storefront you didn't build. In it your creative doesn't run, your site never loads, and your brand does almost no work. The agent reads structured data. Titles, specs, price, availability, review counts. Then it chooses.
Two months ago the problem with agentic commerce was being invisible to it. Shopify solved that for you without asking. The problem now is being visible on terms you didn't set.
So the call is whether you want to compete on the axes an agent actually uses.
For some brands the honest answer is no. If you win on brand, story, or an experience you control, agent orders stay a small and closely watched slice while you keep investing where you're strong.
For others this is cheap incremental demand arriving before competitors have noticed, and the move is to fix the feed while the field is thin.
Either answer works. What doesn't is finding out in January that 8% of Q4 orders came through an agent at a margin nobody modeled.
One caveat worth holding onto: 404 Media reports some Muse tasks still get handed to Meta staff behind the scenes. The agent is early, and partly human. Early is when you build the plumbing, well before the volume shows up.
If you want an outside read on whether your feed and margin structure can absorb agent-driven demand, that's what our audits cover.
Ask Your Team
"Can we tell an agent order from a human one in our reporting?"
Green flag: yes, and someone can show you the order attribute or tag that does it.
Red flag: "we'd see it in Shop Pay." That's a payment method, not a source. If agent orders blend into your direct revenue, every efficiency number you read this quarter is slightly wrong in a direction you can't measure.
"If an agent compares us against three competitors on price and specs, do we win?"
Green flag: a specific answer by product category, with a named reason. Price, availability, review volume.
Red flag: an answer about brand positioning. The agent isn't reading your brand.
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.
What Good Looks Like
What separates a store an agent picks from one it skips.
Signal | What the agent reads | Where brands fail |
|---|---|---|
Product title | Exact spec, size, model, variant | Marketing language, no attributes |
Price | Current, shipping resolved | Stale feed, shipping revealed late |
Availability | Real-time stock | Backorder discovered at checkout |
Reviews | Count and rating in structured data | Reviews only rendered in page HTML |
Specs | Complete structured attributes | Incomplete, or living inside an image |
None of this is new advice. It's the product feed hygiene that's been sitting on your backlog since 2023, when it only affected Google Shopping. It now decides whether a second storefront exists for you at all.
Ignore This Week
Loud right now. Safe to skip at your size.
Google's antitrust remedies got unsealed. A six-year behavioral regime, no breakup, AdX must bid into rival ad servers. Third time this story has surfaced in a month. Nothing changes about how you buy.
Google's Customer Lifecycle Goals. Skip them unless your customer list is at least 1% of your target population. At your size a simple customer-list exclusion does the same job without the misconfiguration risk.
Not ignorable: OpenAI is testing a cross-site cookie that can persist for a year and send page and conversion data back from any site running its ad pixel.
It reportedly fires under analytics-only consent, even when a visitor has declined marketing cookies.
If you're testing ChatGPT ads, that exposure sits on your site rather than OpenAI's. Get a consent-platform check and a legal read before anyone scales the pixel.
Are you treating agents as a channel yet, or waiting for the first attributed sale? Hit reply.

