The Customer Who Never Opens the Site
A shopping agent will not admire your hero. It will ask whether the SKU exists, what it costs, and whether it ships.
A shopping agent will not admire your hero. It will not wait for the parallax. It will ask whether the SKU exists, what it costs, whether it ships, and whether it can be returned
then it will buy, or it will leave, without ever seeing the page you paid to design.
You built a store for a person with a thumb.
The next customer may be a person with an agent.
They will say "buy the refill, the one I got in March, if it's in stock and under twenty." The agent will not scroll.
It will not watch the film.
It will query.
If your product is a photograph and a paragraph, the agent has nothing to hold.
If your product is structured data — title, variant, price, stock, dimensions, policy — the agent can finish the job.
This is not the same as last year's point that sites now have two readers, humans and AI.
That was about being found.
This is about being bought by something that never loads your CSS.
The size of the shift, without the fog
McKinsey (17 Oct 2025) puts agentic commerce — agents that compare, assemble, and check out — at up to $1 trillion of U.S.
B2C retail revenue by 2030, and $3–5 trillion globally.
They also write that half of consumers now use AI when searching the internet.
The page you designed is already downstream of a machine.
OpenAI and Stripe published an Agentic Commerce Protocol.
Shopify is building agent access to catalogs.
Perplexity launched in-chat buying in late 2024.
This is not a mood.
It is plumbing.
Gartner's 2025 prediction, reported at IT Symposium/Xpo, is the B2B twin: agents intermediating 90% of B2B purchases and more than $15 trillion by 2028.
If you sell to businesses, the buyer of record may not have a retina.
What an agent can actually use
An agent cannot be seduced.
It can be informed.
It needs a feed that does not lie: the variant that exists, the price that will be charged, the stock that can ship, the return rule that will apply, the VAT that a German buyer will see.
It needs IDs that survive a rename. It needs policies in language a model can quote without inventing a 30-day window you do not offer.
A hero video is invisible.
A cookie wall that never resolves is a closed door.
A price that only appears after four clicks is a missing field.
The brands that will be bought by agents are the ones that already treated the product page as a specification — the same discipline we argued for returns.
The agent is just a faster auditor.
What this does not mean
It does not mean you delete the store.
People will still walk the rooms you built, for the things they want to feel. McKinsey's follow-up (28 Jan 2026) is explicit: agentic commerce is a curve of delegation, not a flip.
Reorders and refills go first.
A sofa still wants a human.
It means the store has a second door, and that door is an API whether you named it one or not.
How we think about it at BuonaLabs
We build the human page and the machine page from the same object.
One catalog.
Structured variants.
Honest stock.
Policies a crawler and a customer can both read.
The film stays for people.
The feed is for everyone else.
The customer who never opens the site is still your customer.
They just sent someone who does not blink.
Sources
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-automation-curve-in-agentic-commerce
https://www.digitalcommerce360.com/2025/11/28/gartner-ai-agents-15-trillion-in-b2b-purchases-by-2028/
BuonaLabs is a digital innovation studio in Lübeck, Germany, building product pages, digital twins, and commerce experiences that try to tell the truth about the object. If this is the work you are trying to do, let's talk.