Field note2026-09-14T15:38:25.120Z

What it means to be found by an agent that already knows the person asking.

A standing AI agent that already knows its person skips the broad question a search engine expects entirely. It arrives already narrowed, checking a shop's facts against one specific person's specific needs.

4 min read

The Selfe team
Agentic commerce infrastructure
Say someone's set up a standing AI agent that already knows their size, their usual budget for outdoor kit, and that they run cold on long walks. Asked to find a new waterproof jacket, it doesn't start from "waterproof jackets" the way a search engine would. It starts from exactly the version of that question its person would ask: this size, this budget, warm enough for someone who runs cold. A shop being checked this way was never being compared on the general category in the first place. It's being checked against a specific, already-known set of requirements, whether it knows that or not.

Why general visibility does nothing for this case

Say someone's set up a standing AI agent that already knows their size, their usual budget for outdoor kit, and that they run cold on long walks. Asked to find a new waterproof jacket, it doesn't start from "waterproof jackets" the way a search engine would. It starts from exactly the version of that question its person would ask: this size, this budget, warm enough for someone who runs cold. A shop being checked this way was never being compared on the general category in the first place. It's being checked against a specific, already-known set of requirements, whether it knows that or not.

This changes what "being found" even means. A shop found by a person typing "waterproof jacket" into a search engine is competing on broad visibility. A shop checked by an agent that already knows its person is competing on whether its specific facts, this size in stock, this price, warm enough for someone who runs cold, match a request that was never typed out loud anywhere the shop could optimise for.

A shop that's excellent at ranking for "waterproof jacket" broadly has no advantage here, because nothing here was searched for broadly in the first place. The agent went straight to checking specific shops' specific facts against a request it already had fully formed. Being generally findable and being specifically matchable are different achievements, and this is the one case where only the second counts.

What gets checked

The exact things the agent already knows about its person, translated into exact things it can check about a product: is this specific size in stock, not just "sizes available"; does this specific price match the stated budget, not just "affordable"; does anything in the product's own facts address the warmth question directly, not just a generic comfort claim. A shop with only general facts has nothing to match against a request this specific, and gets skipped in favour of one that does.

Nothing about this looks like a sale from the shop's side

The shop never sees a browsing session, a cart being filled slowly, or any of the usual signals a sale is close. The agent checks the facts, finds a match or doesn't, and either completes the order in one pass or moves to the next option, all without anything that would normally show up as interest. A shop watching for the moment a customer is about to convert has nothing to watch for here. The only signal available beforehand is whether the facts themselves would have passed the check, knowable in advance, unlike a browsing session that can't be seen until it's already happening.

Why this is a growing, not a niche, case

A shop being checked this way looks unusual today. Most shopping still runs through a browsing person or a one-off search. It's also the shape everything else is moving towards: an agent that already knows its person, acting on their behalf, rather than a person searching fresh each time. The same specific, checkable facts this whole hub has been about serve this case too, aimed at a request that arrives already narrowed rather than broad, not a separate job on top of it. That's the layer Selfe exists to build: keeping a shop's facts specific and current enough that whichever kind of request arrives, broad or already narrowed by someone's own agent, the match holds.

How would we even know this kind of customer exists for us?

Check repeat-purchase patterns with no browsing activity attached. That pattern is the tell.

Do we need different facts for this versus a person searching normally?

No, the same specific, checkable facts serve both. A standing agent just has less patience for anything that isn't already specific and current.