Recently, I made the mistake of going into Target during peak back-to-school week. I don’t have kids, so this was 100% an unforced error. I’m a lifelong urban dweller who lives in three places and gets “it” delivered whenever I can. But I still love a trip to Target. Usually. This time I stood trapped in a sitcom, watching a mom with three irascible under-10s negotiate their way to a chocolate bar, a bag of gummies, and a pack of Tic Tacs in the candy aisle. She gave in as the checkout line inched forward.
I have never once bought clothes on target.com. Not once. I have also never left a Target store without a handful of clothing. One top has been in rotation for upwards of a decade and pulls multiple compliments every single time I wear it (thank you, WhoWhatWear capsule collection, circa 2014). Target clothes are my candy aisle.
Today, people’s baskets don’t match their list. The chocolate bar, the top, the Trader Joe’s run where you went in for two things and came out with a full bag of stuff you didn’t know existed until it was in your hand. The difference between the list and the basket is a lot of the business and the revenue and joy for the consumer. When everything is agentic, I’m going to miss the top I spied out of the corner of my eye on my way to the Swiffer dusters and the Colgate Max.
The agents are coming for the errand, and mostly, good. The list, the compare, the reorder: an instruction instead of a trip. But an agent shops the list, and the list is the smallest version of what a consumer might want. It leaves a lot of potential on the table.
Teaching AI the “10% tolerance”
Most of us have run this experiment ourselves with AI. I asked Claude to find me a hotel and forgot to include a budget and a few other choice criteria, so it came back with a list of places to stay, none of which actually worked for all the reasons I forgot in the first prompt wave. I added more rules and got obedience: the letter of the law, nothing I loved. So I added a third line: it’s okay to go 10% over or 5 miles farther, because I’d rather find a great place that doesn’t quite fit than get exactly what I asked for. That last sentence really matters, having a tolerance. It lets me tell the machine how much I’m willing to be surprised.
We’ve been thinking in these mental models for years as humans and businesses. Now we call it governance. The gas pump preauthorizes more than you’ll probably fill. Corporate cards run on categories and opt-ins, spend thresholds built on the assumption that a rigid plan won’t survive contact with the week. And sometimes we look for opportunities to give up control. If you order the Tasting Menu, you hand a stranger your entire dinner. Take it out of our hands. But only so much: I picked the place, and it has a Michelin rating.
We surrender happily inside a container we chose. Machinery for bounded surrender already exists in commerce; we haven’t pointed it at wandering yet.
Why serendipity matters for commerce
There’s a real argument (Kyle Chayka has made it best) that serendipity can’t be engineered at all, because serendipity requires agency—the finding has to be yours. As Brian Christian points out in Algorithms to Live By, human decision-making relies on a deliberate balance between exploration and exploitation:
“The explore/exploit tradeoff tells us how to find the balance between trying new things and enjoying our favorites… Optimism minimizes regret.”
An agent can manufacture surprise; it can’t hand me the moment my own eye snags on a mustard-and-black pattern nobody will miss in the store or, as it turns out, on the street. A decade of compliments started in that moment. So did my relationship with the store and a category of wares. Serendipity isn’t a nice-to-have in commerce. It’s how the business gets fed.
The three futures for brands
If you’re a brand, this sorts your future into three positions:
- The Name Brand: You’re the one somebody asks for by name, which will increasingly be earned after the sale—via the product, the packaging, the experience, the reason to want it again.
- The 10% Tolerated: You’re the one an agent finds inside somebody’s margin of error, which is a pricing and positioning job played entirely on the agent’s terms.
- The Trusted Curator: You’re the one people hand the keys to—the tasting menu, the curated subscription box, or the Trader Joe’s endcap where you walk in for almond milk and leave with chili-lime rolled tortilla chips you didn’t know existed. This is the rarest position, sitting upstream of the agent entirely.
They take different work, and most GTM plans haven’t picked a lane. The open questions belong to everyone else in the chain. How much surprise will people actually permission, and who asks them? What does the tolerance setting look like on a checkout rail built to verify intent in advance?
For marketers, the winning move isn’t resisting agentic commerce—it’s optimizing for it while aggressively protecting the spaces where delight happens. Teach the machines how to respect human curiosity, design your GTM strategy around tolerance parameters, and build experiences worth stumbling upon.
My agent can have the Swiffer dusters. The candy aisle is mine.