Everybody keeps telling you the future of shopping is hands-off. The fridge notices the milk’s low and reorders it, and you never think about detergent again.
I’m sold on the boring half of that. Buying the same laundry powder every week is not a task I’ll miss.
But then somebody asks the real question — hand an AI your card and let it buy whatever it decides is right? — and my hand stops halfway to the mouse.
That “I’d love to, but not with everything” feeling is more common than it looks. And a payments company just went and measured it across six markets. So let’s look at whether hands-off shopping is actually going to happen, from both the wanting side and the trusting side.
”Could I Really Just Hand AI My Credit Card?”
The people who ran the numbers are Checkout.com, a payments company. Their June 2026 report surveyed consumers across six global markets, plus merchants in the US and UK.
They’re a payments firm, so they’re rooting for this future. But the numbers themselves are more interesting than that caveat makes them sound.
is basically this: an AI does the shopping for you, from hunting down the product to putting the order through. The report’s whole question is how far people will let it.
Start with the wanting side. A third of people (33%) expect at least a tenth of their shopping to be AI-led within a year. Not one in four — one in three thinks this is coming, and soon.
The top reason was the dullest possible one: saving time, at 25%. And when you ask what they’d hand over, it’s the stuff nobody wants to think about — groceries (41%) and household basics (31%). The autopilot starts with the boring restocks, not the fun purchases.
But Trust Is Nowhere Near Caught Up
Read only the demand numbers and you’d think this is a done deal. Then the story turns around.
Nearly one in four (24%) say they’ll never let AI make a purchase for them. That’s a shutter coming down at the door. And 27% say they don’t trust the organizations running AI shopping agents at all — not the AI, the companies behind it.
Ask how much people would let an AI spend without checking in first, and the average lands at just £177 per transaction. That’s a “fine, a small experiment” number, not a “here’s my wallet” one.
And this caution isn’t just a gut flinch. Researchers at Columbia and elsewhere built an audit called ACES that watches what shopping agents actually pick — and it found the picks bunch up on a handful of “default” products, and lurch around when the underlying model changes.
One fitness tracker makes the point. It got chosen 45% of the time on Claude Sonnet 4, jumped to 77% on Claude Opus 4.5, then cratered to 6% on GPT-5.1. Same product, three wildly different fates. What the AI buys can swing on the mood of whichever model it’s running — and if shoppers half-sense that, of course they won’t hand it the wallet.
What They Actually Want Is A Brake Pedal
So what would it take? The survey asked that too, and the top three safety features come out looking suspiciously alike:
- Being able to set a spending cap (30%)
- Being able to cancel an order instantly (29%)
- Being able to cancel easily (28%)
Every one of them is a brake. Not the accelerator — go faster, buy easier — but a way to stop.
What people are really asking for is a feeling of control. They don’t want to let go of the wheel entirely; they’ll let the AI drive, but they want a pedal on their own side of the footwell, driving-instructor style. That’s the temperature here.
And that’s the useful bit for a brand. The thing that makes people comfortable isn’t “the AI is smart” — it’s “a human can always cut in.” The order matters.
The Sellers Aren’t Ready Either
Now the merchant side, the people doing the selling. There’s a gap here too.
72% of merchants admit shoppers will move to agent-led buying faster than most stores are ready for. They can see it coming and they know their own feet aren’t under them.
The proof is in their own answers. 89% of merchants say they’re “actively preparing” — but AI agents are involved in just 3% of actual transactions right now. That’s a big space between what they say and what they do.
And 75% of merchants agree they need a way to pull an agent’s permissions in real time. Which points the same direction as the shoppers’ “let me stop it” — demand, trust, and readiness are all moving, just at three different speeds.
Conclusion: Build The Stop Button Before The Smart One
Pull it together and it goes like this. Hands-off shopping has real demand behind it. But trust and seller-readiness haven’t caught up yet. The thing holding it back isn’t horsepower — it’s whether people feel safe handing it over.
So the move for a brand or ecommerce marketer gets pretty clear. Before you race anyone on flashy AI experiences, put the plain controls in first — a spending cap the shopper sets, and an order they can kill on the spot, wired right into your buying flow. The safety features shoppers are begging for and the ones merchants admit they need are the same features. That’s a rare alignment, and a good place to spend early.
The other quiet habit worth starting: check, now and then, how AI agents are actually treating your products. ACES showed the picks move hard on a single model update. “We were on the list last year, and somewhere along the way we dropped off the shortlist” is much easier to miss than a search-ranking slip. So it’s worth watching on a steady cadence.
Maybe I’m reading too much into one survey from a payments company that’s rooting for this future. But whether agentic commerce arrives fast or slow, those two — the control design and the steady watching — don’t go to waste either way. I’d start there, and not in a hurry.
