Market Commentary

Agentic Commerce: who the shopping agent trusts now decides what gets bought

AI agents that browse, compare and check out on a shopper's behalf are moving from pilot to infrastructure fight. This reads the market across discovery, payment rails and trust — not any single platform's move.

Commissioned by · Rafael Pires

1. The read

Retail's slowest step has always been comparison — the tabs, the reviews, the second-guessing between them. That step is being handed to an agent. Modern Retail expects the AI shopping-agent contest to intensify through 2026, with Amazon, OpenAI and a run of startups each angling to become the default place a purchase happens. What is shifting is not just who sells, but who decides — and a decision, once delegated, tends to stay delegated. Behavioural economics has a name for this: once a default is set, most people never revisit it, because revisiting costs effort and the default already feels like the safe, chosen path.

That default is worth more than any single sale — it is the anchor every future comparison gets measured against, whether or not the agent says so out loud.

2. Why now

Three moments in five months mark the shift from pilot to infrastructure question. First, the framing: Modern Retail's outlook (6 January 2026) named the coming year as the one where the agent-commerce contest goes live. Second, a retraction: Forrester's read (7 March 2026) of OpenAI sunsetting Instant Checkout, after only around thirty Shopify merchants had integrated it, exposed how early the plumbing still is. Third, a scale bet: Amazon's Alexa-based shopping agent, Buy for Me (13 May 2026, TechCrunch), showed a platform with purchase history and device reach moving in regardless.

The sequence reads less like a single wave than like two different bets on where trust gets built first — inside a conversation, or inside a purchase history nobody else holds.

3. How the market is structured

Three layers are forming, and they are not converging on one owner. An agent layer handles discovery and comparison. A retailer layer supplies product data increasingly written for a machine reader rather than a human one. A payment and identity layer decides who is authorised to complete the transaction and who is liable when it goes wrong. Right now, no single actor holds all three — which is exactly why the field looks like a contest rather than a settled stack.

The in-scope column is where the money is already moving. The out-of-scope column is where the market's real risk sits, and it is the larger of the two.

4. Who's moving

The players are not converging on one playbook. Amazon is betting on depth of purchase history and device footprint — a retail memory graph a rival cannot quickly copy, per PYMNTS (14 May 2026). OpenAI has pulled back from owning checkout and leaned into discovery, treating research as the agent's core value rather than the transaction, per Digital Commerce 360 (6 March 2026). Merchants sit downstream of both, adapting product data to whichever reading logic wins ground.

None of these positions is a lead yet. Each is a different theory of where the switching cost will actually sit once shoppers stop comparing for themselves.

5. The evidence

The numbers so far describe appetite outrunning trust, not trust outrunning appetite. Interest in agent-led shopping is well ahead of the share of consumers actually willing to complete a purchase that way, and the one large checkout product withdrawn this year came from the platform framed as the category leader.

Read together, the evidence points to a market that has solved discovery faster than it has solved trust — and trust, not capability, is the binding constraint on the next leg.

6. Where it concentrates

The shift is not landing evenly across the basket. It concentrates hardest where a purchase is repeatable and low-stakes, and lightest where a shopper still wants to see, compare and choose for themselves. Naming the layers where value is actually settling makes the pattern easier to track than treating agentic commerce as one undifferentiated category.

The reorder layer is where agentic commerce is already real. The trust layer is where it is still being negotiated — and it sits underneath every other layer's growth.

7. How it could play out

Two forward states are visible from here, and the market is not committed to either yet. The first keeps agents as a comparison aid, with humans still confirming the final purchase. The second lets agents complete transactions unsupervised for a defined, low-risk category of goods — the reorder layer stretching outward. Which path dominates depends less on what agents can technically do than on whether shoppers accept an unseen ranking logic as a fair one. Framed as projection, not fact: the second path only holds if the trust layer above gets resolved first, and nothing in the evidence so far shows it has been.

The distance between those two states is not a technology gap. It is a trust gap — and closing it is what every platform named above is actually competing to do first.

8. What could reprice it

The read above assumes the trust gap narrows. Several forces could widen it instead, and each would reprice how fast agentic commerce actually scales beyond the reorder layer. The clearest risk is not technical failure — it is a shopper discovering, once, that an agent's top pick was the best-paying placement rather than the best match, in the same way an unclear default can trigger a sharper loss-averse reaction than a merely disappointing one.

Each of those codes is a live repricing risk. A single well-publicised case of undisclosed paid placement would do more to slow adoption than any checkout-tool withdrawal has so far.

9. What to watch

The market's next signal will not be a bigger feature launch — those have already shown they can be withdrawn as fast as they ship. It will be whether any platform discloses its ranking logic, and whether regulators or merchants force the question before shoppers do. Watch the willingness-to-buy figure move past a third; watch whether a second major checkout product gets pulled back; watch whether any platform publishes how it ranks a recommendation.

Until that disclosure happens, agentic commerce keeps growing where the stakes are low and stalling where the choice still matters.

How we read this

This read draws on three cited signals — Modern Retail's 2026 outlook, Forrester's analysis of the Instant Checkout withdrawal, and reporting on Amazon's agentic Alexa launch (TechCrunch, PYMNTS) — plus supporting discovery-trend data (Elogic Commerce, Digital Commerce 360, Opascope). The willingness-to-buy figure and the merchant-integration count are the only hard measurements here; the forward paths in section 7 and the risk scenario in section 8 are Folka's editorial projection, not observed fact. The single number that would most change this read: a disclosed, verifiable figure on what share of agent recommendations reflect paid placement rather than unpaid ranking — no such figure exists in the public record yet.

Grounded in Folka’s corpus: Shopping, One Sentence In · Shopping Through the Agent · Checkout By Voice