AI agents can now search, compare and buy on a shopper's behalf, at scale, inside the platforms consumers already use. The brands that win this channel will not be the ones with the cleverest chatbot. They will be the ones that decide, deliberately, what an agent is allowed to see and buy from them, and who owns that decision.
Most consumer businesses have not made that decision yet. In the last three weeks the market made it harder to postpone.
What changed in the last three weeks?
The infrastructure for agent-led buying went live almost at once. On 30 September, Mastercard expanded Agent Pay with a probability score that estimates whether a transaction was started by an AI agent, now in testing in the US. On 5 October, TikTok introduced a conversational shopping assistant and one-click in-feed checkout, built with Salesforce, Shopify and Stripe among others, and initially offered to advertisers through eligibility-based testing. On 13 October, Shopify opened its Catalog, which it says holds billions of products, to personal AI agents that can search, prepare checkouts, place approved orders and reorder within spending limits the shopper sets.
Agents are not waiting politely for brands to get ready. Security firm HUMAN, quoted by Digiday, logged 5.6 million requests from Meta's Muse shopping agent in under two days after it started tracking it, a volume that took ChatGPT 11 months to reach. 84% of those requests targeted product and search pages.
That is the gap. The pipes are ready; the commercial policy is not. The same PYMNTS survey, as reported by FinTech Weekly, found that 93% of merchants believe the AI or agent provider should bear the loss when an agent makes a wrong purchase. Checkout.com's own merchant research, cited in the same piece, puts agent-involved transactions at just 3% today, while 89% of merchants say they are preparing. Those are self-reported vendor figures, but the direction is clear: almost everyone is getting ready, and very few have decided what "ready" means.
Why is this a commercial decision, not a technology project?
Because an agent is a new kind of shelf, and someone has to set its terms. Call it the agent offer: the range, prices, pack sizes, promotions, return policies and product data a brand deliberately exposes to AI shopping agents.
Today most brands have an agent offer by accident. It is whatever their product feeds, retailer listings and site markup happen to say. Nobody in commercial leadership signed it off. Yet it already shapes outcomes. Jellyfish, also cited by Digiday, found that when Alexa for Shopping surfaced toys from 177 brands, nearly every one came from Amazon's own store, while ChatGPT pointed shoppers to 24 retailers and Google's AI Mode to 37. Each agent has its own routes to product, and a brand absent from those routes is simply not considered.
The value of getting it right is also showing. Adobe data reported by Digiday shows AI-referred shoppers generated 53% more revenue per visit than other traffic to US retail sites in July. That is a per-visit figure, not a share of total sales, but it says these shoppers arrive further down the funnel. The agent has already done the comparing.
Which decisions actually need redesigning?
Four, and they sit squarely in the commercial function.
- Range. Which SKUs should agents be able to buy directly, and which should stay in channels where the brand controls the experience? A full-range, same-terms policy is a choice, not a default, as the 28% figure shows.
- Price and promotion. Agents compare relentlessly. Price-pack architecture designed for a physical shelf may expose awkward per-unit comparisons once a machine lines products up. Revenue growth management teams need an explicit view of how each pack and promotion reads to an agent.
- Terms and liability. Returns, substitutions and wrong orders will happen. With 93% of merchants expecting the agent provider to carry the loss, and providers unlikely to accept that wholesale, brands and retailers need a negotiated position, not an assumption.
- Channel conflict. Agents will shift volume between retailers, marketplaces and direct channels without asking. FinTech Weekly reports that Amazon began blocking purchases through Meta's Muse in September. Brands selling through retail partners need a position on where they want agent demand to land before the platforms decide for them.
This is why AI pays where the decision is redesigned, not where a model is bolted on. Plugging a catalogue into an agent protocol takes weeks. Agreeing range, price and liability rules for a channel that sits between retailers and consumers takes leadership. The question of who owns the AI P&L arrives here in its most practical form.
What to do now
- Name an owner for the agent offer (CCO, with eCommerce and RGM). Within 30 days, one accountable leader for what agents can see and buy, with a cross-functional team behind them.
- Audit what agents see today (eCommerce lead). Within 60 days, test your top categories in the major assistants and agent platforms. Record where you appear, at what price, from which seller, with which product claims.
- Write an agent range and pricing policy (RGM director). Within 90 days, decide which SKUs, packs and promotions are agent-eligible, and how per-unit comparisons will read.
- Agree terms with key retail partners (Sales director, customer teams). Within 90 days, open the conversation on agent-led orders, returns, substitutions and who carries the cost of errors.
- Fix the product data (Data and digital lead). Within 90 days, make structured attributes, pack sizes, ingredients and policies complete and consistent across feeds. For an agent, the data is the packaging.
The bottom line
Agentic commerce has moved from demos to default settings in a single month. Volumes are still small, which is exactly why now is the time to decide. The brands that treat the agent as a commercial channel, with an owner, a range, a price logic and terms, will shape how agents buy. The rest will be shaped by them.
Questions leaders ask
What is agentic commerce?
Agentic commerce is buying in which an AI agent searches, compares and completes purchases on a person's behalf, usually within limits the person sets, such as a spending cap or approval step.
Do consumer brands need to sell directly to agents?
Not necessarily. Many will reach agents through retailers and marketplaces. But every brand needs a deliberate policy on which products, prices and terms agents encounter, whichever route they take.
Is it too early to act if agent transactions are only a few percent?
No. The rules of a new channel are set while it is small. Range, pricing and liability positions agreed now are far cheaper than renegotiating them once agent volume is material.
