When AI agents do the shopping

English · Nederlands

Ink drawing of a butler walking through a shopping mall with a short list while shop assistants lean out to catch his attention; inset, the customer relaxes at home.

The next visitor to your product page may not be a person. It may be an AI agent comparing your specs, price and delivery terms with ten competitors, on behalf of a customer you never meet. In 2012 I wrote about closing the gap between clicks and bricks for the mobile shopper. Reading it again in 2026, that gap has largely closed, and a new one has opened between the customer and the shop.

What I saw then

The 2012 post started with everyday frustrations. A coupon in a local magazine for a nearby shop, while the brand I bought there pushed a mobile discount for a webshop. A billboard for a weekend in Turkey with no way to book it. Standing in a store, scanning a barcode and buying on Amazon instead. I argued for a level playing field between pure-play e-commerce and bricks and mortar, and against innovation built in silos.

Google’s rumoured move into same-day delivery was my example of how ambiguous new technology looked to retailers: a lifeline and a threat at the same time. The bigger picture was online and offline merging around the customer, with the phone learning our rhythms and preferences. The end game, I wrote, was production and sales matched so closely to demand that waste from imperfect information would disappear.

What is the same now

The customer still wants the answer without the fragmentation. The AI agent is the logical next step of the “automatic, contextual world” I quoted in 2012, and it is now concrete. In September 2025 OpenAI launched Instant Checkout in ChatGPT, built on the Agentic Commerce Protocol it developed with Stripe. The merchant stays merchant of record and handles fulfilment and returns (OpenAI). Google announced its Agent Payments Protocol the same month, developed with more than 60 organisations, using signed “mandates” as proof of what the user actually authorised (The Paypers).

Retail media follows the shopper, wherever that shopper sits. Amazon started placing sponsored products inside the answers of its Rufus shopping assistant in 2024 (TechRadar), and its advertising revenue passed $68 billion in 2025 (Marketing Dive). Whoever controls the assistant controls the shelf.

And the ambiguity is back. In November 2025 Amazon sued Perplexity over its Comet browser agent shopping on customers’ Amazon accounts; Perplexity responded that Amazon was using its market position to stifle competition (Retail Insight Network). Same question as Google’s delivery plans in 2012: is the new intermediary your channel or your competitor?

What is different

In 2012 the phone kept the customer in the loop. They scanned, compared and decided. An agent takes them out of the comparison step. What persuades a human, such as packaging, shelf position or a beautiful product photo, counts for less. What a machine can read, such as attributes, availability, price and terms, counts for more.

That changes the economics, not just the marketing. When an agent compares on price, delivery and returns, undifferentiated products get squeezed on margin. Brands with a clear reason to be chosen, a specification nobody else has or a service promise the agent can verify, keep their price. That is a pricing and positioning question before it is a technology question.

Adoption is also slower than the hype. In March 2026 OpenAI scaled back Instant Checkout, saying the first version “did not offer the level of flexibility that we aspire to provide”, and shifted its focus to product discovery while merchants use their own checkout (TechCrunch). So the agent shapes the choice long before it handles the payment. Plan for agent-assisted research now and agent-completed purchases later.

What to do about it

  1. Fix your product data first. Complete attributes, product identifiers, real availability, price, delivery times and returns terms, structured and consistent across your site, feeds and marketplaces. An agent cannot be charmed. It can only read.
  2. Make your policies agent-readable. Warranty, returns, compatibility, volume pricing: if it only lives in a PDF or behind “contact sales”, the agent will pick the competitor that published it.
  3. Decide your agent policy on purpose. Block, allow or partner. Amazon and Perplexity show that this is a commercial decision, not an IT setting. Agree it with legal, payments and your channel partners.
  4. Move retail media budget with measurement. Sponsored placements are moving into assistants. Test them, but demand reporting on what the assistant showed and why, and keep tracking your organic visibility next to it.
  5. Invest in what agents cannot do. Owning the product today, expert advice, an easy return at a counter. When comparison is automated, physical presence and service become the differentiator.

The new level playing field

In 2012 I wanted a level playing field between the webshop and the store. In 2026 the playing field is a data feed that an agent reads in milliseconds. Make sure yours is the cleanest one it finds.

Sources