AI
Agentic checkout is quietly rewiring the front door of online retail
Assistants that browse, compare and buy on a shopper's behalf are moving from demo to production. Retailers now face a harder question than adoption: who owns the customer relationship when software does the clicking?
Key takeaways
- Agentic shopping assistants are completing purchases in single machine-readable transactions
- Merchants are shifting feed quality work from marketing teams into engineering teams
- Card networks are piloting delegated credentials with hard spend ceilings for AI buyers

For two years, agentic shopping lived in keynote slides. This quarter it started showing up in payment logs. Large merchants report a small but fast-growing share of sessions that never touch a product page: an assistant queries a catalogue feed, resolves stock and price, and completes a purchase in a single machine-readable round trip.
The interface is disappearing
When an agent is the buyer, merchandising loses most of its leverage. Hero banners, bundles and urgency cues are written for human attention. Structured data, delivery promises and returns policy are what an agent actually reads — and the merchants adapting fastest are the ones treating their product feed as the new storefront.
That shift reprices the whole funnel. Paid placement matters less when discovery happens inside a model's context window; conversion depends on whether your feed answers the question cleanly. Several retailers we spoke with have moved feed quality out of marketing and into engineering.
Trust becomes the bottleneck
The unresolved problem is authorisation. An agent that can buy can also over-buy, and dispute frameworks were designed around a human pressing a button. Card networks are piloting delegated credentials with hard spend ceilings, but the rules are still being written in public.
“The winning merchants will be the ones whose data is boringly correct.”
Expect the next 12 months to look less like a consumer revolution and more like plumbing: catalogue hygiene, machine-readable policies, and settlement rules that assume software is on the other end of the transaction.
Why it matters
Retailers and payment providers must adapt product data and authorisation rules for software buyers rather than humans. Shoppers may see purchasing decisions increasingly made by assistants rather than storefront browsing.
Sources and references
- Advisable analysis
- Updated 17 Sept 2026 · 09:10 UTC
- AI-assisted draft, edited and fact-checked before publication · Reviewed by Mara Vance.
Corrections and updates
- Update
Added detail on delegated card credentials after card-network briefings.
In this story
- Agentic commerce
- Online retail
- Card network
Related stories

Inference costs fell again this quarter — but not for everyone
Headline per-token pricing keeps dropping. Long-context and tool-heavy workloads are moving the other way, and that gap is reshaping product roadmaps.
Daniel Okafor · 20 hr ago · 6 min read

Small models on device start winning real workloads
Latency, privacy and cost are pushing classification, extraction and routing off the cloud — and product teams are noticing users prefer it.
Daniel Okafor · 3 days ago · 5 min read

Cross-border duty changes reprice fast fashion overnight
Low-value import exemptions are narrowing across several major markets. Direct-from-factory pricing models lose their structural edge.
Inês Ferreira · 4 days ago · 6 min read
Daily Brief
One email each morning. The stories that actually change decisions.
Five minutes of e-commerce, AI, payments, business and startup news, edited by humans and sent before markets open.
No spam. Unsubscribe in one click.