Agentic Commerce Meets the Last Mile: When the Shopper Is an Agent, Who Owns the Delivery Promise?
The 2025 holiday season settled the question of whether agentic commerce is real. Salesforce's holiday data put AI-influenced online sales at $262 billion between November 1 and December 31 — 20% of global online sales — inside a record $1.29 trillion season. Retailers operating their own shopper-facing agents grew sales 59% faster than retailers without them, and those agents took direct action on shoppers' behalf at a sharply rising rate, handling 142% more tasks than in the prior two months [1].
The momentum carried straight into 2026. Adobe Analytics, tracking more than one trillion visits to US retail sites, reported AI-referred traffic up 138% year over year in May 2026 — a 1,324% increase since it began tracking in October 2024 — and, more strikingly, that AI-referred visitors now convert 54% better than traffic from non-AI channels, a full reversal from a year earlier when AI referrals converted at roughly half the rate of everything else [2].
Shoppers are delegating the finding, the comparing, and increasingly the buying. The infrastructure race to serve them is well underway. And almost all of it stops at the buy button — which is exactly where the next competition begins.
The Rails Got Built Fast
Two protocol families now define how agents transact with merchants.
The Agentic Commerce Protocol (ACP), co-developed by OpenAI and Stripe, launched in September 2025 with Instant Checkout in ChatGPT — Etsy sellers first, with Shopify merchants to follow — as a single integration through which merchants can sell inside an AI conversation while keeping control of the transaction and fulfillment [3]. The model has already evolved: in March 2026, OpenAI shifted emphasis away from Instant Checkout as a consumer-facing feature toward merchant-controlled checkout, where the agent handles discovery and intent but the merchant retains the final payment step — a pivot that came as Shopify confirmed to Forrester that only around 30 of its merchants were live on Instant Checkout as of February 2026 [4][7].
The Universal Commerce Protocol (UCP) arrived in January 2026, announced by Google at NRF and co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, endorsed by more than 20 partners including Visa, Mastercard, and Stripe [5]. By March, Google had added agent-facing cart support, real-time catalog access — variants, inventory, pricing — and identity linking for loyalty benefits, with simplified onboarding through Merchant Center [6]. UCP-powered buying is live for eligible US merchants across Google's AI Mode, Gemini, and, since May 2026, YouTube Shopping [5].
Note what both protocols have converged on: the agent doesn't replace the merchant — it interrogates the merchant. Catalogs, carts, prices, and inventory are being restructured into formats a machine can query, compare, and act on. That restructuring is the real story of 2026.
Everything Below the Buy Button Is Still Illegible
Here is the uncomfortable counterpart to all that progress. Adobe's May 2026 analysis found that even at the top of the funnel — product content, the thing retailers have optimized for two decades — only about half of retail site content is readable by AI systems: 63% in cosmetics at the high end, 56% in electronics, 51% in sporting goods and apparel, 48% in grocery, 47% in furniture and home [2].
Now consider what sits below the buy button: delivery options, delivery promises, shipping prices, pickup availability, tracking status, returns. At the typical merchant, none of it was ever designed to be read by anything but a human eyeball. Delivery options live in a commerce platform's shipping settings and render as a checkout widget. The delivery promise is a line of text — "3–7 business days" — with no structure behind it. Tracking is a carrier page in a carrier's vocabulary. Returns start in a portal built for a person with a mouse.
An agent can't weigh what it can't parse. When a shopping agent evaluates two merchants with comparable products and prices, the tiebreaker is the thing agents are explicitly built to optimize: total cost and time to the doorstep. A merchant whose delivery options are a rendered widget rather than structured data doesn't lose that comparison — they're absent from it.
The Delivery Promise Becomes an API Contract
This is the deeper shift hiding inside agentic commerce. For a human shopper, the delivery promise is persuasion — a line at checkout that nudges a decision. For a buying agent, the promise is a machine-readable commitment: a price, a date, and conditions the agent evaluates against its instructions ("under $50, arrives by Friday") and then holds the merchant to.
That changes the engineering standard a promise has to meet. A static "3–7 business days" string fails twice: it's too vague for an agent to compare, and too disconnected from operations to be reliable. The promise agents can act on has to be computed — from the fulfillment location that would actually serve the order, real carrier service levels, cut-offs, and working days — and expressed as structured data with a price attached. Merchants who can produce that today are rare. They are also, not coincidentally, the merchants whose human checkout conversion already benefits from credible delivery dates.
And the agent relationship doesn't end at purchase. The same agent that placed the order will be asked "where's my order?" — and will expect an event stream or a queryable status in a consistent vocabulary, not a tracking page. It will initiate the return when the customer asks, and will expect that to be an API call that produces a trackable reverse shipment, not a portal session. Post-purchase, in other words, is about to get its own machine-readability audit — and the industry's baseline is worse there than anywhere Adobe measured.
Who Owns the Promise When the Agent Buys?
One more consequence deserves naming. When an agent chooses a merchant on the strength of a delivery promise and the promise fails, the shopper doesn't blame the agent — the agent was following instructions. The failure lands on the brand, with an extra twist: the agent remembers. A merchant who misses promises becomes, in effect, algorithmically down-ranked at the exact moment of the next purchase decision — not by a marketplace's opaque policy, but by every individual customer's software.
Loyalty in agentic commerce is therefore operational before it is emotional. The merchants who win repeat agent-mediated purchases will be the ones whose promises are computed honestly, kept reliably, and whose failures are visible and recoverable in machine time.
What Merchants Should Do Now
The protocol wars will take years to settle; the direction will not change. Four moves are worth making regardless of which rail wins:
1. Make delivery options structured data. Options, prices, conditions, and pickup availability should exist as an API response before they exist as a widget. 2. Compute the promise; stop asserting it. A delivery date derived from real fulfillment locations, carrier service levels, and cut-offs is the only kind an agent can trust — or compare. 3. Normalize tracking into one vocabulary. Agents (and humans) need canonical statuses across every carrier, exposed as events and queries, not carrier-branded pages. 4. Treat returns as an API. A return request should programmatically create a trackable reverse shipment — the self-service portal becomes one client of that capability, and an agent becomes another.
Two decades ago, merchants rebuilt their stores to be legible to search engines, and the ones who moved early compounded the advantage for years. The same rebuild is now due below the buy button. The agents are already converting 54% better than every channel you've spent the last decade optimizing [2] — the only question is whether, when they reach the delivery section, there's anything there for them to read.
Carriyo — The Intelligent Commerce Platform — makes the journey from checkout to doorstep machine-readable: structured delivery options and computed promises at checkout, canonical tracking across 100+ carrier integrations, and API-first returns.
---
Sources
1. Salesforce. "Salesforce Reveals 2025 Holiday Shopping Data" (November 1 – December 31, 2025: AI influenced 20% of global online sales, worth $262B; record $1.29 trillion global online holiday sales; retailers running their own shopper agents grew sales 59% faster than retailers without; agents handled 142% more tasks than in the prior two months). https://www.salesforce.com/news/stories/2025-holiday-shopping-data/ 2. Adobe Analytics, via Digital Commerce 360. "Adobe: AI-referred traffic to retail sites doubles in a year" (June 17, 2026 — based on more than 1 trillion visits to US retail sites: AI-referred traffic to US retail sites up 138% YoY in May 2026 and up 1,324% since tracking began October 2024; AI-referred traffic converted 54% better than non-AI sources, a reversal from the prior year; share of retail site content readable by AI, May 2026: cosmetics 63%, electronics 56%, sporting goods 51%, apparel 51%, grocery 48%, furniture & home 47%). https://www.digitalcommerce360.com/2026/06/17/adobe-ai-referred-traffic-to-retail-sites-doubles-in-a-year/ 3. OpenAI. "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol" (September 29, 2025 — Instant Checkout launch with Etsy sellers, Shopify merchants to follow; ACP co-developed with Stripe). https://openai.com/index/buy-it-in-chatgpt/ 4. Digital Commerce 360. "OpenAI shifts checkout plans in its agentic commerce strategy" (March 6, 2026 — OpenAI pivot toward merchant-controlled checkout). https://www.digitalcommerce360.com/2026/03/06/openai-shifts-checkout-plans-agentic-commerce-strategy/ 5. TechCrunch. "Google announces a new protocol to facilitate commerce using AI agents" (January 11, 2026 — UCP announced at NRF; co-developed with Shopify, Etsy, Wayfair, Target, and Walmart; endorsed by 20+ partners including Visa, Mastercard, Stripe, American Express, Best Buy, The Home Depot). https://techcrunch.com/2026/01/11/google-announces-a-new-protocol-to-facilitate-commerce-using-ai-agents/ — see also Sundar Pichai's NRF 2026 remarks (https://blog.google/company-news/inside-google/message-ceo/nrf-2026-remarks/); YouTube Shopping availability per Google's Marketing Live shopping update, May 2026 (https://blog.google/products-and-platforms/products/shopping/shopping-updates-google-marketing-live/); secondary explainer: Semrush Blog, "Universal Commerce Protocol (UCP): What You Need to Know" (https://www.semrush.com/blog/universal-commerce-protocol/) 6. Google. "Universal Commerce Protocol updates improve AI shopping for retailers" (March 19, 2026 — cart support, real-time catalog access including variants/inventory/pricing, identity linking; simplified Merchant Center onboarding; implementation partners Commerce Inc, Salesforce, Stripe). https://blog.google/products-and-platforms/products/shopping/ucp-updates/ 7. Forrester (Emily Pfeiffer & Sucharita Kodali). "What It Means That The Leader In Agentic Commerce Just Pulled Back" (March 7, 2026 — "Shopify confirmed to Forrester a month ago that the number was closer to 30 and climbing"). https://www.forrester.com/blogs/what-it-means-that-the-leader-in-agentic-commerce-just-pulled-back/