Blog Folders

 

Introduction

The Agentic Commerce Protocol launched in 2025 as a minimum viable standard — single items, basic checkout — and 2026 is its maturation year: multi-item cart support is becoming standard, because when a user tells an agent to "order everything I need for taco night," the agent must compose ground beef, tortillas, cheese, salsa, and limes into one transaction (MetaRouter, 2026). This piece unpacks what protocol maturity actually demands from retailer data infrastructure — the unglamorous layer that decides who can participate.

Why Protocols, Why Now

Agents are live on ChatGPT, Gemini, Copilot, and Perplexity, completing real purchases — ChatGPT users buy directly from Etsy sellers, with over a million Shopify merchants following, and Copilot Checkout is live in the US (OpenAI, 2025; Ekamoira, 2026). Open protocols are the connective tissue determining who gets discovered and who gets skipped (Ekamoira, 2026). The stakes are quantified: 4.4× higher conversion potential for AI recommendations — but realized only by merchants whose infrastructure supports agent requirements; the gap between potential and median reality is the infrastructure opportunity (McKinsey via MetaRouter, 2026).

What Protocol Maturity Demands — Four Infrastructure Layers

Composable catalog data

Multi-item carts mean agents assemble baskets, not products. That requires machine-readable relationships: substitutes, complements, quantities, pack sizes, dietary/compatibility attributes. A catalog that can't express "this salsa pairs with these tortillas; this is the 12-pack" can't be composed into taco night.

Real-time state

Carts fail on stale data. Inventory, price, and promo state must be current where agents read it — an agent that builds a basket of two in-stock and one phantom item produces a failed checkout and a learned distrust of your data. (Real-time synchronization of inventory, dynamic pricing, and promotional mechanics is exactly what 2026 retail-media strategists are prioritizing — Mirakl, 2026.)

Agent identity & trust plumbing

As agent traffic scales, the first mile of data infrastructure becomes where legitimate agent traffic gets fast-tracked and suspicious traffic gets scrutinized — with transaction logging capturing agent identity, authorization context, and decision signals (MetaRouter, 2026). Retailers need to distinguish good agents from scrapers-with-intent, which means identity-aware ingress, not blanket blocking.

Measurement

2026's forcing mechanism is measurement: boards will demand rigor — AI initiatives live or die by outcomes, not experimentation theater (MetaRouter, 2026). That means instrumenting agent-channel visibility, composition share (how often your products enter agent baskets), and conversion — a new analytics surface most stacks don't have.

The Trajectory — and the Honest Caveats

Projections run from sub-1% of traffic toward 15–25%, with the scaling concentrated in 2026–27 (MetaRouter, 2026). Counterweights worth respecting: about half of consumers remain cautious about fully autonomous purchases (Bain, 2025), and the near-term emphasis — visible in major retailer-AI collaborations — is discovery and comparison rather than instant checkout (industry reporting, 2026). Strategic translation: the discovery layer is where 2026 is won; checkout protocols are where 2027 is won. Build the data layer that serves both.

Where External Data Fits (Our Lane, Stated Plainly)

Three protocol-era problems are measurement-and-data problems:

Composition-share benchmarking — which merchants' products enter agent responses for your category's queries, tracked over time (our agent-readability audits).

Competitive feed intelligence — what attribute depth and freshness the surfaced competitors maintain.

Feed construction — normalizing your catalog into agent-parseable, relationship-rich, fresh feeds (our agent-ready feeds service).

FAQs

Is ACP an official standard everyone must adopt?

It's an emerging open protocol among several connective-tissue efforts in the agent economy; adoption is being driven by platform integrations rather than mandate. The infrastructure requirements above hold across whichever protocols win.

Should retailers block AI agent traffic?

Blanket blocking trades short-term server relief for channel invisibility. The 2026 posture emerging across the industry is differentiated access: identity-aware handling that fast-tracks legitimate agents — paired with a deliberate crawler policy.

What's the first practical step?

Measure your current state: run a purchase-intent query panel against major assistants and see whether your products appear at all. Everything else prioritizes from that baseline.

Does this matter outside the US?

The protocols are global by design; rollout is US-led. Multi-market retailers should treat US agent-channel data as their 12-month preview.