What is agentic commerce?
Agentic commerce is shopping where an AI agent (not the buyer) finds products, compares options, assembles carts, and in some flows completes the purchase, acting on the buyer's instructions. The buyer states intent ("restock my espresso setup under $80"); the agent does the clicking. Your store participates by being machine-readable, or it doesn't participate.
What is Google's Universal Cart?
Universal Cart aggregates products from multiple retailers into one cart inside Google surfaces, starting with Search and Gemini, with YouTube and Gmail support planned. Google handles discovery and price comparison; the transaction still completes on the merchant's own site. It launches in the US in summer 2026, with checkout experiences expanding to Canada and Australia, then the UK.
Two details matter for merchants. First, launch partners are heavyweights: Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify, so the surface will have buyer trust on day one. Second, stores must be Universal Commerce Protocol (UCP)-enabled to appear in carts at all. UCP readiness is the immediate to-do for brands that want in. A related Agent Payments Protocol (AP2) adds budget and brand-preference guardrails for agent-driven purchases.
What are "information agents"?
Agents inside Google Search (AI Pro and Ultra subscriptions, launching summer 2026) that keep monitoring the web after an initial query and deliver synthesized updates with the ability to take action. For a store, that means a buyer's agent might be watching your category for a restock or price change, and acting the moment it happens. Stale availability data doesn't just lose a sale; it misinforms a standing watcher.
What is WebMCP?
WebMCP is a proposed open web standard, now in an origin trial as of Chrome 149 (June 9, 2026), that lets a website declare structured tools and element purposes so AI agents can complete tasks without guessing at the interface. Think: telling an agent explicitly which form field is "first name" versus "full name," or exposing a "check order status" action directly. Google is positioning it for any agentic browser, not just Chrome.
It's early (origin trial, not GA), but the direction is unambiguous: agent-readiness is extending beyond structured data into on-site task completion. Sites that declare their tools will be operable by agents; sites that don't will be operated by screen-scraping guesswork, badly.
Security is part of the spec, not an afterthought. On June 11, 2026, two days into the origin trial, Chrome published security guidance warning that agents operating WebMCP tools can be hijacked two ways: malicious instructions hidden in tool manifests, and contaminated tool outputs (user-generated content like reviews carrying indirect prompt injection). Chrome's position is that model-level safety can't reliably stop these attacks, so defenses must be architectural. For a store exposing tools, that means: mark untrusted content with untrustedContentHint, declare read-only tools with readOnlyHint, and restrict tool access to trusted origins via exposedTo. Tools that return customer reviews or Q&A content are the highest-risk surface; treat their output as untrusted by default.
How do you get a store agent-ready?
- Feed and identifier completeness. GTIN, brand, and MPN on every item: identifiers are how agents match products into carts. Empty optional attributes are queries you silently don't match.
- Three-way consistency. Feed, on-page copy, and schema must agree on price and availability. Contradictions don't just cost Shopping eligibility, they cost agent confidence, and agents don't give second chances.
- Complete Product schema on every product page, kept current. Structured data is the operating layer agents read first.
- Merchant Center hygiene. The 2026 requirements stack: revised ToS June 15, product video via the video link attribute serving June 30, 500×500 minimum images (warnings now, enforced January 31, 2027). Diagnostics are uptime monitoring now.
- Watch UCP. Universal Cart participation runs through it. If you're on BigCommerce or another major platform, expect platform-level support, but verify rather than assume.
The machine-readable layer (schema, feeds, identifiers) is the same foundation our technical SEO work builds for AI citations, which is why we treat them as one program. For the deeper integration work (UCP readiness, agent-facing tooling, custom storefront APIs), that's BigCommerce development and custom AI app territory: the same engineering, pointed at the buyer's agent instead of the buyer.