What's Actually Live
Three protocols are emerging as the plumbing for this: UCP (Google and Shopify, covering the full shopping journey), ACP (OpenAI and Stripe, focused on checkout), and MCP (Anthropic, giving models real-time data access). Consumer-facing agents already shopping on a user's behalf include Perplexity's Comet browser, Amazon's Rufus, and Google AI Mode shopping. Notably, OpenAI deprecated its Instant Checkout feature in March 2026 — the current ACP model is product discovery plus a merchant-site redirect, not an agent silently completing checkout end-to-end on arbitrary sites.
The Real Scale So Far
McKinsey forecasts $900 billion to $1 trillion in US retail revenue moving through agentic commerce by 2030. Morgan Stanley's AlphaWise survey put LLM adoption near 50% in the US, with AI agents already capturing an estimated 10-20% of ecommerce activity — early, but not hypothetical.
What This Actually Requires From a Storefront
An agent evaluating a product can't infer price, availability or return policy from layout the way a human shopper can — it needs that data structured and explicit: clean product schema, accurate real-time stock data, and policies stated in machine-readable form, not just a PDF linked in the footer. This is the same integration depth already required for solid PIM and marketplace API work under Ecommerce Platforms — agentic commerce raises the stakes on it rather than introducing a new discipline.
Why This Isn't Optional for Long
As agent-driven discovery captures a growing share of purchase research, a catalog that's only readable by human eyes — JS-rendered pricing, image-only stock status — becomes invisible to an increasing slice of the funnel, the same way a page invisible to AI crawlers is invisible to GEO. The businesses affected first will be the ones with the least structured product data, not the smallest ones.