The Agentic Commerce Report

Network-level support for agent-initiated payments — Visa Intelligent Commerce, Mastercard Agent Pay, Stripe Agent Toolkit. None of the three rails is currently interoperable; each handles agent identity, token scoping, and dispute liability differently.

Agent and LLM discovery of products: ChatGPT shopping, Google AI Overviews carousels, Shopify's agent-readable product feed format. Convergence on a single feed standard is unsettled — Schema.org has competing drafts under dispute.

Open specs that define how agents interoperate: Anthropic's Model Context Protocol, Google's Agent-to-Agent protocol, the in-progress Agent Commerce Protocol working group. Movement here lags movement at networks by 6–9 months.

Cryptographic agent identity, delegation chains, and revocation. The W3C published a working draft of Verifiable Credentials for autonomous agents in week 17; Auth0 shipped agent delegation flows earlier in the year.

Prompt injection in product descriptions, agent-purchase fraud, OWASP Top 10 work for agentic systems. Wiz Research demonstrated three live attacks on retail agents in week 17.

EU AI Act enforcement, CFPB notices on agent purchases, the first US class action over agent-initiated unauthorised purchases (filed in California, week 16). Regulatory cadence is now monthly rather than quarterly.

Live deployments at retailers — Walmart's agent-shopping pilot, Mastercard's confirmed three EU pilots, Shopify's agent-checkout integration on Plus plans. Conversion lift on agent-driven sessions is reportedly real, but dispute volume runs higher than expected.