New AI-shopping protocols promise to work together like Lego bricks — but are merchants actually snapping them together yet?

This asks whether merchants are actually combining the new agent-shopping protocols into one working stack, or whether the 'stack' is still mostly an idea. The short answer: the corpus has a strong argument for why merchants *should* stack them, but very little evidence that merchants *are* doing it. The case for stacking is clearest in Do agentic commerce protocols compete or complement each other?. It describes the protocols as rungs on a ladder rather than rivals: one helps agents find products, one handles checkout, one governs payment authorization, and one proves who the agent is. Each rung takes on a different level of fraud risk. That's a design argument, not an adoption count.

The parts that look most real are the individual layers, each built by a large player with partners. Google's AP2 chains signed 'mandates' (a record of what the user intended, then a record of what's in the cart) into an audit trail. More than 60 payment firms helped build it How can signed mandates prevent payment fraud by agents?. Visa's Trusted Agent Protocol, built with Cloudflare, lets a merchant check a cryptographic signature to tell a trusted agent from an unknown bot Can cryptographic signatures help merchants identify trusted AI agents?. McKinsey's view is that signed mandates like these will have to replace the human signals that fraud systems used to depend on Does agent-initiated commerce break existing payment infrastructure?. But 'partners signed on to build it' is not the same as 'merchants run it in production.'

The clearest evidence on what merchants actually do points the other way. When Amazon blocked Meta's Muse shopping agent, it did so through terms-of-service enforcement, not by negotiating through any identity protocol. That suggests the identity layer of the stack isn't working in practice yet What identity standards do shopping agents actually need?. The same analysis makes a subtler point: secure credential vaults protect the shopper but don't tell the merchant who's knocking. A layer that looks finished from the consumer side can still be missing from the merchant side Can credential vaulting alone solve merchant-facing agent identification?.

Two lateral threads help explain why the stack may come together slowly or unevenly. In a nearby area of the corpus, teams building production AI agents often drop protocol layers like MCP for direct function calls because the protocols added unpredictability. 85% of surveyed practitioners build custom agents Why do protocol-based tool integrations fail in production workflows?. If developers do that inside their own systems, merchants may also pick single layers that solve their immediate fraud or checkout problem instead of adopting the whole ladder. Meanwhile, coordination standards tend to spread by wrapping existing systems rather than replacing them Should coordination protocols wrap existing systems or replace them?. Payments seem to follow the same pattern: agents are expected to use familiar virtual cards before newer stablecoin options Will agents adopt cards before stablecoins for payments?.

The question you may not have known to ask is *who* decides whether there's one stack at all. Waller expects agentic commerce to split into open and closed systems, the way payment networks did Will agentic commerce platforms adopt open or closed models?. O'Reilly argues that protocols shape markets rather than sitting neutrally underneath them Do protocols reshape markets by enabling multi-vendor competition?. If that's right, a merchant's 'integrated stack' may end up being whichever bundle its dominant platform hands it. A large retailer like Amazon may choose to build none of it. The corpus doesn't yet have merchant deployment data that would settle this, so treat any claim of integrated adoption as unproven for now.