Should one company control the rails for AI shopping agents, or should the system be open for any agent to use?

This explores what changes when agentic commerce runs on rules owned by one platform versus open standards any agent can use. The clearest framing in the collection comes from payments history. Card networks split into open-loop systems like Visa, where many banks take part, and closed-loop systems like early American Express, where one company controls both sides. Agent commerce looks set to split the same way. Some platforms will let any agent shop. Others will only work with their own agent Will agentic commerce platforms adopt open or closed models?. That choice matters for more than engineering. It decides who controls the customer, how concentrated the market becomes and which standards win.

The Amazon–Meta Muse dispute shows what happens when no shared standard exists. Amazon blocked Meta's shopping agent, and nothing like a negotiation took place. Retailers enforce access through terms-of-service bans because agents have no standard way to say who they are or prove that a shopper authorized them What identity standards do shopping agents actually need?. Without an interoperable identity layer, each platform makes its own rules, and access depends on whoever owns the storefront. The UK's Competition and Markets Authority (CMA) pushes this argument further. It treats data portability as basic market infrastructure, not a convenience. If your purchase history and preferences stay locked inside one ecosystem, an outside agent can't act well for you, and switching to another agent becomes too costly to bother Should data portability be infrastructure for agent-driven markets?.

The interoperable side turns out not to be one standard competing with platform standards. It is a stack of protocols, each covering a different job. Boboev maps the new protocols from OpenAI, Google, Visa and Mastercard (ACP, AP2, UCP, TAP and Web Bot Auth) onto separate layers: running the transaction, governing what the agent may do, product discovery, and agent identity. Merchants are meant to combine them rather than pick one, and each layer handles a different level of fraud risk Do agentic commerce protocols compete or complement each other?. A lesson from wider agent research supports this. Coordination standards spread faster when they wrap and connect existing protocols than when they try to replace them, because adopters get value step by step without rewriting everything Should coordination protocols wrap existing systems or replace them?.

The less obvious point is that open, protocol-based commerce brings its own problems. EU consumer law protects shoppers partly through screen design: required "buy" buttons, mandatory disclosures and clear withdrawal steps. When agents transact with each other through APIs, there is no screen, so those protections stop working. Busch proposes rebuilding them inside the protocols themselves, for example with a standard withdrawal API Can EU consumer law rules survive agent-to-agent API transactions?. On the engineering side, teams running agents in production have found that tool access through a general protocol (MCP) caused unpredictable failures, which went away when they switched to direct, custom function calls Why do protocol-based tool integrations fail in production workflows?. That is a quiet argument for why closed, platform-specific integrations stay attractive: they are more predictable.

Taken together, the trade-off looks like this. Platform-specific standards give tighter control and more reliable behavior, at the cost of lock-in and access decided by whoever owns the storefront. Interoperable standards offer competition and the freedom to switch, but only if identity, data portability and consumer protections are written into the protocols too. The collection doesn't yet show which model is winning in practice. It mostly offers arguments and frameworks rather than evidence from real markets.