For AI agents to actually shop and pay on our behalf, which companies have to open their doors first?

This explores who has to adapt (payment networks, big platforms, merchants and the businesses behind them) before AI agents can shop and pay at scale. The corpus doesn't offer a neat list of companies. It describes a stack of layers, and each layer has its own gatekeepers. McKinsey's projection of up to $1 trillion in US agent-driven retail by 2030 assumes agents can coordinate, keep context and complete multistep purchases through protocols like MCP, A2A and AP2 Could agentic commerce reach one trillion dollars by 2030?. The question is who has to plug into those protocols.

The clearest answer is payment networks. Visa's Trusted Agent Protocol and Mastercard's Web Bot Auth sit at the identity layer. They let a merchant tell a legitimate shopping agent apart from a scraper or a fraudster. Commerce protocols such as ACP, AP2 and UCP cover checkout, permissions and product discovery. Boboev's point is that these don't compete. They form a ladder that merchants climb one rung at a time, and each rung absorbs a different level of fraud risk Do agentic commerce protocols compete or complement each other?. A parallel track is growing on crypto rails. On x402, AI agent wallets went from a 73-day median wait between funding and first spend to 18 minutes, and most of the volume runs through US wallets paying for model inference Are AI agent wallets becoming genuinely operational instead of experimental? Which countries dominate agent-to-agent AI payments on chain?.

The less obvious answer is that the biggest platforms don't need to become agent-compatible. They need to avoid making compatibility exclusive. Rothschild and colleagues argue that Apple, Google, Meta and Microsoft could recreate app-store gatekeeping by controlling the layers that let agents work together Will agent economies develop as open webs or closed gardens?. Waller makes a related comparison to card networks: some systems will be open and accept any agent, while closed ones will only work with their own Will agentic commerce platforms adopt open or closed models?. The UK's competition regulator (the CMA) adds another requirement: whoever holds your purchase history has to let it move. Without data portability, an agent can't act for you outside the ecosystem that holds your data, and switching stays locked Should data portability be infrastructure for agent-driven markets?.

Merchants and brands are in an awkward position. BCG expects $1.3 trillion of e-commerce spending to shift toward agents, with consumers and networks gaining the most and merchants coming out worse off Who wins and loses as AI agents reshape shopping?. Accenture finds that consumers say they would prefer agents built by retailers, but that evidence comes from surveys, not working systems Will brands succeed by building their own AI shopping agents?. For any company, becoming agent-compatible is more than adding an API. Deloitte finds that agent projects fail when they are bolted onto old workflows. They work when the data architecture and processes are redesigned around them Does agentic AI succeed through process redesign or workflow layering?.

The takeaway is that scaling probably won't come from one company winning a standards war. Coordination layers tend to get adopted by wrapping and bridging the protocols that already exist, not by replacing them Should coordination protocols wrap existing systems or replace them?. The open question is governance: whether the companies that control identity, data and distribution leave those bridges open.