If one company controls your data, your shopping AI, and the payment that completes the sale, what actually keeps it in check?

This explores what could stop one company from controlling your data, the AI assistant you buy through, and the payment that completes the sale, all at once. The concern is concrete. Is Google building a chokepoint across data, chatbots, and payments? argues Google is doing exactly this. It is applying its old search-advertising chokepoint strategy to agents by combining personal data, chatbot distribution, and deals with retailers and payment networks. The corpus offers three kinds of counterweight. It is also frank that each one is partial.

The first is making data portable. The UK competition regulator argues that data portability is not a consumer convenience but basic market infrastructure (Should data portability be infrastructure for agent-driven markets?). If your preferences and purchase history are locked inside one ecosystem, a rival agent can't act well for you, and switching agents means starting from zero. Portability attacks the data leg of the triad directly. The idea extends to consumer rights. Can EU consumer law rules survive agent-to-agent API transactions? points out that EU protections like mandatory 'buy' buttons and disclosure screens disappear when agents transact machine-to-machine. Busch proposes writing rights into the protocol layer itself, such as a required 'withdrawal API', so protection doesn't depend on whoever owns the interface.

The second is splitting the stack into separate layers. Do protocols reshape markets by enabling multi-vendor competition? argues that protocols are not neutral plumbing. A shared standard like MCP can turn a winner-take-all race into a web where many vendors can plug in. Do agentic commerce protocols compete or complement each other? describes the current agent-commerce protocols as separate layers: discovery, checkout, governance and identity. Merchants stack them rather than pick one. The IMF makes a similar point from the payments side. It argues that the probabilistic AI layer that decides what to buy should be kept apart from the deterministic layers that authorize and settle payment (Should AI agents make payment decisions independently?). Their reason is reliability, not competition, but separated layers are also harder for one firm to fuse. The catch: the protocols that are supposed to open the market are written by the would-be gatekeepers. Google's AP2 was built with more than 60 payment firms (How can signed mandates prevent payment fraud by agents?), and Stripe and OpenAI co-wrote ACP (Can payment tokens let AI agents checkout safely?). A protocol can be either a lock or a key, depending on who controls how it changes.

The third is diversity of agents, a check that's easy to miss. In pricing experiments, identical LLM agents quietly settled into prices 22% above competitive levels. Mixing in agents with different time horizons, data access or underlying algorithms weakened or broke that collusion (Does agent heterogeneity prevent LLM pricing collusion?). The CMA warns that the opposite case, many businesses running similar pricing agents in a concentrated market, could reduce competition without any explicit agreement (Could autonomous pricing agents dampen competition without explicit collusion?). So keeping many different agents and models in the market isn't just a fairness goal. It may directly protect competition.

One thing works against all three. The hardest problems in agent commerce are trust problems: proving an agent has authority, deciding who pays when it goes wrong, and catching new kinds of fraud (What blocks autonomous agents from handling purchases at scale?). Big integrated players are well placed to solve them, which pulls the market toward concentration. Convenience pulls the same way: a single 'allow always' setting can quietly let an agent buy and share your address without asking (Do broad consent settings pose bigger risks than prompt injection attacks?). These retrieved notes don't directly cover classic antitrust remedies such as breakups or bans on owning several layers at once. The defenses they describe are portability, separated layers and agent diversity, not structural separation by law.