Module 3 opened with the Deutsche Bundesbank's conditional: if agents can switch between providers, protocols and rails with almost zero friction, the traditional lock-in effects in payments and the price-setting power of established intermediaries lose their significance. Module 3 asked who steers the switch. This lesson asks what the switch does to the people being switched away from, and who is building the layers where the switching happens.
Lock-in was the business model
The Bundesbank's framing is economic before it is technical. By largely automating payment-related decisions, agentic payments could significantly reduce coordination costs: the search, communication, negotiation and monitoring costs a footnote traces back to Williamson. Much of the value an intermediary captures today is the coordination cost it saves the parties. Take that down and the margin has to come from somewhere else.
The report says where the pressure lands. In combination with instant payments, agents could identify cheaper and faster rails and systematically prioritize cost-efficient means of payment over means with wider margins. Existing sources of revenue and business models would come under pressure. Agentic payments could tend to avoid existing, currently profitable payment networks, because those networks have higher costs than alternatives such as instant payments. Cross-border is singled out, because routing is where the biggest fee differences are.
The structural prediction follows: barriers to entry fall, market coordination becomes highly granular, and payment intermediaries shift from gatekeepers to interoperable infrastructure providers. That last phrase is the one for an executive to sit with. Gatekeepers charge for access. Infrastructure providers charge for throughput. The margins are not the same.
An agent that is indifferent to your brand and sensitive to your price is the customer every intermediary has spent 30 years making sure it does not have.
Module 5 added the customer's side of the same trade: the cheaper rails are the ones without a route back. The margin argument and the dispute-right argument are two views of one optimizer. I wrote the pricing view up in Agentic Commerce Will Reprice Payments, and the grocery aisle, where pay-by-bank already competes with interchange for a human customer, is the live version in The Interchange War Has a Front Line.
The fight is for the interface
If the rail is commoditized, where is the margin? The Bundesbank's answer is the customer interface. Control over interfaces with end customers could become a fiercely competitive field, it says, and many entities are developing solutions to tap the market potential its footnote attributes to the consultancies' estimates.
It gives its own view of the starting positions. Established payment service providers have technical, regulatory and organizational experience and are, in the Bundesbank's words, in a good starting position. AI model providers, technology firms and cloud and infrastructure providers compete by supplying decision logic, platforms, interfaces or scalable infrastructure. And partnerships between the technologically leading companies and the traditional providers are increasing, aimed at integrating advanced solutions into existing systems.
Module 3's case study is the evidence that owning the model does not settle the interface fight. OpenAI's pivot away from the first Instant Checkout left discovery with the model provider and checkout with the merchant (TechCrunch), and UCP was designed so that the business remains the merchant of record (Google). The interface that matters for payments turned out to be the merchant's, not the chat window. That is one data point from the first wave. The Bundesbank describes a contest, not a result.
Four groups, four interests
The Bundesbank describes four kinds of actor and takes care to say their interests differ. Module 2 introduced three of them at the communication layer. Here they are across the whole architecture.
Two of the Bundesbank's observations deserve their own paragraph.
The upstream control instance. Agent architecture and authentication logic determine which actions an agent is allowed to perform before the protocol is actually used. Whoever runs the framework decides what the agent can even attempt. Module 2 called that the gatekeeper. Here it is the reason the technology group can hold central roles without a payments license: by the time a payment protocol is invoked, the important decisions have been made.
The regional players. Alongside the global networks, the Bundesbank notes regional players building their own solutions, and its footnote names the National Payments Corporation of India and the Chinese platform companies. Existing user relationships and integrated trading platforms let them embed agentic payments into daily digital life quickly. The cost is fragmentation: regional or platform-specific solutions that do not interoperate. That is the Bundesbank's sovereignty concern in reverse. Other jurisdictions are solving it by building their own stack.
Concentration, and the data behind it
The Bundesbank's worry is not that one group wins. It is that a small number of protocols win, whoever runs them. If a small number of central protocols are dominant, market structures could become more concentrated. A small number of technology providers could control core functionalities from interpreting user intentions to initiating and executing payments. Providers and users could become equally dependent on external systems.
The mechanism of influence is the one Module 3 covered: agentic payment providers could indirectly influence the choice of merchant or means of payment through the models used, and limited auditability makes that hard to see. The Bundesbank adds the input side here. Access to data will become a key competitive factor, and asymmetric availability of data may lead to market distortions. Closed systems like SWIFT are currently the dominant means of communication in payments, so open interfaces are not fully in place, and integrating agents requires more openness and harmonization.
There is a quieter line for anyone selling to merchants. Making offerings accessible to agents, through machine-readable catalogs and agent-compatible interfaces, could benefit particularly large retailers and online marketplace providers, while small retailers would have to overcome significant technical and operational challenges. The readiness scan in Module 2 is a weekly measurement of that gap, and the 92 storefronts it covers are the large ones.
The cost nobody has priced yet
One passage in the Bundesbank's conclusion cuts against the whole market-structure section. Adoption depends not only on technical performance, user acceptance and regulatory treatment, but also on how the costs of AI services evolve. If costs rise significantly, agentic payments are initially likely to be used only for complex cases or where the value added is particularly high, and not everything technically feasible would be implemented.
Read that against the margin argument. The agent that routes around a card network to save a percentage point is itself running on inference someone is paying for. If that cost rises, the rail arbitrage stops being worth it for a €40 grocery order. The prediction that agents will prioritize cost-efficient means of payment assumes the agent is itself cost-efficient. For the retail cases the article is most optimistic about, that is open, and the footnote cites the business press on rising AI bills rather than any settled figure.
What position to hold
The Bundesbank describes the field; it does not advise. I will go one step further, because the course has given enough to say what the defensible positions are.
Be the rail the agent prefers. The position the instant payment systems hold by regulation. A throughput business with thin margins and no dispute right, and the Bundesbank expects it to carry more volume.
Be the interface the merchant keeps. The position UCP's design and the Instant Checkout pivot point to. The merchant of record holds the customer, the order record and the return path. A provider that gives the merchant the agent-facing surface without taking the relationship is selling the thing the Bundesbank says is most contested.
Be the verifier at step 6. The position the burden-of-proof prediction creates. Someone has to verify the anchor, test the mandate and keep the record a dispute will need. The party that does it reliably is the party agents are routed to, because the others will decline to execute.
Be the registry. Module 4 ended on the observation that every anchor relies on a registry of identifiers, keys or tokens, and whoever runs the registry decides which agents exist. The Bundesbank's cloud-provider paragraph says the same from the other side: verifying agent identities at the protocol level, like domain certification, is particularly attractive. Very few parties can take that position, and the Bundesbank's sovereignty concern is largely about who they are.
Each of those is a role in the architecture rather than a product. Europe's version of the dependency problem, banks building agents on infrastructure they do not control, is in Europe's banks are building agents on borrowed infrastructure. The last lesson takes the central bank's view of the same architecture.
Next lesson: under what conditions. The Bundesbank's three roles for central banks, the sovereignty question, the rule it wants applied, and the course's closing question. Who wrote the conditions your agent pays under?
Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.