When the Bank Becomes Infrastructure: What the IFZ Finding Reveals About Agentic Commerce
On June 1, Andreas Dietrich described how ChatGPT is becoming a financial companion and how banks risk being reduced to mere settlement providers. This is not a banking story. It is the same movement publishers have lived through for two years — third industry, same pattern. And the layer deciding what gets recommended is unoccupied in all three cases.
On June 1, Andreas Dietrich published an analysis on the IFZ Retail Banking Blog that was read across Swiss banking and almost nowhere else. That is a shame, because it does not primarily describe a banking problem. It describes, for the third time in eighteen months, the same movement — this time in an industry that assumed it was exempt. Anyone who watched what happened to publishers in 2024 and 2025 reads it with an uncomfortable sense of déjà vu. What actually happened On May 15, OpenAI rolled out a Personal Finance Manager inside ChatGPT in the US. Users connect bank accounts, credit cards, and investment accounts directly to the chat. The assistant sees balances, transaction histories, portfolios, recurring payments, and liabilities, stores "financial memories" for follow-up conversations, and answers questions in natural language. The numbers Dietrich cites: Metric Value Swiss residents using AI chatbots regularly 60 % People asking ChatGPT financial questions monthly 200M US financial institutions connectable via Plaid 12,000+ The assistant cannot execute transactions yet. That is the decisive half-sentence — and the announced next step. The number nobody quotes The 200 million is the headline. The third number is the one that matters. Plaid is not the product. Plaid is the rail. It is the data interface through which 12,000 institutions are connected to an assistant that none of them controls. Not one of those 12,000 negotiated how its product is presented in conversation, in what order alternatives appear, or whether its name comes up at all. Anyone working at Nexbid recognises this construction. We have seen it twice already, under different names. The same movement, three times This is our reading, not Dietrich's — he writes about banking, not about agentic commerce. But the structure is identical in all three cases: Publishers (2024–2026) Retail (2025–2026) Banking (now) The asset holder Publisher with content Merchant with products Bank with customer data What they lose The click The checkout The customer relationship The rail Crawlers, AI Overviews UCP, ACP Plaid Who controls it US platforms US coalition US platform What remains Content supplier Fulfilment Settlement In all three cases the asset holder owns the thing that matters — and loses the relationship anyway. Not because the asset lost value, but because a layer inserted itself in front of it that decides when and how it surfaces at all. Dietrich's own concern captures it precisely: that banks are …
- Author
- Holger von Ellerts
- Published
- 2026-07-17
- Topics
- Agentic, Protocol Commerce, Research