Who owns the AI agent? It sounds like a technology question. It is a relationship question. In Episode 2 of TreasurUp Talks, host Joost Kevelam asked Marien van Baren, TreasurUp’s co-founder and CTO, why it matters so much that banks own the agent. Marien’s answer fits in one line: whoever hosts the agent owns the relationship.
Put the AI in a TMS, and the TMS owns that slice
Marien’s example is the treasury management system. Put a company’s financial AI inside a TMS, and the TMS owns that slice of the company’s financials. The same logic runs through every channel a company works in: the ERP platform, the accounting package, the fintech app. The bank may still execute the payment and price the hedge underneath. But the questions, the preparation, and the decision move to wherever the agent lives.
The shift is already visible. Businesses are beginning to use AI assistants for financial decisions and banking tasks. The assistant they use becomes the front door. The bank becomes the back end. Fintechs did not win business clients by replacing banks; they took the daily workflows of payments, FX, and liquidity. An agent raises the stakes, because it does not take one workflow. It becomes the interface to all of them.
Who owns the AI agent when things go wrong?
Marien’s test for where the agent should live is not technical. “When things go wrong, who is accountable? Who owns the automation flow?”
In regulated banking, those questions cannot stay open. A hedge proposal, a cross-border payment, a liquidity sweep: each one carries financial and regulatory consequences. A supervisor will ask who validated the model, who set the limits, and who approved the action. The party that hosts the agent is the party that answers. If that party sits outside the bank, so does the answer.
The bank is the trust layer
“From our perspective, the bank is the trust layer,” Marien says in the clip. “That’s where the data is. That is where we have the controls and the accountability already live.”
Banks should use that position. Three things already sit inside the bank and nowhere else. The data: account flows, positions, limits, client history. The controls: mandates, entitlements, approval flows, audit trails. The accountability: a regulated institution that answers to its supervisor and to its clients.
This is the design point behind TreasurUp’s agentic business banking platform. Agents run inside the bank’s own channels, branded as the bank’s and governed by the bank’s policies. Every agent action with financial, regulatory, or accounting impact passes through an explicit human approval, logged with the agent’s reasoning, the input data, and the alternatives considered. Agents propose. The company user or the bank user disposes. Where the platform runs is the bank’s choice: managed SaaS, single-tenant in the bank’s cloud, or hybrid, with bring-your-own-LLM across all three.
Where the agent belongs
“So that is, from our perspective, where the agent belongs.”
McKinsey reached a similar conclusion for retail banking in April 2026: gen AI agents threaten to place themselves between banks and their customers, and banks that want to keep the relationship need the agent inside their own brand.1 In business banking, the workflows at stake are the ones fintechs are already capturing: payments, FX, and liquidity.
Who owns the AI agent turns out to be the same question as who owns the daily relationship. The question for every bank is not whether AI agents will handle daily business banking. It is whose brand the agent carries when it does.
Watch the full clip on our LinkedIn page. For the platform behind the argument, read Inside the agentic business banking platform and Agentic banking: why 2026 is the year banks must act. Curious where the agent should sit at your bank? Feel free to reach out.
