My short answer: The interface may change, but I don’t think banking apps will disappear. Announcements from payment networks since April 2025 show that the groundwork for agents that can transact is being laid. They do not show that all of banking is moving to agents.
This is the long version of the first issue of AFTER MONEY. I will keep evidence, interpretation and forecast clearly apart.
From apps to intent
Doing something in a banking app today means navigating menus: the right tab, the right form, the right confirmation screen. The idea of an agent reverses this. The user describes a goal; the agent carries out the steps with the tools it is allowed to use.
I think of it as a three-step flow: intent, agent, approval. The user says what they want, the agent prepares and executes, and at the critical point the user still has the final say. To be clear from the start, this is a scenario. I am not presenting data showing that this kind of use is widespread.
- 01IntentThe user says what they want.
- 02AgentPrepares and executes with permitted tools.
- 03ApprovalAt the critical step, the user has the final say.
A chatbot is not an agent
This is where the debate gets muddled most often. A chatbot answers a question: “How much did I spend on groceries this month?” An agent can act, using tools within defined permissions: it prepares a payment, sends it and records the result.
That is how I use the word agent in this piece. It is a working definition, not a universal technical classification. But it has an important consequence: Not every chat interface is an agent. Adding a chat window to an app does not turn it into one.
Payments are moving ahead
The first concrete steps toward agents that transact came from payment networks. Three announcements, newest first:
- 10 Jun 2026
Mastercard Agent Pay for Machines ↗
High-frequency, very low-value payments between agents and machines. Settlement across cards, accounts and stablecoins; permission rules and spending limits.
- 30 Apr 2025
Visa Intelligent Commerce ↗
Agents that find and buy products on a user’s behalf. Consumers set spending limits and conditions; tokenized digital credentials replace card details.
- 29 Apr 2025
Mastercard Agent Pay ↗
Agents that make purchases on a user’s behalf. Consumers set limits on what an agent can buy; agents must be registered and verified before they transact.
Two caveats. First, these are company announcements. They do not tell us how widely the products are used, where they are available or how they perform under independent review; I make no claim about availability in Turkey. Second, Agent Pay for Machines focuses on machine-to-machine payments rather than consumers, so its link to banking apps is indirect.
Payments are not all of banking
There is a long way between “payments are moving” and “banking apps are over.” A banking app is more than a payment screen; it also carries lending, account management and transaction disputes.
None of the announcements above shows those areas moving to agents. That needs separate evidence, and this piece does not claim it.
Control is part of the product
If an agent can act, the real product question is: Who is in control, and how? A December 2024 report by the Financial Stability Institute of the Bank for International Settlements (BIS) offers a useful frame.
According to the report, AI can amplify existing risks such as model risk and data privacy, while generative AI brings risks of its own, such as hallucination and anthropomorphism. It finds that most financial authorities have not written separate AI rules for financial institutions, since existing frameworks largely cover these risks. It lists governance, model risk management, data governance and third-party AI providers among the areas that need further regulatory attention. The views in the report are the authors’ own.
Building on that frame, my design proposal has four parts. These are product design proposals, not legal requirements:
- Explicit permissions: What the agent can and cannot do should be visible at a glance.
- Transaction limits: Limits on amount, frequency and recipient should stay with the user.
- Approval at critical steps: For transactions that are hard to reverse, the user keeps the final say.
- Reviewable records: What the agent did, and why, should be traceable afterwards.
One detail stands out: Both Visa and Mastercard put the power to set limits in the consumer’s hands. Control is central to how these products are described, too.
My scenario: coexistence
Here I move from interpretation to forecast, and I want to flag that clearly. In my scenario, the agent coordinates everyday, repetitive tasks: bills, recurring transfers, budget reminders. The app remains the place to verify, review and intervene.
In this scenario the app does not disappear. It turns from a storefront into a control panel where you can see what the agent did and stop it when needed.
There are ways this could prove wrong. Users may not trust agents enough, and apps may stay much as they are today. Or approval steps may become so simple that the app recedes even further into the background. No one can say with data today which of these will happen.
Sources
- Visa — Find and Buy with AI: Visa Intelligent Commerce ↗30 Apr 2025
- Mastercard — Mastercard unveils Agent Pay ↗29 Apr 2025
- Mastercard — Agent Pay for Machines ↗10 Jun 2026
- BIS — FSI Insights 63: Regulating AI in the financial sector ↗12 Dec 2024
Sources checked on 29 October 2026.