How Will Banking UX Change in the Age of Chatbots and Agents?

My short answer: Menus will not disappear, but the banking interface will split into three: conversation to express intent, screens to stay in control, and an interface machines can read for agents. The job of UX will shift from guiding users through menus to designing trust and control.

In the first issue of AFTER MONEY, I wrote that with agents the banking app could turn into a control panel. This piece is the design side of that scenario: How will screens, flows and websites change? As before, I will keep evidence, interpretation and forecast apart.

01ConversationTo express intent.
02ScreenTo stay in control and approve.
03Agent interfaceStructured information machines can read.

The three-layer banking interface I expect in this piece: a scenario, not an adoption claim.

From menus to intent

In 2023, UX researcher Jakob Nielsen called AI “the first new UI paradigm in 60 years.” In his account, we have worked with computers through commands for more than 60 years: The user tells the computer what to do, one step at a time. With AI, the user describes the outcome they want, and the system works out the steps. Nielsen calls this “intent-based outcome specification.”

Nielsen also lists the weak spots. Describing what you want in writing is a hard skill. Results are not easy to refine step by step. And when users do not know how a result was produced, correcting errors gets harder. That is why he expects future interfaces to be hybrids that combine intent-based design with today’s graphical interfaces.

For banking, the meaning seems clear to me: Conversation will not replace the screen; it will sit in front of it.

Lessons from the chatbot era

Banking went through the first wave of this shift with chatbots. In a June 2023 report, the US Consumer Financial Protection Bureau (CFPB) estimated that about 98 million people, roughly 37% of the US population, used a bank’s chatbot in 2022.

The same report lists the problems: wasted time, loops users cannot get out of, inaccurate information, disputes that go unrecognized or unresolved, and privacy and security risks. The CFPB’s sharpest criticism is aimed at loops with no way out to a human. According to the report, customers who cannot reach a live agent lose trust.

Klarna shows both sides. According to figures the company itself announced in February 2024, its AI assistant handled two-thirds of customer service chats in its first month, or 2.3 million conversations, and average resolution time fell from 11 minutes to under 2 minutes. In May 2025, Klarna’s CEO said the company was hiring human agents again so that customers could always reach a person; he acknowledged that cost had been too dominant a factor and had lowered service quality.

The lesson, as I see it: The automation rate is a success metric, but the moment that defines the UX is the moment a user needs to get out of the automation.

A new kind of user: the agent

With agents, banking interfaces gain a new kind of visitor: software acting on a user’s behalf. When Visa announced its Trusted Agent Protocol in October 2025, it said AI-driven traffic to US retail websites had grown sharply and that merchants struggled to tell trusted shopping agents from malicious bots. According to Visa, bot detection can mistakenly block legitimate agent purchases.

This raises a new design question for banks’ websites and apps, too. A product page will now have two readers: a person and the agent working for them. The agent needs information that is structured, explicit and verifiable: fees, limits, terms. The person needs to see what the agent read and what it did.

Six UX components that will change

These are product design proposals, not legal requirements. For the last two, EU rules point the way:

From conversation to a confirmation card: Users can describe their intent freely, but before anything is executed they should see a structured card with the amount, recipient, date and fees. What gets approved should be the card, not the sentence.
The permissions screen moves to the center: Which agent can do what, up to which limit and until when should be visible on one screen and revocable with one tap. Consumer-set limits also feature prominently in Visa’s and Mastercard’s agent announcements.
An activity log: The answer to “What did the agent do, and why?” should sit in a readable feed, with an undo option right there for actions that can be reversed.
Handoff to a human: The way to reach a person should be visible on every screen, and the context of the conversation should be passed to the agent who picks it up, so users do not have to start over.
Saying it is AI: Under Article 50 of the EU AI Act, AI systems that interact directly with people must be designed so that people are told they are interacting with an AI system, unless that is obvious from the context.
Accessibility and plain language: Since 28 June 2025, the European Accessibility Act has applied in the EU to banks’ websites and mobile apps, among other things. It requires information to be understandable and not to exceed level B2 of the Common European Framework of Reference for Languages. A conversational interface can help accessibility, but it should not be the only way in.

What does not change

Even as the interface changes, three user needs stay the same: understanding what happened, being able to fix it when something goes wrong, and reaching someone when needed. The hybrid Nielsen expects says the same thing: Intent is expressed in conversation, and control still happens on the screen.

My scenario: the home screen becomes a feed

From here on, this is a forecast. In my scenario, the banking app’s home screen turns from a grid of menus into a three-part structure: a field at the top to type or say what you want, a feed in the middle of what the agent has done and what is waiting for approval, and a hub at the bottom for managing permissions.

Websites also become two-layered: narrative for people, structured information for agents. The fee table on a product page turns into data an agent can read directly.

There are ways this could prove wrong. For routine tasks, menus may stay faster than conversation, and users may not let go of them. Regulators may limit what agents can do in banking. Or conversational interfaces may stay confined to customer service, as they largely did in the chatbot era. No one can say with data today which of these will happen.

Sources

Sources checked on 30 October 2026.

AUTHOR’S NOTEThe views in this piece are my own and do not represent my employer or any institution. It relies only on public sources. Not investment or legal advice.

← All posts