In 2012, Jamie Dimon told JPMorgan’s board that mobile banking wasn’t a feature — it was a channel. At the time, fewer than 20% of retail banking customers used a mobile app. By 2020, mobile accounted for more than 75% of all consumer banking interactions at major US banks. The institutions that treated mobile as a strategic channel — not a nice-to-have — captured outsized market share. The ones that waited spent years playing catch-up, losing customers steadily to digital-first competitors. We are now standing at the beginning of a shift that is at least as significant, and probably faster: AI assistants are becoming the next primary banking channel.

The channel shift is already happening

Here is a fact that most bank executives have not yet internalized: your customers are already asking ChatGPT about their money. They are asking it to help them budget. They are asking it to explain fees. They are asking it to compare savings rates across institutions. And right now, the bank has no presence in that conversation.

This is not hypothetical. OpenAI’s ChatGPT surpassed 200 million weekly active users. Anthropic’s Claude, Google’s Gemini, and a growing ecosystem of specialized AI agents are collectively training hundreds of millions of consumers to interact with information — and increasingly, with services — through natural language. A growing share of consumer financial interactions could start with an AI assistant rather than a bank’s own app or website — some industry observers project this could exceed 50% by the end of the decade.

If this sounds like the kind of breathless prediction that has preceded every overhyped technology trend of the past decade, that skepticism is warranted. But there is a structural difference this time: AI assistants are not asking consumers to adopt a new device or learn a new behavior. They are meeting consumers inside a tool they already use every day.

The parallel to mobile banking is instructive, but the timeline is compressed. Mobile took roughly a decade to go from early adoption to dominance. AI assistants are on a trajectory measured in years, not decades. The infrastructure is already in place. The user behavior is already forming. The only question is whether banks will meet their customers where they are, or cede that surface to someone else.

Why this is different from “just another app”

When banks think about new digital channels, the instinct is to think about building something — a new app, a new portal, a new widget. That instinct is wrong here. AI assistants are not a destination that banks build. They are an environment that customers already inhabit. The strategic move is not to build another app. It is to expose your existing services through the surfaces your customers are already using.

Consider what a customer actually wants to do with their bank on any given day:

  • “What was my spending on restaurants last month?”
  • “Dispute this charge from Uber.”
  • “Move $5,000 from savings to checking.”
  • “Am I on track for my savings goal?”
  • “Show me my last three mortgage payments.”

Every one of these is a natural-language request. Every one of these is something a customer could ask ChatGPT right now — if their bank had the infrastructure to securely receive and execute it. The technology exists today — what is missing is the bridge between a bank’s existing APIs and the AI assistants customers are already using.

This is a critical distinction. Banks have already invested heavily in the underlying digital infrastructure — APIs for balance checks, payment initiation, dispute management, account transfers. The servicing layer is built. What is missing is the bridge that makes those capabilities accessible through the AI assistants customers are already using.

The next wave of banking interactions won’t come from humans

There is a second dimension to this shift that most banks have not yet internalized. The next wave of entities interacting with banking infrastructure will not be humans typing on phones. They will be AI agents acting autonomously on behalf of humans.

Think about what is already emerging: AI agents that manage personal finances, that automatically optimize savings allocations, that negotiate bills, that execute recurring financial tasks on a schedule. These agents do not log into mobile apps. They do not navigate web portals. They call APIs. And they expect structured, authenticated, compliant access to do so.

Banks that only think about human customers using AI assistants are already thinking too small. The real scale of this channel includes every AI agent that will eventually need to interact with financial services on a customer’s behalf. That is a fundamentally larger addressable surface than any previous banking channel.

The window is shorter than mobile

Here is the uncomfortable truth about channel transitions in banking: the early movers do not just get a head start. They get compounding advantages that become nearly impossible to overcome.

When Chase launched its mobile app early and invested aggressively in mobile-first features, it didn’t just attract mobile-savvy customers. It trained an entire generation to associate “banking on my phone” with “banking with Chase.” That association became a moat. The banks that launched mobile apps three or four years later were technically competitive but psychologically secondary — customers had already formed their habits.

The same dynamic will play out with AI assistants, but faster. Here is why:

  1. AI adoption curves are steeper. Mobile required new hardware (smartphones) and new behaviors. AI assistants layer onto devices people already own and behaviors they are already developing. The adoption friction is dramatically lower.
  2. Customer switching costs increase with AI. Once a customer sets up their bank inside ChatGPT and starts relying on that interaction model, the cost of switching banks increases. The AI assistant becomes a sticky distribution layer.
  3. AI-native customers are disproportionately valuable. Early adopters of AI tools tend to be higher-income, more digitally engaged, and more likely to hold multiple financial products. Losing them is not losing average customers — it is losing your best customers.

Banks have a window measured in quarters, not years, to establish presence in the AI assistant channel. The institutions that move now will be the ones that AI-native customers associate with AI-accessible banking. The ones that wait will find themselves competing on rate alone — the most commoditized and least defensible position in financial services.

What does “being in the channel” actually mean?

Being present in the AI assistant channel requires solving several interconnected problems simultaneously:

  • Authentication: How do you verify that an AI agent is acting on behalf of a legitimate, authorized customer? This is not the same as a login flow. It requires a fundamentally different security model.
  • Authorization: How do you control what actions an AI agent can take? Checking a balance is different from initiating a wire transfer. The permissioning has to be granular and configurable.
  • Compliance: Every interaction through an AI assistant needs to be logged, auditable, and compliant with the same regulatory framework that governs every other customer interaction. OCC, CFPB, FCA — and other regulators do not care whether the customer used your app or ChatGPT.
  • Business logic: If a customer asks a question that represents a cross-sell opportunity — “Do you offer a higher-yield savings account?” — the bank needs to be able to respond intelligently within the conversation, not redirect the customer to a different channel.

This is exactly the problem we created MidLyr to solve — purpose-built middleware that connects existing bank APIs to AI assistants with authentication, compliance, and business logic built in, deployable in weeks rather than years.

What banking looks like when every interaction starts with “Hey ChatGPT…”

Picture this: it is 2028. A customer wakes up and asks their AI assistant to summarize last week’s spending. The assistant pulls transaction data from the bank, categorizes it, flags an unusual charge, and asks whether the customer wants to dispute it. The customer says yes. The dispute is filed. The assistant then mentions that based on the customer’s spending patterns, a different credit card from the same bank would save them $400 a year in fees. The customer says “tell me more.” The conversation continues.

In that interaction, the bank never opened an app. The customer never navigated a menu. There was no hold time, no chat widget, no branch visit. But the bank was present in every moment of the interaction — authenticating, executing, advising, and earning.

That is not science fiction. The technology to deliver that experience exists today. The question is not whether banking will move to AI assistants. The question is which banks will be there when their customers arrive.

Early movers are already seeing this play out. Banks that have connected their existing APIs to AI assistants are finding that the experience transforms not just the channel, but the nature of customer service itself.

“What we realized after talking to MidLyr is that a lot of what we considered independent servicing issues can be addressed naturally through the LLM interaction with our backend systems.”

— COO, top 50 U.S. Regional Bank (name withheld by request)

The mobile banking revolution rewarded the institutions that recognized a channel shift early and moved decisively. The AI assistant revolution will do the same — but the window is shorter, the stakes are higher, and the customers who care most are the ones you can least afford to lose.

MidLyr is AI-native banking infrastructure that connects your bank’s existing APIs to ChatGPT, Claude, and every AI assistant — with bank-grade authentication, compliance guardrails, and cross-sell tooling built in. MidLyr powers the customer intraction, as well as the internal operations to support these interactions with dedicated modules for dispute handling, fraud investigations and more. Integration takes 2-6 weeks for banks with existing API infrastructure. Book a demo to see what your bank looks like inside ChatGPT.