Inheriting a huge fortune and Stark Industries when his father passed away, Tony Stark landed in the number four slot in Forbes Fictional 15 rich list; and while a fully functioning Iron Man suit hasn’t quite landed in reality yet, one element of Stark Industries’ success does resonate with the real world today: J.A.R.V.I.S.
The Age of Ultron instalment of the Marvel franchise was released in 2015, six years before the emergence of ChatGPT. However, now in 2025, it seems agentic AI is all anyone can talk about (stablecoins aside). Tony Stark told Bruce Banner in the film: “Started out, J.A.R.V.I.S. was just a natural language UI. Now he runs the Iron Legion. He runs more of the business than anyone besides Pepper.”
The Marvel films never really explore how J.A.R.V.I.S (aka Just A Rather Very Intelligent System) is capable of doing what it does, but as viewers, we can piece together that it is not through a bank app. Now, though, while there are some challenges down the road, a world in which every individual and business has their own “J.A.R.V.I.S” or equivalent AI agent to run their finances is in view.
Arguably, AI is the biggest revolution in financial services since the introduction of cheques and promissory notes in medieval Europe. While that innovation laid the groundwork for modern financial instruments, the developments of agentic AI flip the balance of power between customer and bank on its head.
From embedded finance to embedded agents
In order to evolve into this AI world, banks should not think of agentic AI as a purely technological challenge. Nor should they work towards some expectation that the same bank processes and systems that are currently in place would simply be more cost-effectively run by LLMs end-to-end.
After all, agentic AI from the customer end is on the same continuum as embedded finance. It puts more information and more power in their hands and allows them to use it on their own terms. Embedded finance puts the product at the point of need for the customer to make a decision. Embedded AI agents will identify that point of need independently and act on it, having been given some prescribed goals earlier.
One challenge for embedded finance is that consumers must trust that vendors have selected the best banking partner with the best products and services for them. Embedded agents will solve this by validating the partners and merchants or pointing to an alternative.
So what can banks do? Fully conversational embedded agents are not necessarily widely available for consumers, nor very capable yet. Therefore, experimenting is difficult. However, some banks are already beginning to transform their business models with embedded finance. They are learning to build from the outside in, not as they used to, from the inside out.
They are co-creating more specific solutions to verticals’ specific needs with the shared knowledge of vertical partners. They develop cross-functional expertise (think compliance and onboarding), working one or two steps removed from the customer, and they develop an understanding of the risks involved. They realise they must change to compete.
Near-perfect information is taking away banks’ knowledge advantage
Financial services is an industry built on information asymmetry. Admit it, banks know more than their customers. About their business, the economy, the likelihood of them paying their debt, and how they make money. That’s probably why on Penny Lane, only the banker in the corner has a motorcar.
Times are changing. The information required to understand your financial options and reach an informed decision is much more readily available and accessible. What started in 1999 with MoneySupermarket.com is being rocket-fueled by AI, providing users with abundant (admittedly, sometimes hallucinated) information. Forget flying exosuits, the future arrived as a clunky price-comparison site.
Alarm bells should be ringing
The change is not just in the availability of near-perfect information, though. AI is also changing how we browse the internet. Since Google launched its AI mode, users have stopped visiting the websites the information was gleaned.
The numbers tell the story. Healthcare sites’ traffic dropped 31 per cent in the year to June 2025, according to Similarweb, while science and education are down 10 per cent. Pew Research found that “Google users are more likely to end their browsing session entirely after visiting a search page with an AI summary than on pages without a summary.”
Moreover, ChatGPT can now browse the web in real-time, access current information, and is being integrated with other platforms to allow for tasks like booking travel or making payments within the chat environment.
In an interview with Peter Diamandis, ex-Google CEO Eric Schmidt said: “User interfaces are largely going to go away.” When the man who built the internet’s front door says doors are disappearing, it’s time to worry.
Are banks next? Banking executives should be wondering: if customers no longer come to your website, or to any other, for that matter, how do you sell to them? How do you ensure they even see your products?
Relevance is about taking a step back
Embedded finance builders like myself have long argued that distribution models are changing and that financial products need to be available at the point of need. That is where purchase decisions are made and where propensity to buy can be accurately assessed.
However, successful embedded finance use cases can only be seen once banks give up the ‘customer ownership’ they once had. To compete, they have to make their products and infrastructure available within a separate, non-financial user journey beyond their walled gardens.
Embedded finance is just the first step in collapsing the legacy process of purchasing financial products that is disconnected between the point of need and the point of purchase. There are already a number of successful examples in the wild of embedded finance being integrated fully. In those journeys, it is no longer necessary for consumers to visit a bank, physically or virtually, to complete a financial product purchase, and the number of successful transactions is only growing.
This is only the beginning. Embedded agents have the potential to further collapse the process into a single conversational flow. AI agents are going to do more: they’ll handle research, outreach, follow-ups, and purchase recommendations. In time, it is highly likely they will act independently on behalf of the customer to reach a given goal. Just as Marvel keeps finding new crossovers, banks have to accept they’re no longer the star but a part of the ensemble performing for the customer.
The illusion of progress
Banks are catching up on the wave, I hear you say. Not a day goes by without a new AI use case in banking revealing itself. This can be seen in CBInsights latest report, 100 Real-World Applications of GenAI Across Financial Services and Insurance. Yet, the cacophony of financial institutions’ AI-related announcements seem to be missing the point.
Another efficiency point here, another employee chatbot there… Meanwhile, distribution itself is evaporating. If customers never step foot in your branch or log in to your app, what exactly are you optimising for?
Your customers are not going to want to converse with a bank, not even using a natural language UI. They want to do that first with the provider of the product or service they actually need. And later on, they will most likely want their own agent who will do the talking. Spoiler: that agent won’t be wearing your bank’s cape.
Make no mistake. I believe banks should go ahead and license an LLM platform for their workforce. Let employees experiment and automate workflows with it. Employees are, after all, at least in part representative of your customer base, which will soon use similar tools and agents. If utilised correctly, it can be a means of preparing for the future.
Having said all that, the road is still full of quite big bumps. Regulation, AI understanding of product context, understanding what’s behind the headline APR number in terms of risk appetite, the legalities behind agency – these are all challenges yet to be overcome.
Part of a bigger universe
JPMorgan’s Open Banking fees announcement aside, banks are slowly, through embedded finance, learning to operate in the background. They are structuring their services to be used by third parties in a much more modular fashion. They realise they can no longer ‘own’ the customer and have to cooperate with those who have a stronger claim to that.
In Iron Man’s post-credits scene, Nick Fury tells Tony Stark: “You think you’re the only superhero in the world? Mr. Stark, you’ve become part of a bigger universe. You just don’t know it yet.” Like Stark, banks might think they’re the heroes of their agentic AI story. In reality, they’re just discovering they’re characters in a much bigger universe.
-
Shaul is a London-based fintech advisor with a global outlook on embedded finance, banking-as-a-service and payments infrastructure. Shaul works with startups, financial institutions and governments to accelerate innovation across industry verticals, launch new product propositions and enter new markets.
Previously, Shaul was head of banking, growth at Railsbank where he led international expansion to the US and Australia. Prior to Railsbank, Shaul was the UK Government’s first fintech advisor reporting to HM Treasury where he played a role building the UK fintech ecosystem, cementing its global leadership position. Earlier in his career, Shaul held various roles within retail banking technology and finance. He is also a basketball junkie.View all posts