The Trust Business: The Future of Banking in the Age of AI

Every wave of automation in finance has been predicted to end the same way, with the humans gone. Every wave has been wrong about how. Here is what the next twenty years actually hold, and why the bank that survives will be the one that remembers trust is the only product it ever truly sold.

---

When the automated teller machine arrived, the forecast was unanimous: the human teller was finished. The logic looked clean. A machine that hands out cash around the clock will obviously replace the person who hands out cash from nine to five.

It did not happen. For about two decades, as the economist James Bessen documented in the IMF's Finance & Development, the number of bank tellers in the United States held steady and even rose, climbing from roughly 500,000 in the 1980s toward 600,000 by 2010, even as some 400,000 ATMs were installed. The machine cut the number of tellers each branch needed, which made branches cheaper to run, so banks opened far more of them. The job changed, from counting notes to building relationships, but the people stayed. The obvious first-order prediction, machines replace humans, was swamped by a second-order effect that almost nobody bothered to model. The teller count has since fallen as online and mobile banking matured, and that is the real lesson, not that automation never costs jobs, but that its second-order effects are powerful, hard to predict, and run on their own timeline.

I start here because the future of banking is being forecast right now with the same confident, clean, first-order logic. And it is about to make a version of the same mistake.

The consensus is not wrong, exactly. AI will automate enormous amounts of routine financial work, and the headline numbers are real. The consensus is just incomplete. It stops at the sentence where the interesting part begins. Strip out everything that can be automated, commoditized, and decentralized, and what remains is not nothing. What remains is trust. The future of banking is a long argument about who earns it and who loses it, and that is the part the spreadsheets cannot forecast.

A pattern, and possibly a break

Banking has survived this story before. Electronic funds transfer was going to empty the back office. Online banking was going to close the branches. Mobile was going to kill the desktop. Fintech was going to eat the incumbents whole. Each wave was predicted to gut the workforce. Each one, instead, moved the place where humans added value and grew the total amount of activity.

There is an honest reason to think this wave might be different. Previous technologies automated execution, the moving of money and the processing of forms. This one targets judgment itself. Citi's 2024 report on AI in finance made the point bluntly: banking is among the most exposed sectors in the entire economy, because so much of its work is the kind of cognitive labor that large models now do well. That is the case for a genuine break in the pattern.

So which is it, pattern or break? The honest answer is both, and the timeline tells you where each applies.

Where banking stands in 2026

Look closely at a large bank today and you see a contradiction held in a single breath. Record investment in AI. Record profit forecasts. And, very politely, in rooms with the doors closed, record planning for a smaller workforce.

The reason the contradiction is surfacing now is that the technology has crossed from demonstration to deployment. Deloitte's 2026 banking and capital markets outlook frames the year as the moment banks must push agentic AI out of the pilot phase and into governed, enterprise-scale production. Most banks have pilots. Few have scaled them. That gap is the story of the next twelve months. And it is not theoretical: FIS, working with Anthropic, has built a Financial Crimes AI Agent that compresses anti-money-laundering investigations from hours into minutes, the sort of task that used to occupy whole floors of analysts. Multiply that across compliance, operations, servicing, and reporting, and you have the shape of the next five years.

Three horizons

Anyone who tells you what banking looks like in 2040 with a straight face is guessing. The trick is to be precise about confidence. The near term is forecastable from real data. The middle term has a genuine institutional consensus. The far term is open, and that is where judgment, not data, has to do the work.

2026 to 2027: already happening

This horizon needs no imagination, only attention.

The back and middle office contract first. Citi estimates that around 54 percent of banking jobs sit in the high-potential-for-automation category, with another 12 percent open to augmentation, the highest exposure of any industry it studied (insurance follows at 48 percent, capital markets at 40 percent). Bloomberg Intelligence put a number on the human cost in early 2025: global banks could shed as many as 200,000 jobs within three to five years, a net cut of around 3 percent of the workforce, concentrated in operations, the back office, and the middle office.

Customer service starts to wear a synthetic face. AI avatars and voice agents take the routine inquiries, the balance checks, the card replacements, the password resets. Done well, this is not the dystopia people fear. It is the beginning of a split that matters later: as machines absorb the routine, reaching a real human becomes the premium tier.

The crypto question stops being hypothetical. After the GENIUS Act of July 2025, the first U.S. federal law for stablecoins, the regulator moved fast. In December 2025 the Office of the Comptroller of the Currency granted conditional national trust-bank charters to Circle, Ripple, Fidelity Digital Assets, BitGo, and Paxos, following Anchorage Digital as the first chartered crypto bank, with Coinbase among those that had filed. These are limited charters, weighted toward custody rather than full deposit-taking and lending, so it is a doorway, not a finished building. But the direction is unmistakable. The story of the decade is not crypto collectives replacing banks. It is crypto firms becoming banks, and banks adopting crypto rails. The two are converging.

Security moves from cost center to core function. The World Economic Forum already ranks cybersecurity among the fastest-rising skills of the decade, and the scale of the problem is staggering: the United Nations estimates roughly 2 trillion dollars in illicit funds move through the global financial system each year, while U.S. institutions alone spend 35 to 40 billion dollars a year fighting money laundering. As money becomes faster, more automated, and more autonomous, protecting it stops being plumbing and becomes the product.

And one disruption that has nothing to do with AI deserves a line. In a growing list of regions, climate risk is pulling insurers out of the market. No insurance means no mortgage, and no mortgage means a core banking product simply cannot be sold there. The risk models the whole industry runs on were not built for a world where entire postcodes become uninsurable.

Toward 2030: the consensus

This is where the credible forecasts cluster, and where the numbers get specific.

The profit-and-people split widens into a moral question. The same reports that promise mass displacement also promise abundance. Citi projects that AI could add about 170 billion dollars to global banking profits by 2028, a 9 percent lift that would take the total pool close to 2 trillion dollars. Bloomberg Intelligence sees pre-tax profits running 12 to 17 percent higher by 2027, as much as 180 billion dollars added to the industry's bottom line. So the surplus is real, and the displacement is real, and they arrive together. The question almost no one in the consulting literature will touch is the one that matters most: where does the surplus go, and what does an institution owe the people it automates? That is not an economic question. It is a values question, and banks will be asked it in public.

The job mix inverts. The World Economic Forum's Future of Jobs Report 2025 has bank tellers among the fastest-declining roles globally, while roles in AI, big data, fintech, and cybersecurity grow fastest. It is worth noting the report is not a doom forecast: it projects 170 million new roles created and 92 million displaced by 2030, a net gain of 78 million jobs. The shape of the change is a shift, not a collapse. And it carries a counterintuitive finding that points straight at the human premium: demand for human skills keeps climbing. McKinsey estimates demand for social and emotional skills rising by around a quarter by 2030, 26 percent in the United States and 22 percent in Europe, even as routine processing falls away. The market is automating the head and bidding up the heart.

Management itself gets automated. The quiet target of this wave is the middle manager. Scheduling, resource allocation, performance tracking, status reporting, coordination, all of it is exactly what agents do well. What is left for a human manager is the part a machine cannot fake: coaching, meaning, judgment under ambiguity, and trust. The risk worth naming, which most forecasts skip, is that algorithmic management can curdle into algorithmic surveillance. How a bank handles that line will say more about its culture than any mission statement.

The deposit base, banking's quietest advantage, becomes contestable. As balances drift toward Big Tech wallets and stablecoins, the cheapest and stickiest funding in all of finance is suddenly in play.

And here is the most important tension of the whole decade, hiding in a demographic statistic. Cerulli projects that roughly 124 trillion dollars will pass to heirs and charity through 2048, the largest movement of private wealth ever recorded. Now layer on a second finding from the same research: more than 70 percent of heirs are likely to drop their parents' advisor after they inherit. Read those two facts together. At the precise moment AI can do the analytical work of wealth management, the relationships that hold a client are at their most fragile, and the client is changing generations. The machine arrives exactly when the trusted human becomes both scarce and decisive. That single tension is the entire argument for the human premium, written in numbers.

2030 to 2040: the open questions

Past 2030, no serious institution will plant a specific flag, because it stops being forecasting and starts being imagination. That is not a weakness. It is where a futurist earns the title. So these next ones are mine, offered as wagers rather than predictions, and they share a single thread: each one ends at the same question, which is who holds the off-switch.

My first wager is that housing as a service kills the mortgage. The mortgage has been the anchor product of retail banking for a century. If ownership gives way to subscription living, build-to-rent at scale, and institutional landlords, the product that defined the consumer bank simply shrinks. The freedom this could buy younger people is real, and so is the risk that they end up owning nothing and renting everything from balance sheets they will never see. A bank that wants to matter in that world has to decide which side of that bargain it is on.

My second wager is that banks become the tax collector. This sounds like science fiction until you learn it already has a name. The economist Edgar Feige proposed the Automated Payment Transaction tax in 1989, and presented it to a U.S. federal tax-reform panel in 2005: a tiny flat levy assessed and collected automatically by the payment system itself, every time money settles, replacing income, sales, and corporate taxes. Because the wealthy move far more money, they carry far more of the load despite the flat rate. Switzerland's microtax initiative ran on exactly this idea, proposed abolishing income tax and value-added tax, and won more than 23 percent of the vote in 2016, tied directly to the universal basic income debate. Now connect the dot. If the rails become the tax collector, the bank stops being an intermediary and becomes an arm of the state. That is the end of financial privacy and the beginning of programmable enforcement, money that can be taxed, frozen, or set to expire by rule. Whoever controls that switch holds a kind of power no bank has ever held. We should decide who that is on purpose, not by default.

My third wager is that the word bank becomes a license you rent rather than an institution you are. Embedded finance is already pushing banking into other people's apps, so that you bank inside your car, your accounting software, your favorite store, and the bank itself vanishes behind someone else's brand. Combine that with the crypto-charter convergence and you reach a world where holding a banking license is infrastructure, available to anyone with the rails and the compliance. And the moment anyone can hold the charter, the only thing that separates one provider from another is the one thing that cannot be chartered: trust.

A few more belong on the list, briefly. The rise of the solopreneur, with AI letting one person run what used to take a team, hands banks millions of micro-clients instead of mid-sized firms, and a generation that expects its finances to run themselves. Programmable money tied to verified digital identity creates rails that are efficient and, in the wrong hands, total. And a generation raised on transparency will increasingly choose a bank for where its money goes and who it harms, not for the rate, which makes ethics a competitive product rather than a press release.

The dots most forecasts miss

Two connections deserve their own moment, because almost everyone walks past them.

The first is the AI-powered client. Buried inside Citi's own report is a warning the industry has barely absorbed: AI in the hands of customers could intensify price competition and shift the balance of power toward them. Every forecast models the bank pointing AI at the customer. Almost none model the customer pointing AI back, with personal agents that shop rates continuously, switch providers overnight, and negotiate without sentiment. That collapses loyalty and margin from the demand side, and it is coming for the comfortable parts of retail finance first.

The second is the trust inversion. As deepfakes and AI-enabled fraud scale, the scarce and valuable thing a bank can offer is no longer speed or convenience, which everyone will have. It is verified trust and human recourse: a real person to call, and a real institution to hold accountable, when the machine gets it wrong. The bank of 2035 may make less of its money moving funds and more of it guaranteeing truth.

The human premium

Notice what every one of these scenarios circles back to. As the mechanics get automated, commoditized, and decentralized, value migrates to the capacities a machine cannot hold.

In the framework I use with the leadership teams I work with, those capacities have names. Belonging, the relationship that the 124-trillion-dollar handover desperately needs and AI cannot carry. Purpose, the values that decide what a bank is actually for. Adaptability, the steadiness a client pays a human for when the avatar cannot reassure them. Clear judgment, which is exactly the cognitive work Citi calls commoditizable, with the highest tier of it the part that stubbornly is not. And the human energy to keep caring under load, which turns out to be the entire recourse layer of the trust inversion.

The market is making this case in cold numbers. The same forces automating the analytical work are bidding up the relational work. Banking is quietly becoming one of the most human industries in the economy, precisely because so much of it is about to stop being done by humans.

What to do on Monday

For leaders, the move is redeploy rather than simply replace. The institutions that win the next decade will treat the capacity AI frees up as fuel for the human work that is becoming scarce, and will reskill toward judgment, relationship, and trust faster than they cut. A wave of layoffs is the easy decision and, in a trust business, often the expensive one.

For individuals, the skills that compound are the ones that sit furthest from automation: building trust, exercising judgment where the data runs out, and holding the human-recourse roles that grow more valuable as machines handle more. The career risk is not being replaced by AI. It is spending the next five years competing with it on the very ground where it is strongest.

What a bank is for

The teller did not disappear when the ATM arrived. The job became more human, not less, and the institutions that understood that got the next two decades right.

We are standing at the same kind of moment, only larger. When execution is free, when intelligence is cheap, when money moves itself, the only durable asset an institution can hold is trust, and trust is still built, earned, and broken by people. The banks that survive the age of AI will not be the ones with the best models. Everyone will have those. They will be the ones that remembered, while everyone else was counting what machines could do, that they were never in the money business. They were always in the trust business.

---

Bradley Hook writes and speaks on human performance in the age of intelligence, and is the founder of the FLAME Method.