The Banker's Question

I — The Question That Needed a Banker

The defining feature of financial advice, for most of the period in which it has existed as a paid service, was not the difficulty of the advice but the difficulty of reaching anyone qualified to give it. The substance of what a good adviser told a typical client was rarely arcane: spend less than you earn, diversify, mind costs and taxes, do not panic in a downturn, match the risk you take to the time you have. The value was not in the secrecy of this knowledge — most of it has been printed in personal-finance books for a century — but in its application to a particular situation by someone who could be reached, trusted, and held to account. And reaching that someone was gated: by minimum investable assets, by the cost of the relationship, by membership of the right institution. The advice was, in principle, available to all. Access to its competent application was not.

This gate had a second feature that the industry has never been comfortable discussing plainly. The person at it was usually paid, in whole or in part, to distribute product. The adviser who recommended a fund often earned from the fund; the banker who proposed a structured note often booked a margin on it; the broker who suggested a trade was compensated by the trade. The advice and the sale were entangled, and the client could not easily tell where one ended and the other began. Regulation has spent decades trying to separate them — disclosure rules, suitability requirements, fee-based models, fiduciary standards — with partial success, because the entanglement was structural rather than incidental. The intermediary's economics depended, to varying degrees, on the client transacting.

So the median client confronted a service that was gated by access and compromised by incentive. They paid for proximity to expertise, and accepted, knowingly or not, that the expertise came with a thumb on the scale. For a long time there was no alternative, because the only way to get the knowledge applied to your own circumstances was to go through a human who had a reason to want you to act. That is the arrangement now being dissolved, and the dissolution is more consequential than the industry's framing of "AI tools for advisers" suggests, because it does not improve the gate. It removes it.

The scarce thing was never the advice. It was the gate — and the person standing at it was paid, as much as anything, to sell you what lay beyond.

II — The Default Becomes Non-Advised

A capable conversational model will now discuss a person's financial situation at any hour, for free, with patience that does not run out, and — this is the decisive part — with no product to sell. It will explain the trade-off between paying down a mortgage and investing the difference. It will model the tax treatment of competing retirement vehicles. It will talk someone through whether their portfolio matches their stated risk tolerance, and push back when the two do not align. It does not do this better than the best human adviser. It does it better than the adviser the median person can actually reach — which, for most people most of the time, is no adviser at all, or one whose minimum they do not meet, or one whose incentive they cannot fully trust. The relevant comparison is never AI against the ideal adviser. It is AI against the realistic alternative, and against that alternative it wins on availability, on cost, and on the absence of a thumb on the scale.

The migration this produces is not confined to the mass market. Sophisticated individuals — people with the means to retain good advisers and the knowledge to use them — are increasingly consulting a model first, and arriving at the human, if at all, already informed, already holding a view, already able to interrogate what they are told. The model becomes the first port of call and the human becomes a second opinion, which is a profound inversion of the historical order, in which the human was the source and the client had nowhere else to go. When the default mode of financial decision-making becomes a conversation with an unconflicted machine, the human adviser is no longer the gateway to knowledge. They are a checkpoint the informed client may choose to visit, and increasingly may not.

For the mass-affluent advisory model, this is an economic problem, not merely a competitive one. That model was built on charging a recurring fee — often a percentage of assets — for a service whose core deliverables were precisely the things a model now supplies for nothing: a financial plan, an asset allocation, a fund selection, periodic reassurance. When the deliverables are free, the fee has nothing to attach to. The adviser can continue to charge it only for as long as the client does not realise the deliverables are available elsewhere at no cost — which is to say, for the duration of a transition period that is already underway and will not last. The honest description of where this leads is not that advice gets better. It is that advice, for the median client, stops being a paid service at all.

One caveat belongs here, and it is not a small one. The model's great present advantage — that it has no product to sell — is not guaranteed to persist. The economics of distribution will exert their usual pull, and an AI advice layer captured by product manufacturers, subtly steering toward what pays to be steered toward, is an entirely plausible future. The unconflicted sparring partner is the current reality and the reason for the migration; it is not a law of nature. If it is lost — if the free advice quietly reacquires a thumb on the scale — the dissolution described here reverses, and the gate is rebuilt in software. Whether that happens is a question of regulation and market structure, not of technology, and it is among the most important open questions in the whole transition.

III — What Actually Survives

If the default becomes non-advised, the human advisory relationship does not vanish; it bifurcates. On one side is the commoditised default, where the model does the work and the human adds nothing the client could not get for free. On the other is the genuine niche, where something survives that the model cannot supply. The analytical task — and the place where the industry tends to flatter itself — is to be honest about which of its claimed strengths fall on which side. Most of what the industry advertises as its value is on the wrong side. Access to information, asset allocation, portfolio construction, market commentary, the curation of research: these are exactly the legible, reproducible outputs a model now generates on demand. An adviser whose offer is "I will tell you what is happening in markets and how to position for it" is offering what has just become free.

What genuinely survives is narrower, and each surviving thing survives for the same reason: it is a legal or institutional fact, not an informational one — something that cannot be reproduced by a model because it is not, at bottom, knowledge. The first is fiduciary accountability: a licensed human or institution that bears responsibility for the outcome, that can be sued, sanctioned, or struck off, that has capital at risk and a reputation that can be destroyed. A model cannot bear this, not because it is not clever enough but because there is no entity to hold accountable — a point of structure, not of capability. For decisions large enough that someone must answer for them, the accountable human is not a conduit for information; they are a bearer of liability, and that role does not commoditise.

The others follow the same logic. There is access to genuinely restricted assets and opportunities — allocations, instruments, and deals that are gated by relationship, licence, or regulation, where the value is the access itself rather than the advice about it. There is complex multi-party structuring and execution — the assembly of a transaction across counterparties, jurisdictions, and legal forms, where the work is coordination and bespoke construction rather than recommendation. There is the behavioural and commitment function — the discipline of an external party who stops a client from selling at the bottom, not by informing them but by standing between them and the impulse, a service whose value is precisely that it is not self-administered. And there is execution that requires a licensed human in the loop — the tax, legal, and jurisdictional acts that a model can describe but cannot perform, because performing them requires authority the model does not and cannot hold.

What unites the survivors is that none of them is information. Each is a thing the industry has historically bundled with information and under-priced relative to it, because information was the visible, marketable part and these were the quiet accompaniments. The bundle is now being unbundled by force: the information component is racing to zero, and what is left is the accountability, the access, the structuring, the discipline, and the licensed execution. An advisory business with a genuine claim to one or more of these has a future. An advisory business whose real offer was information dressed as advice does not — and the transition period, during which clients have not yet fully noticed, is the interval in which that distinction becomes destiny.

Which side of the line

Commoditised (the model does it for free): market views, asset allocation, portfolio construction, fund and product selection, research curation, financial planning, periodic reassurance — the legible, reproducible outputs that were the industry's main advertised value.

Surviving (legal or institutional facts, not information): fiduciary accountability and borne liability; access to genuinely restricted assets; complex multi-party structuring; the behavioural commitment function; and execution requiring a licensed human. None of these is knowledge, which is why none commoditises.

IV — The Venue Follows the Same Logic

The dissolution of the advice gate has a structural sibling in the trading venue, and the two are driven by the same force. Markets have historically operated in sessions — open in the morning, closed at night, shut at weekends — and the convention is old enough to feel natural, but its causes were practical and are being removed one by one. Trading hours existed because the participants who made markets work were human: market-makers who needed to sleep, back offices that settled in batches overnight, exchanges that reconciled and netted between sessions. The closure was not a feature anyone designed for its own sake. It was the accommodation a market made to the humans inside it.

Remove the humans from the mechanism — replace the market-maker with an always-on algorithm, the overnight batch with continuous settlement, the human reconciliation with automated clearing — and the practical reasons for closure fall away. Tokenisation pushes in the same direction, because a tokenised instrument settling on a ledger that never closes has no native session; its natural state is continuous. And the demand side is converging too: when the marginal participant is an automated strategy, or an AI agent acting for a retail user who expects to transact whenever they decide to, the appetite for markets that are open only part of the time becomes a friction rather than a norm. Continuous, around-the-clock trading stops being a peculiarity of crypto venues and starts looking like the direction of travel for markets generally — not because anyone has decreed it, but because the frictions that made closure necessary are being automated out of existence.

But the convergence is not costless, and the discipline that applies to the advice argument applies here too: what is gained is availability, and availability is not the same as resilience. Market closure performed functions beyond accommodating tired humans. The overnight and weekend gaps were cooling-off periods, intervals in which panic could subside and information could be digested before trading resumed — a circuit-breaker built into the calendar. Batch settlement between sessions netted enormous volumes of offsetting trades, an efficiency that continuous settlement partly forgoes. And the daily close provided a clean, universally agreed reference price, a coordination point that a market without sessions must reconstruct by other means. A market that never closes is more available and, in certain respects, more fragile: it concentrates the management of liquidity and risk into systems that must now operate without pause, removes the enforced intervals in which stress could dissipate, and asks always-on infrastructure to do what the rhythm of the trading day once did for free. The direction of travel is real. So are the things given up to travel in it.

The trading day was never sacred. It was the hours the humans could stay awake. Remove the humans and the day dissolves — along with the cooling-off the night used to provide.

V — The Same Shift, Twice

Set the two movements side by side and they are revealed as one. In the advice relationship, automation removes the human gatekeeper who stood between the client and the knowledge. In the trading venue, automation removes the human gatekeeper — the market-maker, the back office, the session itself — that stood between the participant and the market. Both are disintermediation by automation, and in both the thing being removed is the human whose limitations the old arrangement was built to accommodate: the adviser the client had to reach, the trader who had to sleep. When the limitation is automated away, the accommodation built around it dissolves, and the question in each case becomes the same: once the human conduit is gone, what is left that still requires a human at all?

The answer, in both domains, is identical, and it is the through-line of the whole argument. What survives is never the information — the information is precisely what automation makes free and abundant. What survives is what is not information: accountability, the bearing of liability by an entity that can be held to it; licence, the legal authority to perform acts that knowledge alone does not entitle one to perform; and structural privilege, genuine access to things that are gated by something other than ignorance. In the advice relationship these appear as the fiduciary, the restricted allocation, the licensed execution. In the trading venue they appear as the regulated infrastructure, the entity that bears settlement risk, the licensed operator answerable when the always-on system fails at three in the morning. The vocabulary differs; the residue is the same. Automation competes away everything that was only information, and leaves standing everything that was a fact of law or institution.

This is the same conclusion that the analysis of financial infrastructure keeps arriving at from other directions: that the binding constraint, once information and capability become abundant, is governance and accountability — who answers, under what authority, when something goes wrong. Technology has a long record of dissolving intermediaries who were really just conduits for information, and a matching record of failing to dissolve the ones who bear responsibility, because responsibility is not a technological function. The financial-services industry is now meeting that pattern at scale. Most of what it does is being competed toward zero, because most of what it does was information. The part that survives is the part it has historically undervalued — not the glamorous part, not the part that filled the marketing, but the quiet, accountable, licensed, privileged remainder that was doing the real work all along.

For the industry, the implication is uncomfortable but clarifying. The firms that survive will not be the ones with the best market views or the slickest research — those are now commodities. They will be the ones that can credibly bear accountability, that hold genuine access, that can structure what cannot be structured by a model, and that operate the licensed infrastructure the automated market still requires. Everything else is being unbundled and given away. The advice was always the smaller thing underneath the information — and now that the information is free, the smaller thing is all there is left to sell.

Automation dissolves the conduit and spares the one who answers for the outcome. Information was always the easy part to sell. Accountability was always the hard part to replace — and it is the only part that is left.

— What Is Actually Being Sold

The transition has a shape, and naming it is the practical payoff. For the individual, the lesson is that the free, unconflicted sparring partner is now the right first port of call for the great majority of financial questions, and that paying a human for what the machine supplies for nothing is a habit worth examining — while remaining alert to the one thing the machine cannot yet be trusted to have stayed: its lack of a product to sell. For the genuinely complex, the accountable, the restricted, and the licensed, the human remains indispensable, not as a source of information but as a bearer of something a model structurally cannot bear.

For the firm, the lesson is to stop defending the part that is already lost and start protecting the part that survives. Resources spent sharpening market views and polishing research are spent defending a commodity. Resources spent deepening genuine accountability, securing real access, building structuring capability, and operating licensed infrastructure are spent on the future. The unbundling will not be gentle, and the transition period — during which clients still pay for information out of habit — is precisely the window in which a firm decides which side of the line it will be standing on when the habit ends.

And for the market as a whole, the lesson is that availability and resilience are not the same thing, and that a financial system disintermediated by automation — advice without an adviser, venues without a close — will be more accessible, more continuous, and more efficient in the ways that information abundance delivers, while concentrating accountability and systemic risk into a smaller number of licensed, regulated, answerable points. That concentration is not a flaw to be engineered away. It is the residue the whole process leaves behind: the part that could not be automated because it was never information in the first place. The banker's question can now be answered by anyone, at any hour, for free. What still cannot be had for free is someone to answer for the outcome — and that, stripped of everything the machine has taken, is what the industry was always really selling.