When I Thought Investment Banking Was Technology-Proof
In 2016, fintech looked like a payments problem. Ten years later, AI has challenged what I thought made investment banking uniquely human.
In 2016, I worked in the investment-banking division of a Malaysian branch of a Singaporean bank.
One morning, during our regular meeting, my boss asked a question that felt important at the time:
“What effect will fintech have on our work?”
We worked across debt capital markets, equity capital markets, mergers and acquisitions, loan syndications, and project finance. Much of our work involved research, analysis, financial modelling, structuring, preparing materials, reviewing transaction documents, and using our knowledge and judgement to propose and execute deals.
Fintech was already a fashionable word. But, from where I sat, it seemed to belong somewhere else.
It was about e-wallets, digital payments, mobile banking, and consumer-facing financial platforms. It might change how people paid for coffee, transferred money, or interacted with their banks. But it did not seem capable of advising a company on a bond issuance, evaluating an acquisition, structuring a financing, negotiating with lenders, or preparing a credible pitch to a client.
The most advanced technology in my daily working life was probably the Bloomberg terminal.
So, consciously or not, I reached a reassuring conclusion: fintech would be important, but it would not meaningfully affect investment banking.
Our work required human judgement.
Ten years later, I think my younger self would be horrified—not because artificial intelligence can already replace an investment banker in every respect, but because it can increasingly perform work that I once regarded as inseparable from human professional judgement.
The error was not that I failed to notice technology. The error was that I noticed it, saw its limits, and assumed those limits would define its future.
Fintech was the wrong question
In 2016, the question was whether fintech would disrupt investment banking.
Looking back, the more important question was whether technology would eventually disrupt knowledge work.
Fintech, as we understood it then, mainly changed how money moved. It improved payments, access, interfaces, and retail financial services. Its most visible effects were transactional: faster transfers, digital wallets, online customer onboarding, and new ways for consumers and small businesses to access financial products.
Investment banking appeared to sit outside this disruption.
Our work was not mainly about moving money. It was about making sense of complex information under uncertainty. We tried to understand businesses, markets, financial statements, client objectives, funding conditions, investor appetite, comparable transactions, legal constraints, and commercial risk.
We did not merely retrieve information. We interpreted it.
We did not merely build models. We decided which assumptions mattered.
We did not merely produce documents. We tried to persuade clients, investors, credit committees, and counterparties that a particular transaction made sense.
That seemed fundamentally human.
And perhaps it still is, at least in its most consequential form. A board deciding whether to acquire another company, a lender deciding whether to finance a project, or an issuer deciding how to raise capital will still want accountable people involved. They will want someone who understands the context, challenges assumptions, handles competing incentives, and accepts responsibility for the outcome.
But that is not the same as saying the profession is protected.
AI does not need to replace the investment banker completely to change investment banking. It only needs to automate enough of the work surrounding the banker’s judgement to change the economics of employing one.
The work around judgement
A great deal of investment-banking work is not a single act of inspired judgement. It is a chain of activities that gradually leads toward a recommendation or transaction.
It involves gathering information, reading reports, comparing companies, extracting data, summarising markets, reviewing precedents, building draft models, identifying risks, preparing presentations, writing transaction materials, tracking comments, coordinating specialists, and revising the work repeatedly.
In 2016, much of that work required junior bankers, analysts, associates, vice-presidents, and senior staff. The work was time-consuming because information had to be found, read, understood, formatted, checked, and turned into something useful.
Today, AI can already assist with many parts of that chain.
It can search and synthesise large volumes of text. It can extract information from annual reports and prospectuses. It can draft research summaries, presentation outlines, financial commentary, comparison tables, and first versions of internal memoranda. It can help identify questions, generate scenarios, organise a model, compare documents, and highlight potentially relevant information.
Finance teams are already using generative AI for tasks such as drafting commentary, summarising performance, and supporting scenario modelling.
None of this means an AI system can independently advise a board on a complex acquisition or replace a senior banker’s relationships and commercial instincts.
But it does mean that the work once performed by a large pyramid of junior and mid-level professionals may require fewer people.
That is the point many professionals miss.
The future does not have to be a choice between two extremes:
- AI replaces nobody because humans are indispensable.
- AI replaces everyone because machines become superior at everything.
The more likely path is less dramatic but more economically important.
AI may compress the work beneath senior judgement. It may reduce the time required to prepare a first draft, perform preliminary analysis, collect market intelligence, review a document, produce a client update, or coordinate a process. It may allow a smaller team to produce the output that once required a larger one.
The profession may survive. The job may remain. But the number of people required, the tasks assigned to junior staff, the expected speed of delivery, and the value placed on routine analytical work may all change.
A programmer’s version of my mistake
This is not only an investment-banking story.
Consider a senior software developer today.
They may look at AI-generated code and reasonably conclude that the technology is not ready to replace them. It can produce an impressive function, explain code, suggest a test, or create a plausible application. But it can also misunderstand requirements, introduce security flaws, fail to grasp the wider architecture, and generate code that looks correct while creating subtle problems in production.
The senior developer still needs to review it.
That conclusion is sensible. Trust in AI tools among developers remains limited: only 29% of respondents to Stack Overflow’s 2025 survey said they trusted AI tools, down 11 percentage points from the prior year. A separate 2026 developer survey found that 96% did not fully trust AI-generated code to be functionally correct.
But the reasoning can slide into complacency.
The developer may say: “AI cannot write reliable production software without a senior developer reviewing it. Therefore, software development is safe.”
That sounds familiar.
In 2016, I could have said: “Fintech cannot structure a syndicated loan, advise on an acquisition, or understand a company’s financing needs. Therefore, investment banking is safe.”
Both statements may be accurate in the moment. Both can be dangerous when transformed into permanent predictions.
The question should not be whether an AI coding assistant can replace a senior developer today.
The question is whether the senior developer’s current role is the final human boundary—or merely the next capability that AI systems will learn to support.
Who reviews the AI?
Today’s software-development workflow increasingly looks like this:
Human writes code → Human reviews code → Human deploys and monitors it
Today’s software-development workflow increasingly looks like this:
AI drafts code → AI tests code → AI reviews code → Human oversees high-risk decisions
An AI agent can already be designed to read a codebase, review a proposed change, run tests, inspect dependencies, identify common security weaknesses, compare code against development standards, and propose improvements. Other agents can challenge the first agent’s conclusions, generate adversarial tests, monitor production behaviour, or investigate failures.
The idea of AI reviewing AI may sound absurd only because we are accustomed to treating the human reviewer as the final source of judgement.
But in finance, we should understand the underlying logic.
A transaction is not made safer simply because one person reviews another person’s work. The control environment depends on evidence, segregation of duties, independent challenge, documentation, limits, escalation procedures, and accountability.
The same will be true of AI systems.
An AI agent reviewing another AI agent may not be a sufficient control, especially if both systems share the same blind spots, data limitations, or flawed assumptions. But it may become part of a broader system: one agent produces, another tests, another challenges, a rules engine checks, a security system scans, and a human reviews the decisions that remain ambiguous, material, or irreversible.
A 2026 paper went so far as to argue that coding agents had reached a threshold where traditional human code review could become redundant and be replaced by agent-driven verification. That is an ambitious and contested claim—not an established conclusion—but it demonstrates how rapidly the “humans will always review the AI” assumption is being tested.
The senior developer may remain essential. But their role may become less about writing every line of code and more about defining objectives, setting architectural constraints, identifying unacceptable risks, resolving conflicts between business priorities, and taking responsibility for outcomes.
That is not disappearance. But it is transformation.
And transformation can still be painful for a profession.
The hubris of temporary limits
This is what I mean by hubris.
It is not arrogance in the simple sense. It is not believing that finance professionals or programmers are unintelligent. It is not denying that human judgement has value.
It is the habit of taking a technology’s present weakness and treating it as a permanent boundary.
In 2016, fintech could not do the work I did. That was true.
But I confused a temporary capability gap with a durable division of labour.
I saw that technology was effective in payments and retail financial services but weak in investment banking. From this, I inferred that investment banking was naturally protected.
I did not sufficiently consider that a different class of technology might emerge—one built not around moving money, but around reading, writing, analysing, comparing, modelling, coding, and generating plausible responses to complex questions.
AI has begun to challenge precisely those activities.
The uncomfortable question for finance professionals is therefore not:
“Can AI replace me today?”
It is:
“Which parts of what I do today would I have considered impossible for a machine to do ten years ago?”
The answer may include more than we are comfortable admitting.
Research summaries. Market commentary. First-draft investment memos. Comparable-company analysis. Document extraction. Earnings reviews. Scenario analysis. Due-diligence support. Client materials. Meeting notes. Compliance queries. Workflow coordination.
None of these alone is the entire job of a banker, fund manager, auditor, lawyer, accountant, risk professional, or software developer.
But a profession is not only defined by its highest and most difficult task. It is also defined by the volume of work required to support that task.
If AI reduces that supporting work, the profession’s structure changes.
The shrinking island
The likely impact of AI may not resemble a tsunami.
A tsunami is visible, dramatic, and immediate. It arrives all at once.
AI may instead be like the tide, quietly eating away at the edges of your island.
First, it takes the work we dismiss as routine: research, drafting, formatting, reconciliation, coding, testing, and review.
Then it takes the work we call preliminary analysis: summarising, comparing, detecting patterns, identifying risks, and preparing options.
Then it begins to challenge the activities that used to demonstrate professional experience: recognising precedents, applying familiar frameworks, asking standard questions, and producing a credible first view.
The island does not vanish overnight. But it becomes smaller.
The remaining land may still be valuable. It may contain the work that requires trust, leadership, ethical judgement, deep domain knowledge, commercial instinct, negotiation, and accountable decision-making.
But fewer people may be needed to occupy it.
That is why the question is not whether AI will make humans irrelevant. It is whether we are preparing for a world in which human value moves upward, while the work beneath it becomes automated, accelerated, and less scarce.
What remains human
There are good reasons to believe that experienced professionals will continue to matter.
Someone must decide which question is worth asking. Someone must determine whether the available data is complete, reliable, and relevant. Someone must understand what a client really needs—not merely what they requested. Someone must resolve trade-offs between speed, cost, risk, return, regulation, and reputation.
Someone must challenge a persuasive but flawed answer.
And someone must be accountable.
Those responsibilities may become more important, not less, as AI becomes embedded in professional work.
But they are not a reason for complacency.
The future finance professional will not compete with AI by working like a machine—reading faster, writing more standardised reports, producing more slides, or performing routine analysis more efficiently.
They will add value by doing what institutions will still need people to do: defining the problem correctly, identifying missing information, challenging assumptions, understanding context, communicating uncertainty, managing relationships, making difficult trade-offs, and accepting responsibility for consequential decisions.
In 2016, I thought investment banking was technology-proof because technology had not yet reached the work that I considered uniquely human.
Today, I see the danger in that assumption.
The hubris is not believing that humans still matter.
The hubris is believing that the shoreline will remain where it is.