The debate over artificial intelligence in investment banking is often framed as a simple question: will machines replace bankers? That framing misses the more consequential change already underway. AI is unlikely to eliminate the profession, but it is increasingly capable of automating the repetitive analytical and production work through which investment banking has traditionally been organised.
The scale of the potential disruption is significant. Citigroup has estimated that 54% of banking jobs have high potential for automation, with another 12% offering scope for augmentation. Deloitte, meanwhile, estimates that generative AI could increase front-office productivity at the world's 14 largest investment banks by 27%–35%, with investment banking potentially experiencing an average productivity improvement of about 34%.
The implications extend beyond headcount. Investment banking has historically used junior-level analytical work as both an operating model and a training system. If AI removes much of that work, banks face a less obvious challenge: how to develop the judgment, commercial instinct and relationship skills required of future senior bankers.
Ask this question at a dinner party and you will get two confident answers. The tech optimist says yes: banking is just spreadsheets and slide decks, and machines already do both better. The banker says never, because deals run on trust, and no algorithm has ever taken a CEO to dinner. The stakes behind the argument are not trivial: Citigroup has estimated that 54 per cent of banking jobs carry a high potential for automation, a larger share than in any other sector it examined, with a further 12 per cent open to augmentation (Bloomberg, 2024; Finextra, 2024). Both answers are wrong. The truth is stranger and more interesting: AI will not replace investment bankers. It will replace investment banking as we know it, while leaving a smaller, weirder, more human profession standing in its place. The sections that follow set out what will actually happen.
The Pyramid Is About to Lose Its Base
Investment banking has always been built like a pyramid. At the bottom, armies of analysts and associates grind through 90-hour weeks building financial models, formatting pitch books, and running comparable company analyses. At the top, a thin layer of managing directors closes deals over long relationships and longer dinners.
The dirty secret of the industry is that most of the work at the bottom of the pyramid was never intellectually demanding. It was just voluminous. A discounted cash flow model is not hard. Building forty versions of it before Monday morning is. That is precisely the kind of work AI devours. Drafting a confidential information memorandum, screening hundreds of potential acquirers, summarising thousands of pages of due diligence documents, turning around the twelfth ostensibly final version of a pitch deck at 3am: these tasks are structured, repetitive, and pattern-based. They are not the future of banking. They are the past, and the past is being automated in real time. Deloitte (2023) estimated that the top fourteen global investment banks could raise front-office productivity by 27 to 35 per cent through generative AI, worth roughly US$3.5 million in additional revenue per front-office employee by 2026. So the first thing that actually happens is that the analyst class shrinks dramatically. Not to zero, since someone still has to check the machine's work, but a team of twelve becomes a team of three, each supervising AI output rather than producing it by hand. This is already visible in hiring. Banks have been reported to be weighing reductions of up to two-thirds in incoming junior analyst classes (Business Insider, 2025), and by mid-2026 such reductions were being described as under way across major institutions (Fortune, 2026; Outsource Accelerator, 2026), alongside measurable declines in operations and support headcount (American Banker, 2026).
What AI Cannot Buy: A Reputation to Lose
When a company sells itself, it is not buying a valuation model. It is buying a person who will stake their name on the outcome. It is buying someone who knows which private equity partner is bluffing, which strategic buyer is desperate, and which board member secretly opposes the deal. It is buying a phone call that gets answered on the first ring because of a favour from 2011. This is the irreplaceable core of the business, and it is irreplaceable for a reason that has nothing to do with intelligence: accountability. An AI cannot be sued, fired, shamed, or trusted. It has no reputation to protect, no career to lose, no skin in the game. When US$2 billion changes hands, both sides want a human being whose future depends on getting it right. Deals are also, frankly, emotional. Founders selling their life's work do not need a probability distribution. They need someone to talk them off the ledge at midnight when they get cold feet. Negotiations turn on ego, timing, and reading the room. These are not data problems. They are human problems, and they will stay human.
The Real Disruption: The Apprenticeship Crisis
Here is the twist almost nobody is talking about, and it is the most important part of the story. For decades, investment banking trained its leaders through suffering. You learned to be a dealmaker by spending your twenties inside the machinery of deals: building the models, sitting in the data rooms, watching negotiations unravel and recover. The grunt work was not just labour. It was the curriculum. AI is now eating the curriculum. If analysts no longer build models by hand, how do they develop the intuition to know when a model is lying- If juniors never draft the documents, how do they learn what makes a deal narrative persuasive- The industry is quietly sawing off the ladder its own leaders climbed. The contradiction is arithmetic as wellascultural:banksdrawroughly62percentoftheirownAItalentfromthesamejuniorcohortstheyareshrinking, and senior judgment, as McKinsey's Debasish Patnaik has argued, cannot simply be recruited laterally (Fortune, 2026).
This means the banks that win the next decade will not be the ones with the best AI. Every firm will have roughly the same AI. The winners will be the firms that solve the apprenticeship problem by inventing new ways to manufacture judgment in young people without a decade of spreadsheet suffering. Expect deal simulations, early client exposure, and a career path that looks less like a pyramid and more like a fast-track guild.
The New Job Description
Put it all together, and the banker of 2035 looks like this. They command a suite of AI systems the way a senior banker once commanded a team of analysts, except the team works in seconds, never sleeps, and never quits for a hedge fund. Their scarce skills are not Excel and PowerPoint; they are client psychology, negotiation instinct, regulatory navigation, and the ability to interrogate machine output with a sceptic's eye. Fewer people will do this job. The ones who remain will be more productive, better paid, and, ironically, more human than their predecessors, because everything mechanical about the role will have been stripped away. Banking stops being a volume business of documents and becomes a pure judgment business of decisions. Mid-tier firms face the harshest reckoning. When elite advice becomes cheaper to produce, the middle of the market gets squeezed from both sides: boutiques with star rainmakers above, AI-powered platforms below. A note of caution is warranted. Much of the reduction in banking headcount to date reflects post-pandemic over- hiring and cyclical uncertainty rather than automation as such, and some observers argue that the displacement narrative has run ahead of the evidence (Fortune, 2025). The direction of travel is clear; the pace remains contested.
So, can AI replace investment bankers- It will replace the hours, not the humans. It will replace the pyramid, not the peak. It will hollow out everything about the job that was mechanical and leave behind only what was always the real product: trust, judgment, and a name on the line. The investment banker is not going extinct. But the investment banking career, the one built on a decade of all- nighters as the price of admission, is already gone. The question for the next generation is not whether a machine will take their job. It is whether the industry can still teach them how to do the part machines never could.
Key Takeaways
- AI will automate tasks before it replaces professions: Financial modelling, document review, valuation support, research and presentation production are among the activities most exposed to generative AI.
- Human judgment remains the scarce asset: Client trust, negotiation, accountability, reputation and complex decision-making are considerably harder to automate than analytical production.
- The apprenticeship model is under pressure: The junior work being automated has historically provided the practical training through which future senior bankers developed judgment.
- Competitive advantage will shift from technology to execution: As AI capabilities become broadly accessible, firms that redesign training, workflows and client relationships effectively may outperform those that simply deploy another AI tool.
**AI will make financial analysis abundant; the scarce asset will be the judgment to know what deserves to be believed
