AI Is Eating Entry-Level Finance Jobs. Here Is What the Data Actually Shows.

The Prediction That Started the Conversation

In early 2025 Anthropic CEO Dario Amodei said something that financial and consulting circles are still arguing about: AI he predicted could wipe out about 50 percent of entry-level management jobs within a few years.to directly observe what companies asked that system to do

The question worth asking now in mid-2026 is not whether Amodei was right in theory. It's what the data actually shows eighteen months later with actual layoff numbers on the table. Goldman Sachs estimated in a 2026 analysis that AI is already reducing U.S. employment by about 16,000 jobs per month. Carnegie's June market commentary pointed to entry-level white-collar work as one of the places whereAI adoption will be hit hardest in 2026. Factory job cuts that same month approached financial crisis and COVID-era levels and Challenger Gray & Christmas layoff data for June 2026 showed that factory and knowledge job reductions occurred at the same time a breadth that previous cycles rarely did

What Is Actually Being Automated

The Yale School of Management published a surprising analysis in April 2026 about where AI's impacts on work are really concentrated. That's not where the sci-fi version of automation was pointing. Forget assembly lines and truck drivers. The roles on display in 2026 are entry-level jobs in the knowledge economy: analyst work based on data collection and synthesis junior coding jobs that mostly produce repetitive text junior associate workyear based on document review and research administrative tasks that route information between people and systems. Those are exactly the roles that have functioned as career paths in finance consulting law and technology for decades

Specific financial tasks are already well documented. Financial modeling work creating three financial statement models running DCF analysis and filling out comparison tables can be performed by AI coding tools in a fraction of the time needed by a human analyst. Synthesizing research reading 10-Ks earnings call transcripts and industry reports to produce an investment thesis summary comes close to exactly the task that big language models like Claude andGPT-5. Data room due diligence review the arduous task of reading hundreds of documents to detect material problems in an M&A deal is being automated at law firms and investment banks at the same time. These are the tasks that juniors at Goldman McKinsey and Cravath have traditionally performed during their first two years and that is changing within the exact roles I'm applying for this cycle

The Measurement Problem: Why This Is Hard to Count

Before we continue it's worth reflecting on how difficult it really is to measure this because that difficulty is doing a lot of silent work in this debate. When a company lays someone off that shows up. Someone files for unemployment a WARNING notice is posted a headline appears. When a company simply doesn't hire someone it could have hired in a different year nothing shows up anywhere. You can't apply for a job that was never created. Call this the ghost hiring problemThe channel's researchers believe that AI is working right now: reducing hiring rather than active layoffs it is almost invisible in the data that normally tracks the labor market

That's why the numbers above are model-based estimates not headcounts pulled from a ledger. Goldman's 16,000-a-month figure is derived from broader labor market indicators not a count of layoffs attributed to AI because such a count doesn't exist anywhere. The Challenger data captures announced layoffs a real and useful signal but structurally they can't see a paper that a bank simply decided not to publish this year. None of that makesestimates are useless. It means that each number in this article should be read as the best model of a process that is really difficult to observe directly not as a strict count that someone can audit

The "AI Washing" Problem

Not all job losses attributed to AI in 2026 will actually be caused by AI. Harvard Business Review published a major article in January 2026 that documented something it called "AI washing": companies citing AI as the reason for layoffs that were actually driven by overcapacity interest rate pressure or simple strategic restructuring. Nearly 60 percent of U.S. hiring managers surveyed byResume.org said they planned to make layoffs in 2026 with AI as the most cited reason. Only 9 percent said AI had completely replaced certain functions. Sit with that space for a second. Sixty percent cite it and nine percent experience it. This is a strong sign that AI has become a socially acceptable explanation for a restructuring that would have happened anyway. It inflates how big the effect of AI seems from the outside and creates a speciesof anxiety that is far ahead of the real structural change that lies beneath

A Worked Example: Decomposing a Headcount Number

This is the exercise I do when I see a headline saying the analyst class is shrinking again. Let's take a hypothetical bank not a real one. Let's say its incoming class of investment banking analysts was 100 people in 2023 and let's say the class of 2026 is 78. That's a drop of 22 people or a decrease of 22 percent since 22 divided by 100 is0.22. A headline would call it an AI story. I don't think you can not without more work first

Divide the 22-person drop into buckets. Start with a baseline: call it 8 spots left vacant in any soft deal year AI or not since banks have always cut the incoming class when deal volume declines. That baseline isn't something the bank discloses it's an assumption drawn from how staff behaved in the last slow year I can point to 2016. Next let's assume the deal count theM&A announcements and IPOs combined fell by an amount that based on a rough rule of thumb that staff tracks deal volume represents another 9 of the 22. That's the rate cycle not the AI. Let's say 3 more changed instead of disappearing moved to a lower-cost analysis center instead of being eliminated entirely. Add them up: 8 plus 9 plus 3 is 20. Subtract that from 22 and you're left2. Two people from a 22 percent decline in the headlines land in the bucket that would actually attribute to AI doing the job a junior used to do

To make it clear what this example is and what it is not: 100 78 8 9 and 3 are numbers I made up to follow logic not revelations from any real bank. No bank publishes this breakdown which is exactly the problem. The point is not that the real participation of AI is always 2 out of 22. It is that a simple reduction in staff is a set of causes accumulated on top of each other and without a breakdown like this it cannotYou can tell which cause is doing the work. Anyone who gives you the AI ratio of a headcount figure with any real confidence is skipping a step that honestly can't be skipped from outside the company

Case Study: What Happened When Trading Floors Went Electronic

To find a real precedent I like to go back to what happened with floor trading. Stock exchanges used to run on shouts literally people standing in a pit shouting and waving orders. The New York Stock Exchange was for most of the 20th century the noisy physical center of American finance. Then came electronic execution. Decimalization in 2001 reduced the profits that banks and market makers could make on a spread. The 2005 NMS regulationIt forced orders to go to wherever had the best price favoring fast electronic systems over floor brokers. The New York Stock Exchange launched what it called a hybrid market in 2006 combining floor trading with electronic execution and continued to push electronic further in subsequent years

The predicted apocalypse was simple: Trading jobs would disappear and mostly they did. By most accounts the floor went from thousands of traders and brokers in its peak years to a small fraction in about a decade. If a floor trader was told in the late 1990s that this was about to happen the fear would have looked a lot like the fear beginning finance students have about AI right now

What actually happened is more interesting than the drop in the gross job count. Floor jobs in the specific form they existed mostly disappeared. But trading work didn't disappear it moved and changed shape. Electronic marketplace creation became its own industry employing people who understand the microstructure of the market and write the systems that quote prices in microseconds. Compliance and market structure roles grew because someone has to monitor systems complicated enough to need a rule.as Reg NMS. Quantitative and technological roles within banks and trading firms expanded to build the infrastructure that replaced human intermediaries. None of those roles were an individual swap for a floor trader job and the transition was not easy for people whose specific jobs disappeared. But the overall demand for people who understand the markets did not collapse as a headline about the template would suggest. It was reorganized around a different skill set closer to writing and monitoring systems than toshout hand signals across a well

Why Blaming AI Is Harder Than It Looks

Let me make the skeptical case as strong as I can because I think it deserves more than a token paragraph. The claim that AI is reducing entry-level financial contracting has to overcome at least three confounding factors before it is considered established and I don't think it completely survives all three yet

The first is the rate cycleInvestment banking staff have always tracked trading volume and trading volume tracks the cost of capital. When borrowing is expensive M&A slows IPOs are shelved and banks need fewer entities to take on fewer deals. That relationship predates AI by decades. Any year with a tighter Fed wherever the policy rate is is a year in which banks would cut the class of incoming analysts with or withoutsingle AI tool in the building. If hiring in 2026 is smooth the first honest question is not to what extent AI did this but what hiring would have been like at this point in the rate cycle anyway

The second is relocation which is also not new. Banks have been building analysis and support centers in places like Bengaluru Mumbai and Hyderabad for almost two decades moving increasingly senior jobs to lower-cost locations long before large language models existed. A shrinking analyst class in New York or London could just as easily reflect staff quietly moving to an existing offshore center or it could reflect the disappearance of staff in a chatbot. Both produce the same local headline about layoffs. They are not the samehistory

The third is ordinary. cost discipline the least interesting and probably the most underrated explanation. Banks operate in cycles of overhiring during a boom and cutting afterward on a timeline that has nothing to do with technology. Wall Street hired aggressively during the pandemic deal boom then spent 2022 and 2023 making big cuts once that boom ended long before generative AI tools were incorporated into analyst workflows. A company that reduces its class of2026 could simply be doing what companies have always done after a hiring spree and turning to AI as an explanation because it's more flattering to shareholders than "we overhired in 2021."

None of this means that AI isn't doing anything. I think it's doing something real because task-level evidence convinces me more than aggregate staffing numbers. But it means that every time I see a bank staffing chart with a downward slope and a headline that blames AI I want that headline to defend itself against rates offshoring and the ordinary boom-and-bust cycle before I believe that AI's share of the slope is large. Most captions don't even try

What This Means for Finance Students Specifically

The honest read for someone entering finance in 2026 is neither "AI will take over every job" nor "AI doesn't change anything." It falls somewhere in the middle and is less comfortable than either extreme. The mix of skills required for entry-level finance jobs is changing and the number of roles exclusively focused on execution is declining. A first-year analyst who can only build a model will find that role compressed. A first-year analyst who can direct AI tools verifyproduction capturing the edge case that the model missed and turning the result into a client-ready narrative with real judgment behind it is worth more than before because that person is effectively doing what previously required two or three people

The skills that increase in value are judgment customer relationship management qualitative synthesis and the ability to spot an AI result that is incorrect or subtly misleading. Those are exactly the skills that entry-level roles have historically been weaker in teaching because junior analysts spent their time performing jobs that never required them. The awkward thing is that the skills needed to succeed in finance in 2026 used to take about five years to develop on the job. Now you need to come up with moreof them or build them faster than the previous schedule assumed. The on-ramp is compressing. Destiny has gone nowhere

Where I Have Skin in This Game

I should clearly state the obvious instead of pretending to be a neutral observer. I'm a college student applying for exactly the entry-level finance jobs this article is about. If the pessimistic version of this story is true this is about my hiring cycle not some abstract cohort I'm reporting on from a distance. That gives me a real incentive to want the optimistic reading to be correct and I think the honest thing to do is to mention that bias out loud instead of writing about it

Here's what I really believe bringing that bias to light. I don't think the aggregate numbers - the Goldman estimate the Challenger data the Resume.org survey - are accurate enough on their own to tell me how worried I should be. They're the kind of numbers that are really hard to construct and easy to misinterpret for reasons laid out in the measurements section above. What moves me most is the evidence at the task level which tasks are being automated andto what extent. Comparing Yale's analysis to my own experience with these tools I think first-pass modeling and first-pass research synthesis are actually being compressed not as a corporate excuse. That's not a story I tell myself because it's comfortable. If anything it's the least comfortable reading because it says that the compression is specific and real rather than a vague deniable vibe

What I do with that in practice is treat the tools as something to be good at rather than something to resent. I use them in the same categories of tasks that this article describes a first-pass model a document summary a structured research question then I spend the remaining time checking the result and pressing the parts that seem out of place. I'm not sure that's optimal and I want to be clear that this is a description of what I do not a recommendation for anyone else. I don't know what specific optionsthey protect a given student's odds in this hiring cycle and I'm suspicious of anyone including a version of myself in a different state of mind who confidently claims to know that

The Bottom Line

Dario Amodei's prediction that AI could eliminate about half of entry-level white-collar jobs came from someone with a real point of view and eighteen months later the aggregate numbers Goldman's estimate of about 16,000 jobs a month the Challenger data the Resume.org survey are consistent with something happening without being precise enough to say exactly how much of that is AI. The task-level evidence is more compelling:Yale analysis and targeted automation of modeling research synthesis and document review point to a real mechanism-level shift in what entry-level finance work is. But the workforce is confounded by the rate cycle offshoring and ordinary cost discipline that follows every hiring boom and most headlines that blame AI don't even attempt to separate those threads. Floor trading is the precedent I keep coming back to: Most job-specificThey disappeared and the overall demand for people who understand the markets did not collapse as the headline suggests but rather reorganized around a different skill set. My own reading offered as a college student with an obvious interest in the answer is that the on-ramp to finance is compressing rather than disappearing. The change in what is rewarded describes where the market is moving it is not a strategy I am prescribing and I could be wrong even on that