Artificial Intelligence on Wall Street and the Demand for Human Skills

You give your disciples not truth, but only the semblance of truth; they will be hearers of many things and will have learned nothing; they will appear to be omniscient and will generally know nothing; they will be tiresome company, having the show of wisdom without the reality.

— Plato, Phaedrus, trans. Benjamin Jowett

The people that are really going to excel with AI are going to be able to critically think.

— Bob Deibel, Ramirez & Co., August 26, 2026

Generative AI is the latest technology to externalize a human capacity, and across Wall Street it now absorbs the procedural work that surrounds a decision, such as search, synthesis, drafting, and preliminary modeling, while the decision itself stays with the person accountable for the outcome. Because a model supplies the semblance of knowledge rather than accountable judgment, a financial professional’s comparative advantage has shifted to what AI cannot provide: defining the problem worth solving, verifying evidence through critical thinking, using AI fluently without deferring to it, and earning trust through communication, active listening, and relationship cultivation.

The paper places generative AI in a lineage of technologies that externalized human capacities, from Mesopotamian symbols through printing, the calculator, and factory machinery. It then borrows Plato’s warning in Phaedrus that access to an answer is not possession of knowledge, and unlike a calculator, a language model integrates emotion. Then, the paper reports on how financial firms use AI through three cases, with Bank of America adopting AI late but broadly under regulatory constraint, OMERS adopting AI early from the junior deal team upward, and UBS equity research using the technology to compress earnings-season work. A section on provenance, built on Norges Bank’s concern that a model’s account of the world carries its makers’ interests, argues that accountability cannot be delegated to the tool. The second half of the paper asks what skills hiring managers are searching for in prospective professionals: people skills, hard work, fluency with AI without deference to it, adaptability, and active listening for relationship cultivation. The paper closes by applying central ideas across research, wealth management, banking, asset management, and private markets, concluding that AI now does the procedural work while a person must still decide which facts matter and whether an action is prudent, permissible, and defensible.

The evidence is qualitative practitioner testimony from firm visits in August 2026, during An Organized Walk Down Wall Street, an experiential course offered by the University of Denver’s Daniels College of Business. These conversations are supplemented by Federal Reserve speeches, a National Institute of Standards and Technology (NIST) framework, two OpenAI documents, and one Stanford working paper on entry-level employment.

Innovative Technology Shifts Human Attention

Human societies repeatedly develop technologies that externalize specific human capacities and, in doing so, alter the institutions and standards of judgment through which those capacities are exercised. Early writing systems in Mesopotamia, attested by the late fourth millennium BCE, externalized memory and recordkeeping into durable symbolic form (Encyclopaedia Britannica, 2025, n.d.-a). Printing, first developed in East Asia and later mechanized in Europe by Johannes Gutenberg in the mid-15th century, expanded the reproduction and distribution of knowledge (Encyclopaedia Britannica, 2026b, 2026d, 2026e). Mechanical calculators in the 17th century, including Blaise Pascal’s Pascaline (1642) and Gottfried Wilhelm von Leibniz’s Step Reckoner (designed in 1671 and built in 1673), externalized portions of arithmetic procedure (Encyclopaedia Britannica, 2008, 2026a). Industrial machinery of the 18th and 19th centuries, especially under the factory system and steam-powered production, externalized forms of physical labor and reorganized production at scale (Encyclopaedia Britannica, 2026c, n.d.-b).

The foundations of artificial intelligence emerged through 20th-century developments in mathematical logic, computation, cybernetics, statistics, and information theory. The formal field of artificial intelligence was established during the mid-1950s, while later gains in computing power, data availability, neural networks, and optimization produced modern machine-learning systems. Large language models (LLMs) brought machine learning into the medium of ordinary language. Paul Marcussen, Head of Norges Bank Investment Management’s New York Office, called large language models “a generational, big-tech breakthrough” (P. Marcussen, personal communication, August 25, 2026). Moreover, AI is a technology capable of compressing the time between an intention and its realization, so that whatever a person can clearly articulate and verify becomes a deliverable, effectively shifting where human attention must go.

The External Record of Knowledge

In Phaedrus, Plato suggested that the endeavor of writing words posed a serious risk to the individual. His worry was that an individual could possess an external record of knowledge, and therefore feel enlightened, while not undergoing the development essential to understand the piece of knowledge in a method that could be examined, defined, and inherently adopted. This idea is relevant to generative AI because it resembles the distinction between access to an answer and possession of knowledge (Plato, n.d.; University of Michigan, n.d.).

The distinction surfaced at Vontobel Asset Management during the week of the visits, when a colleague defended AI-drafted writing as “like a calculator.” Markus Hansen, the firm’s Portfolio Manager for Europe, rejected the analogy: “No, because two plus two is always four. You can arrive at the answer in different ways . . . But math is math. An opinion has some element of emotion to it because you’re making a statement” (M. Hansen, personal communication, August 25, 2026). A calculator externalizes a defined arithmetic procedure toward a determinate answer; a language model can retrieve, summarize, recombine, and articulate information, but it does not independently establish the soundness, relevance, or economic consequences of the position it produces. In investment research, that difference separates the efficient production of language from the accountable formation of judgment. Generative-AI systems can also produce confidently stated but erroneous content, making source validation and human review especially important when output may influence regulated or consequential financial decisions (NIST, 2024).

How Wall Street Firms Are Deploying Generative AI

Institutions create economic value by addressing problems that matter to their customers. As technology, preferences, and market conditions change, sustaining that value requires reassessing whether an established solution still serves the underlying need. When an organization directs its attention primarily toward preserving an established solution, its operating model can become rigid, maladaptive, and obsolete. The truly resilient institutions hyper-focus on the problem; the offered solution is then malleable and resilient to economic happenstance and marketplace alterations.

Across financial firms, AI is primarily used to search and synthesize research, summarize filings and transcripts, monitor news and press releases, assist with drafting and coding, organize internal knowledge, translate natural-language questions into structured queries, and accelerate initial modeling or document-review work. David Pearson, Vice President of Investment Banking at Goldman Sachs, stated: “the people who use AI will outcompete the people in the same role who do not” (D. Pearson, personal communication, August 28, 2026). Institutions that refuse to integrate AI may impose unnecessary research, coordination, and administrative burdens on their professionals while allowing less constrained competitors to improve their speed and scope.

Three Use-Cases: Bank of America, OMERS, and UBS

Bank of America has disclosed broad internal adoption of Erica for Employees (Bank of America’s virtual assistant) and reported reduced technology-service-desk calls, while its generative-AI deployments have extended to customer service, employee support, data analysis, and internal workflows (Bank of America, 2025, 2026). Its front-office personnel described the potential for AI to streamline research triage and client preparation but also emphasized the role of regulatory supervision and monitored communications, with intentionality of creating synchronicity in company information. Federal Reserve officials have emphasized that banks remain responsible for managing the risks created by their use of AI and their relationships with external technology providers (Barr, 2025; Bowman, 2026). Sybil Walker, the Vice President of Equity Sales who hosted the visit, described the sequencing candidly: “The big banks are always last, basically. Because we are so regulated, we are so big, we’re quite slow moving . . . We let other people experiment and then we proceed with the safest, most proven routes” (S. Walker, personal communication, August 26, 2026). Walker added, “We have Copilot, and we’re highly encouraged to play around” (S. Walker, personal communication, August 26, 2026). Adoption at Bank of America is therefore wide but deliberately conservative, favoring proven, vendor-supplied tools over in-house experimentation, and it proceeds only within the same supervisory and control structures that its front-office staff and federal regulators both emphasized.

Where Bank of America’s caution is imposed from outside the firm, adoption at OMERS, the Ontario Municipal Employees Retirement System, is ordered from within, by the structure of the deal team itself. Practitioners described expanding AI support for investment analysis while also considering how operating companies could use automation, data systems, and AI-enabled workflows to improve performance. Tyler Craig, the Managing Director who hosted the visit, stated: “We’re implementing [AI] through our deal team, starting from the junior team, letting them automate parts of the work” (T. Craig, personal communication, August 24, 2026). Bank of America adopts late but broadly, and OMERS adopts early but from the bottom, yet both firms protect their institutional judgment.

By contrast, the UBS aerospace and defense research team described using AlphaSense, Copilot, custom agents, and data-scraping tools to search prior research, monitor corporate communications, follow expert calls, and surface potential ideas. Nick Labbadia, the Associate who hosted the visit, described how quickly the firm’s posture had changed: “When I first started, AI was really, really discouraged. If you wrote anything using AI, that would have been bad. And now it’s really, really encouraged . . . It’ll help us get work done faster. But it’s still in early stages” (N. Labbadia, personal communication, August 28, 2026).

Labbadia framed the change in terms of speed, especially when an analyst must update a model, compare reported results with estimates, and determine the source of a variance. UBS is encouraging use of AI for the compressed, procedural work of the reporting cycle. The through line between Bank of America, OMERS, and UBS is that AI absorbs the procedural work that surrounds a decision, but the decision itself remains with the person accountable for it.

Provenance and Accountability: Whose Model?

Marcussen has been with the fund since 2002 and now oversees, among other responsibilities, the selection of the external managers who run part of its roughly $2.2 trillion in assets. His office had integrated language models into trading, research, and the fund’s voting of more than 130,000 proxy questions a year. He raised a question that concerns whether a model’s account of the world is true:

I find it troubling when you ask a Chinese LLM, “Tell me about Tiananmen Square,” and it says, “It’s a square outside the People’s Congress.” Not what I mean. The rewriting of history based on where the model is framed is problematic. (P. Marcussen, personal communication, August 25, 2026)

Models report a version of the world shaped by what they were trained on, and what their makers were permitted or inclined to leave out. Every model therefore has a vantage point, and the organization that built it has interests, commercial, regulatory, or national, that the user did not choose and usually cannot see. When a professional adopts a model’s output as the basis for a decision, the model’s unseen interests are built into the foundation of the decision the professional is claiming. This problematic situation is why accountability cannot be delegated to the technological tool. The Norwegian fund’s response was not to abandon the models but to keep a human at the point of decision, and to base AI adoption on which model, built by whom, is credible enough to trust.

Creating Economic Value

Every institution must distinguish its enduring purpose from the mechanism through which that purpose is presently delivered. During the Bessemer Trust meeting, large-cap Portfolio Manager Jeff Rutledge emphasized that “the skill of understanding what it is that we’re trying to solve for is really, really important now” (J. Rutledge, personal communication, August 24, 2026). Although the remark concerned analytical competence in an AI-enabled workplace, it also illuminates the strategic idea that improving a process is valuable only insofar as that process addresses a consequential problem.

What Skills Hiring Managers Are Searching for

Many incoming financial professionals are concerned about current market hiring due to anxiety-inducing evidence, like the following statistic: Employment of workers aged 22 to 25 in the most AI-exposed occupations now sits roughly 19 percent below where it would be had it kept pace with that of their less-exposed peers (Brynjolfsson et al., 2026). Given the increased concern, Labbadia, the UBS Associate, offered a hiring manager’s view. He is a recent entrant himself: a former U.S. Air Force officer who completed the Applied Quantitative Finance program at the University of Denver, joined UBS as an intern in January 2025, and had just helped interview candidates for an open seat on his team. He stated:

I think people skills are still most important. One of the guys on our team left, so we were hiring. And we wanted to hire the person that we liked the most, not necessarily the one who we thought was the best at modeling. Obviously, we want someone smart, but there are a lot of really smart candidates. [If] you can learn, work with a team, grow, be a good person, and fit with the role that you’re in, I think you can be extremely successful. (N. Labbadia, personal communication, August 28, 2026)

Fluency With AI Tools, Without Deference

Bessemer Trust described enterprise AI applications for research support, document synthesis, and administrative relief, emphasizing that investment analysis remains grounded in primary company information, conventional financial data, market research, and professional judgment. Patrick Boyle, the Managing Director responsible for hiring the firm’s client-service and business-development staff, stated:

Proficiency in AI is hugely important. And if you don’t know a whole lot about it, spend a lot of time on it and figure it out and see how it can be helpful. But I don’t think in any way that devalues the human skills. In my side of the business, I want people to like you and trust you and think you’re smart. And I don’t think AI makes you that. You have to be in the living room with the family. No computers, no AI. They’ve got to look at you. You’ve got to pass a smell test. (P. Boyle, personal communication, August 24, 2026)

For an advisor serving multigenerational families, the work is still to understand the client’s circumstances, helping the client remain committed to a suitable long-term plan. Rutledge stated: “If an analyst on the team says, well, ChatGPT says this or that financial statement ratio says we should sell the stock, I’m not ready to believe that. So, you know, prove it” (J. Rutledge, personal communication, August 24, 2026).

Defining the Problem With Responsible Ownership

OpenAI’s Modeling an AI Jobs Transition, published April 25, 2026, identifies that people will still need to “exercise judgement, take responsibility for decisions, understand unusual cases, and build relationships with colleagues, students, or customers” (OpenAI, 2026). As AI lowers the resource barriers between an idea and its execution, entrepreneurial initiative becomes an important complement to technical knowledge. During the meeting at Ramirez & Co., a family-owned investment bank founded in 1971, Bob Deibel, who builds the firm’s data and analytics tools, explained that an individual who can “think through things in a systematic way” can increasingly execute an idea without extensive supporting infrastructure (R. Deibel, personal communication, August 26, 2026). His observation suggests that professionals must identify the worthwhile problems. Each firm’s use of AI depends on these human capacities, and each firm’s control over AI is, in the end, a person exercising them.

An Adaptable Mind for the “Rapidly Changing Economy”

The laboratories building these AI systems say the same thing about their users that the financial practitioners say about their hires. In A Strong and Safe Start with AI: OpenAI’s Teen AI Literacy Blueprint (a policy document addressed to educators and policymakers in November 2025), OpenAI writes, “AI will create jobs we cannot yet imagine, and students who develop the resilience and fortitude to adapt to a rapidly changing economy will thrive” (OpenAI, 2025). The company whose products are automating entry-level analytical work is telling the next generation that the durable skill is adaptability.

Sam Ramirez Jr., Chief Executive Officer of Ramirez Asset Management, stated: “The first ten years of your career, you’ve got to really be a sponge and learn, and it makes you stronger for that second part of your career where you have a base of knowledge” (S. A. Ramirez Jr., personal communication, August 26, 2026). Walker, on the equity sales desk at Bank of America, used the same image for a shorter horizon: “The first three years, you’re just drinking from a water hose and you’re like a sponge” (S. Walker, personal communication, August 26, 2026). Given this advice, the students who will do the best in the age of AI will learn to absorb information like a sponge, while they work hard. They must also develop the resilience and fortitude to rapidly adapt to an ever-changing economy.

Communication and Active Listening

Boyle was direct about communication when he firmly stated: “I still think the ability to communicate, to speak, and to write are hugely important.” He later added, “I still think the ability to communicate one-on-one and in writing are hugely important, and I don’t think AI takes that away” (P. Boyle, personal communication, August 24, 2026). Asked how he handles politically charged conversations with clients, Boyle answered, “Part of my job is just listening to people” (P. Boyle, personal communication, August 24, 2026). Nick Labbadia also mentioned the importance of active listening:

I think the biggest skill is active listening. We talk to the companies a lot. We also listen on earnings calls; we listen to clients. Active listening’s super important to really try to understand what the company is saying. If you’re a good listener, it just helps. (N. Labbadia, personal communication, August 28, 2026)

Rutledge described the sources behind an investment view as “a combination of internal [research] generated by us using trustworthy sources, coming up with opinions, and listening to outside experts, because you can’t be the expert on everything” (J. Rutledge, personal communication, August 24, 2026).

Relationship Cultivation and Reputation Establishment

Bob Hong, who runs corporate capital markets at Ramirez & Co. after 18 years at Citigroup and Bank of America, stated:

I’m always struck by how small the world is once you develop relationships and once you get to know people. I still run into people that I met my second or third year out of college who are still in this business . . . The reputation that you develop is very important because that will follow you everywhere and will pop up in places that you never expect. (B. Hong, personal communication, August 26, 2026)

David Brackett, the Founder and former Chief Executive of Antares Capital, stated the same point: “Play the long game. Build relationships based on trust. Your integrity, to me, cuts the cord of everything. There are no shortcuts. Honor your commitments. If you say you’re going to do something, do it” (D. Brackett, personal communication, August 28, 2026). And on where those relationships are formed: “If you get the right manager in the right environment, the right team, that’s going to pay long-term dividends” (D. Brackett, personal communication, August 28, 2026).

In addition to making relationships, Chris Bond, who covers consumer companies in equity research at Stifel, told the group, “Make sure you go out of your way for help,” because “a collaborative model is vital.” Andrew Wong, a Junior Research Analyst in Fixed Income Capital Markets at Stifel, added, “Everyone wants to help each other.” Bond closed the point with “Team structure is important” (C. Bond & A. Wong, personal communication, August 27, 2026). At Vontobel’s Quality Growth boutique, Igor Krutov, the Managing Director, offered: “Collectively we are stronger than when we work individually” (I. Krutov, personal communication, August 25, 2026). Pearson, the Goldman Sachs Vice President, was asked what he will remember most from his years in the business. “The relationships,” he said. “It’s nice to see that continuation of the relationships that you establish when you’re together and fight through” (D. Pearson, personal communication, August 28, 2026).

Conclusion

Generative AI is continually changing the emphasis of value creation for financial professionals. Across the institutions examined in this paper, AI performs work that previously consumed a substantial share of junior and mid-level professional time: locating documents, reading transcripts, comparing estimates, organizing news, drafting language, querying data, and assembling a preliminary representation of a problem. However, Plato reminds the profession that information is not identical to wisdom (Plato, n.d.; University of Michigan, n.d.). Now, financial professionals must determine which facts matter, whether the evidence is trustworthy, whether the model is suitable, what the system has omitted, how uncertainty should be communicated, and whether an action is economically prudent, legally permissible, and ethically defensible.

Web Publication Note

This web version and its PDF differ from the manuscript submitted for the course in five places. Four sentences are corrected for grammar and precision: the employment comparison from Brynjolfsson et al. (2026) now states its baseline, less-exposed peers, as the working paper does; the sentence on how Norges Bank chooses a model, the sentence on resilient institutions, and the identification of David Pearson are repaired. Two entries that the text never cites, Butler (1863) and Dostoevsky (2008), are removed from the references. The argument, evidence, and quotations are unchanged.