What MIT Sloan Students Should Know
By Mike Chen
Adapted for MIT Sloan CDO. Based on a longer practitioner paper on agentic AI, workflow redesign, and the future of asset management. This version is written for a broader Sloan audience, with additional emphasis on career implications for students.
Artificial intelligence is already reshaping finance, but not quite in the way many people first expected. The early conversation centered on whether AI could build better prediction models, discover new alpha signals, or automate pieces of analysis. Those questions still matter. But the more important shift is broader and more operational. The next wave of advantage in finance is likely to come less from a single breakthrough model and more from redesigning research, risk, and operating workflows around AI + human collaboration.
In other words, the unit of transformation is shifting from the model to the workflow. That distinction matters because most of finance is not a sequence of isolated questions. It is a chain of tasks: gathering evidence, comparing conflicting signals, revisiting prior judgments, documenting decisions, escalating risks, and deciding what deserves attention or capital. Historically, much of this work has been done in a highly person-dependent way. Strong organizations have strong people, but even great teams are constrained by memory, bandwidth, and the fact that much of the work remains sequential. AI changes that. Properly designed, it can make workflows more repeatable, auditable, scalable, and continuously monitored, while leaving humans in control at the points where judgment and accountability matter most.
From tools to workflows
The first visible wave of AI in finance was largely assistive. Systems summarized documents, drafted commentary, cleaned notes, helped write code, and answered questions. Those are useful tools, but they are still point solutions. They improve isolated tasks one at a time.
The next stage is different. AI systems are increasingly able to work across multiple steps: searching, comparing sources, preserving memory, drafting structured outputs, calling tools, routing tasks, and escalating when needed. That makes it possible to redesign workflows rather than merely accelerate one step in a workflow. In asset management, that might mean a quant research process in which one system reads a paper, another maps the required data to internal sources, another drafts replication code, another runs robustness checks, and another assembles a decision memo for a human researcher or PM. Or it might mean a fundamental workflow in which an agent tracks company developments continuously, prepares earnings notes, compares management language through time, and flags inconsistencies before an analyst meeting. Or a compliance workflow in which routine checks, commentary, and exception packs are drafted automatically, with humans stepping in only where policy interpretation or risk judgment is required.
This is what I mean by industrialization. The process becomes more designed, more instrumented, and more scalable. But that does not mean humans disappear. It means humans move to the highest-value parts of the chain.
Why finance is a natural but hard domain for autonomy
Finance is a natural place for AI to create value because there is so much structured work around judgment. Research preparation, paper replication, monitoring, mandate checks, portfolio diagnostics, commentary drafting, and exception management are all decomposable tasks. Much of the surrounding work is repetitive, high-value, and still too dependent on human bandwidth. That makes finance a natural domain for workflow automation.
But finance is also one of the hardest domains for autonomy. In software engineering or many enterprise workflows, the quality of an answer is often observable fairly quickly. In investing, feedback is delayed, noisy, and often inseparable from luck. A recommendation can look smart for the wrong reason, or look wrong in the short run even when the underlying logic was sound. The historical sample is short relative to the number of hypotheses one can test. Regimes shift. Other investors adapt. Implementation frictions matter enormously. A result that looks elegant in research can disappear once transaction costs, constraints, liquidity, turnover, crowding, and portfolio interactions are taken seriously.
That is why finance is not just prediction. It is prediction under costs, under constraints, in live portfolios, at scale. It is also a domain with plural objective functions. Most institutions are not simply maximizing unconstrained expected return. They are balancing alpha, tracking error, drawdown risk, liquidity, diversification, taxes, mandates, sustainability commitments, and client explainability. A technically impressive AI output can therefore still be institutionally wrong.
A useful analogy is Tesla’s Full Self-Driving. A semi-autonomous system may work impressively most of the time, but one rare failure can still be severe. In driving, one mistake can be fatal. In finance, one rare mistake is less likely to kill someone, but it can still be extremely costly: a hidden data leak, an unsupported recommendation, a compliance miss, a crowded position, or an explanation that is polished but wrong. The real risk is not just failure. It is complacency. The most dangerous system is often the one that works well enough to induce trust before it has fully earned it.
That is why I do not think the future of finance is self-driving portfolios in any unconstrained sense. The more realistic future is governed AI + human collaboration, with more of the surrounding workflow industrialized, and with humans retaining authority at the points where capital, mandates, and reputation are actually on the line.
What this means for finance careers
Many finance jobs will not disappear, but they will be reconfigured. A junior analyst who once spent hours assembling information manually will increasingly be expected to spend more time interpreting, challenging, and deciding. A quant researcher will still need technical depth, but will also need to design workflows, evaluate AI-generated work, and distinguish statistically interesting output from economically meaningful insight. A risk or compliance professional will likely spend less time on repetitive reconciliation and more time on interpretation, escalation, and governance.
That means the premium will rise on people who can do three things at once. First, they can work productively with machines – not just by prompting them, but by structuring tasks, checking outputs, understanding failure modes, and knowing when to trust and when to override. Second, they can exercise judgment under uncertainty. In an AI-rich environment, the scarcest resource is no longer raw output. It is the ability to decide what matters, what is robust, what is missing, and what deserves escalation. Third, they can communicate clearly across functions. The future finance professional will increasingly sit at the intersection of investing, technology, and governance.
How MIT Sloan students should prepare
First, build technical literacy. Students do not need to become frontier-model researchers to thrive in this world, but they do need to understand how modern AI systems work at a practical level: what large language models are good at, where they fail, how agentic workflows differ from simple prompting, and why evaluation matters. You do not need to be a model engineer, but you should be able to use AI systems intelligently and critique them credibly.
Second, build real domain depth. In finance, shallow AI plus shallow finance is not a durable combination. Students should still build expertise in markets, accounting, valuation, portfolio construction, risk, incentives, and market structure. The people who benefit most from AI are usually the ones who already know what a good question looks like.
Third, think in terms of workflows rather than just tasks. Ask where the bottlenecks are. Where is work repetitive but high-value? Where is judgment under-supported? Where are decisions poorly documented? Where are humans overloaded? The future edge in finance will not come only from better models. It will also come from better operating design.
Fourth, practice skepticism. In an AI-rich world, the scarce resource is not output. It is trustworthy output. Students should stress-test results, not just generate them. What assumptions are carrying the answer? What would break in production? What looks statistically interesting but economically empty? In many settings, the best use of AI will be to widen the funnel of analysis. The final edge will still come from deciding what survives scrutiny.
Fifth, develop communication range. Finance professionals increasingly need to move across investing, technology, and governance. The people who rise fastest will be able to translate technical possibilities into business workflows, and technical risks into institutional controls.
The bigger opportunity
This should be an encouraging moment for students rather than a discouraging one. The future of finance is unlikely to be fully autonomous. It is more likely to be agentic, governed, and collaborative. That favors people who can combine analytical rigor, technological comfort, and mature judgment.
The firms that win will not be the ones with the loudest AI marketing. They will be the ones that widen the funnel of ideas, cheapen experimentation without cheapening standards, tighten the loop between research and live decision-making, and keep human accountability exactly where it belongs. The students who stand out will be the ones prepared to help build that future.
Key line to remember: The future of finance is unlikely to be fully autonomous. It is more likely to be governed AI + human collaboration.
About the author
Mike Chen is a senior quantitative investor at Robeco Institutional Asset Management and a lecturer at MIT Sloan, where he teaches on AI and finance. His work sits at the intersection of quantitative investing, machine learning, and institutional asset management, with a particular focus on how AI + human collaboration may reshape research, investing, risk, and operating workflows.
