The Long View: What I Predicted About AI in Finance — and What Actually Happened
Larry Cao
12/8/20254 min read
A well-respected researcher once shared a paradox that has stuck with me: “What distinguishes good research is not being the sole believer in an idea, but recognizing it early, before others confirm it.” I believe that is the essence of good research.
And over the past seven years—between 2018 and 2025—the world of AI in finance evolved fast enough for early hypotheses to either collapse under scrutiny or stand firm when the evidence finally arrived. Much of what began as theoretical framework is now measurable, documented, and operational reality.
In this overview, I revisit several predictions I made during that period—particularly predictions embedded in the FinTech Pyramid (2019), T-Shaped Teams (2019-2021), and Adapted Bloom’s Taxonomy (2023–2025). This analysis synthesizes the findings of my latest paper, Ten Structural Shifts in AI Adoption (2018–2025). What I could previously support only with logic, patterns, and field observation is now backed by empirical research, regulatory reports, and industry data.
Here is the scorecard of those predictions—and the moment the market consensus caught up.
Prediction 1: The “FinTech Pyramid”
The Call (2019): I argued that while cost, talent, and technology are merely initial hurdles, the truly scarce resources—and ultimate bottlenecks—would be Leadership Vision and Time.
When I said it: 2019 (AI Pioneers in Investment Management)
When the Market Agreed: 2024/2025
The Evidence: For years, firms obsessed over the cost of compute, talent shortages, or rapid technological obsolescence. But as inference costs dropped roughly 280-fold by 2025, the “cost” barrier has faded into the background. Suddenly, the industry realized what the Pyramid predicted six years prior: the real scarcity was management capability.
The Consensus Shift: In 2025, McKinsey Digital reported that employees are now more ready for AI than their executives are. The bottleneck had officially moved to the top of the pyramid. Furthermore, time has emerged as the critical differentiator. As NVIDIA’s Jensen Huang suggested, leaders are racing not just for technology, but against the clock: late adopters now face a compounding competitive delay, making time the ultimate constraint.
Prediction 2: T-Shaped Teams vs. Silos
The Call (First Proposed 2019): I predicted that standalone AI teams—whether in siloed labs or as Centers of Excellence (CoEs)—would fail to drive organizational change. I argued that success belonged not to isolated technical luminaries, but to “T-Shaped Teams” combining deep investment depth with organizational AI fluency.
When I said it: Concept introduced in 2019 (AI Pioneers); Framework fully elaborated in 2021 (T-Shaped Teams)
When the Market Agreed: 2024
The Evidence: Between 2021 and 2023, many institutions over-invested in specialized, disconnected AI units. This resulted in low measurable business value. This lack of business context is why we’ve seen high-profile departures of AI pioneers even from major tech and financial firms, proving that technical brilliance alone is insufficient without business context.
The Consensus Shift: Boston Consulting Group (2024) data confirmed that nearly 70% of scaling obstacles are people-related, stemming from the lack of cross-functional integration. McKinsey (2025) also realized “AI centers of excellence have reached their limits”. To achieve scale, organizations must shift “from siloed AI teams to cross functional transformation squads”.
Prediction 3: Seniority-Biased Automation
The Call (2018): I predicted AI would disproportionately automate the routine, repetitive tasks of junior analysts while simultaneously increasing the value of the judgment-heavy Portfolio Manager role.
When I said it: 2018 (AI and the Future of Investment Management)
When the Market Agreed: 2025
The Evidence: For years, the fear was that AI would replace the highest-paid decision makers. The reality was the opposite.
The Consensus Shift: The definitive proof arrived with the Harvard study, “Generative AI as Seniority-Biased Technological Change” (2025). This, combined with OECD labor data, shows a clear trend of automating lower-order cognitive tasks (processing) while creating a wage premium and preserving higher-order ones (judgment and evaluation). The market has now priced in the distinction between “processing” and “judgment” that I outlined seven years ago.
The Path Ahead: The Roadmap for 2030
The purpose of this scorecard is not to say “I told you so”; it is to highlight one critical truth: structural frameworks beat hype cycles.
Over the next few weeks, I will unpack each of these predictions in a deep-dive series—exploring the original logic, the conclusive data, and the practical roadmaps they offer for investors, firms, and careers:
T-Shaped Operating Models: The Death of the Center of Excellence and How to Build Integrated Teams.
The Junior Talent Crisis: How to Start a Career in Investment When AI Does the Entry-Level Work.
The Leadership Constraint: Why Your Vision is the Single Biggest Bottleneck to Scaling AI.
Beyond LLMs: Preparing for AI’s Next Evolution, from Neural-Symbolic Architectures to Global Regulation.
If these structural models correctly predicted the last seven years, they also provide a clear roadmap for the next seven. The time for passive observation is over. The future of value will be determined by how quickly and effectively you adapt your organization to exploit intelligence, especially as we prepare for AI beyond LLMs.
This means confronting four critical questions:
The FinTech Pyramid warns that Time is your most scarce resource. Is your leadership vision strong enough to avoid the compounding competitive delay?
The T-Shaped model suggests a broken structure. Are your investment teams designed for integrated collaboration, or are they still trapped in organizational silos?
The Seniority Bias redefines talent. Is your firm prepared to train for human judgment and governance, or are you still hiring for tasks AI can now do better?
The Emerging Frontier is already taking shape. Are you only optimizing for current LLMs, or is your technical roadmap prepared for the post-LLM AI era?
The future of finance is not a passive event. Subscribe now to receive the full roadmap series and discover the answers to these critical questions.
This article first appeared on larrycao.substack.com on December 4, 2025.