Now that AI has made answers easy to produce, senior leaders face a new challenge: How do they develop younger employees who can quickly generate analysis but may not yet know what matters, what to trust, or what to do next?
Traditionally, work did two things at once: It produced an answer and developed the person doing the work. By researching companies, building models, and defending recommendations, junior employees encountered different situations, made predictions, and discovered where they were wrong. Over time, they learned which signals mattered and when familiar patterns didn’t apply. Experience became judgment.
I’ve watched that process from several vantage points. As an award-winning investigative journalist, I learned to interrogate financial statements and management claims the hard way, one wrong assumption at a time. That experience led me to create the AREA Method, a decision-making system for individuals, companies, and nonprofits to solve complex problems, and to found Decisive, a decision-sciences company that advises executives and organizations on navigating complex decisions. In my work teaching decision-making at Cornell University, and previously at Columbia Business School, I see firsthand how much of that judgment is built through doing the work, not just receiving the answer.
Now, with AI, the tools can produce the analysis without providing all the learning that once came from creating it. But asking employees to continue doing everything manually would give up much of AI’s value. The answer is not to preserve the old process. It’s to redesign how younger employees learn and develop while using AI.
Teach the questions that come after the answer
When AI can produce a market analysis, investment thesis, or financial model in seconds, the analyst’s most important work may begin after the output appears:
Is this right? Which assumptions drive the conclusion? Does credible evidence support them? What’s missing? What would make this recommendation fail?
A polished answer can make those questions harder to ask. The analysis arrives complete, organized, and confident. A less-experienced employee may assume that because the output looks rigorous, the underlying thinking must be rigorous too.
An experienced investor is more likely to notice the weak point: an aggressive growth assumption or a misleading comparison. Junior analysts have encountered fewer of these exceptions, so they have fewer reasons to stop and look again.
The solution isn’t to place a senior investor beside every employee using AI. It’s to build practices that give junior analysts repeated opportunities to exercise the judgment AI can’t supply for them.
Here are three ways to develop your young talent:
1. Predict before you prompt
Before consulting AI, ask the analyst to record a preliminary view:
- What do I think is happening?
- What result do I expect?
- Which variables are likely to matter most?
- What am I uncertain about?
Since you aren’t usually sitting next to the analyst when they’re doing the work, have them practice this skill in meetings where you create moments for your younger team members to bring forward their expectations. This will allow you teaching moments and an ability to model how you’ve developed your own process for anticipating the results you expect to see. This doesn’t need to become another lengthy assignment. A few sentences may be enough.
The purpose is to prevent AI’s answer from becoming the analyst’s starting point. Once the system supplies a persuasive explanation, it can anchor everything that follows. The analyst begins reacting to its frame rather than developing an independent one.
A prior prediction also creates a useful learning moment. If AI reaches a different conclusion, the analyst must examine the gap. Did the system identify evidence the analyst missed? Did it make an assumption that doesn’t fit the situation? Which view is better supported?
The comparison, not simply the prediction, is where learning occurs.
2. Evaluate the output
Producing an investment case and evaluating one are different activities. Organize the workflow accordingly.
After AI generates an analysis, require a separate evaluation pass:
- Which assumptions have the greatest influence on the recommendation?
- What evidence supports each one?
- Which evidence is independent, and which sources may be repeating the same information?
- What relevant information is absent?
- Under what conditions would the conclusion stop applying?
- Who would make the strongest case against it?
As a leader, you likely already use these types of questions to interrogate your team’s analyses. Let them know that they can do the same sort of investigative work on AI’s output to develop their critical thinking and judgment muscles.
For consequential decisions, ask another team member to evaluate the output without seeing the original prompt. The wording of a prompt can frame both the question and the range of acceptable answers. A reviewer who sees that framing may inherit it without realizing it.
Separating generation from evaluation creates distance between the machine’s analysis and the human judgment applied to it. It also teaches employees that a polished recommendation is an object to investigate, not a conclusion to agree with.
3. Save your ‘bad AI’ examples
Investment firms already document performance and study decisions that went wrong. They can use the same material to accelerate the development of judgment.
Showcase your ‘bad AI’ examples where AI produced something plausible but wrong, incomplete, or inappropriate to your younger employees. How did you know that your AI tool made a mistake? What was the red flag or weakness that you noticed and how did you fix it?
The most valuable detail isn’t simply what the system got wrong. It’s the clue that could have alerted the person reviewing it:
- A risk disclosed in a footnote but missing from the recommendation
- A return estimate that ignored contradictory information
- Several apparently independent sources that relied on the same underlying data
Turn these examples into short exercises. Give analysts the original output, ask what they would question and, only then, reveal the outcome. Over time, these case studies become a record of the firm’s experiences with AI to train newer employees to learn from experiences they haven’t yet had personally.
Teaching judgment in the Age of AI
AI gives junior analysts access to more sophisticated capabilities earlier in their careers. That can accelerate their contribution, but it won’t automatically accelerate their judgment.
This is the central argument of my new book, The Human Edge: Smarter Decisions in the Age of AI (Cornell University Press): As AI takes over more of the work of generating answers, the competitive advantage shifts to the quality of the human thinking applied to them. Nowhere is that more true than in how firms bring along the people who will one day be making their most consequential calls.
To achieve both contribution and judgment, investment firms must become more deliberate about development. The new apprenticeship shouldn’t recreate every manual task AI can perform. It should preserve the experiences that teach people when to trust an answer, when to challenge it, and when they don’t yet know enough to decide.



