Down the Rabbit Hole (On Purpose)

A couple of weeks ago, a student told me I had a habit of going down rabbit holes.

The class was supposed to be about scaling an e-grocery business, and somehow we’d ended up talking about how to think about the right time to join a startup, and before that, about the discipline of prioritizing when everything feels urgent.

I could see the mild frustration: Can we get back to the case?

I understood the complaint. If I’m completely honest, time management is not my greatest strength in the classroom. But there is a difference between being unprepared and being open to the messy reality of the subject. I script my classes from beginning to end, but I intentionally build in the margins to explore these unstructured detours when a student’s question opens the right door. I knew something the student didn’t yet: the rabbit holes were the lesson.

This is the tension I want to explore.

When students show up to a classroom, they believe they’re there to learn a subject and pick up some methods. And they are.

But they’re also absorbing something far more challenging to name, something I’ll call wisdom, a word that sounds pretentious even as I write it, the accumulated judgment, instinct, and perspective that a teacher brings to every tangent, every aside, every story that seems only loosely connected to the syllabus.

My claim is simple:

Education is as much about wisdom and mindsets as it is about domain expertise and methods.

And in the age of AI, that claim has never mattered more.

Domain Expertise: The “What”

Let’s start with the most visible layer of education: domain knowledge. This is the subject matter itself. If you’re taking a course on digital strategy, you expect to learn about network effects, platform dynamics, and competitive positioning. If you’re studying operations, you want to understand supply chains, capacity planning, and inventory models. Domain knowledge is the content, the facts, the body of accumulated understanding in a field.

It matters enormously. You cannot think well about a problem you don’t understand. A student who doesn’t know how unit economics works in an e-grocery business will struggle to reason about its viability. Domain expertise gives you the raw material: the vocabulary, the landscape, the precedents.

But here’s the thing: domain knowledge is increasingly abundant. Textbooks contain it. Online courses deliver it. And now, large language models can synthesize it on demand with startling fluency. If all education offered were domain knowledge, you could reasonably ask why anyone still needs a classroom at all.

And the research on learning tells us something important here: facts are the most perishable form of knowledge. Classical work on memory by Hermann Ebbinghaus and later modeling shows that forgetting curves are typically steep early and then flatten. Discrete propositions, definitions, formulas, and dates show the sharpest early decline when they are encoded once and not revisited.

Domain knowledge, in other words, is not just abundant; left unrefreshed, it is also fragile.

Methods: The “How”

The second layer is methods: the frameworks and analytical tools that help you work with domain knowledge. Think of ROIC trees for decomposing profitability, decision metrics for evaluating strategic options, or scenario planning for navigating uncertainty. Methods are the machinery of structured thinking. They take a messy real-world problem and give you a systematic way to approach it.

Methods are powerful. They’re what distinguish a trained analyst from someone who merely knows a lot. When I teach students how to build a decision tree under uncertainty, I’m not just teaching them a technique. I’m giving them a way to see that they can carry into any domain, any industry, and any problem they’ll face in their careers.

But methods, too, have limits.

A framework can tell you how to structure a decision, but it can’t tell you which decision actually matters most right now. It can decompose a problem, but it can’t tell you whether you’re decomposing the right problem. For that, you need something else.

Methods can be more durable than raw facts, but only when they are “compiled” through practice, refreshed periodically, and used in varied contexts. Without use, skills can show substantial decay over months to a year.

Their typical failure mode is not “I forgot the steps,” but rather “I didn’t recognize when to use this method,” or “I applied it mechanically to the wrong problem.” That is a transfer problem. Research on analogical reasoning by Gick and Holyoak shows that schema induction, learning an abstract structure from multiple cases, predicts later transfer better than simply being told a principle.

If you want methods to endure, you must teach them through contrasting cases and repeated selection across varied contexts, not through a single clean example.

Mindsets: The “How to Think”

The third layer, one that students rarely recognize they’re absorbing, is what I call mindsets. These aren’t techniques. They’re orientations. Ways of approaching the world that shape how you perceive problems, how you respond to ambiguity, and how you continue learning long after the course is over.

A mindset might be the habit of asking “What would have to be true for this to work?” before committing to a plan. It might be the instinct to seek disconfirming evidence when you feel most certain. It might be a tolerance for sitting with ambiguity instead of rushing to a premature answer, or the discipline of distinguishing between what is urgent and what is essential.

I teach mindsets constantly, though not always explicitly.

When I push back on a student’s confident answer and ask them to argue the opposite side, I’m cultivating intellectual humility. When I interrupt a polished analysis to ask, “But what are you not seeing?” I’m teaching the mindset of constructive paranoia. When I tell a story about a founder who waited two years before scaling, and how that patience was the decision that made everything else possible, I’m modeling a way of thinking about timing and restraint that no framework can fully capture.

Mindsets are what enable lifelong learning.

Domain knowledge gets outdated.

Methods evolve.

But the capacity to adapt your thinking, to reframe problems, to hold complexity without collapsing it into false simplicity, that endures. Mindsets are closely related to what learning science calls metacognition: thinking about your thinking, and to the idea that education should prepare you for future learning, not just current performance.

Mindsets are not motivational slogans. They are cognitive and emotional habits that shape how you engage with uncertainty and complexity: problem-framing over answer-finding; adaptability and willingness to update; resilience and steady iteration; sensemaking in ambiguity; and comfort with being early in understanding. Much of adult professional life is not “apply formulas, get answers.” It is “decide what matters, decide what to measure, decide what to ignore, decide how to proceed when the data will be incomplete until after the decision.”

Research on learning durability confirms that mindsets can persist, but their longevity is context-sensitive. Unlike facts that decay on a steep curve, mindsets endure when they are reinforced by culture, norms, and identity, becoming part of “who I am” rather than merely “what I know”. A large national field experiment found that even brief growth mindset interventions can produce measurable benefits for specific groups, with effects contingent on school norms and opportunity structures.

The durability of a mindset is less about remembering and more about the learned disposition to treat difficulty as information and stay engaged long enough to extract feedback.

This is where certain classroom “detours” start to make sense.

A rabbit hole about how industries shift is not merely trivia; it is training for adaptability. A rabbit hole about managing uncertainty is not simply philosophy; it is practice in sensemaking. A rabbit hole about prioritization is not merely opinion; it is exposure to how experienced people compress complexity into action.

Wisdom: The “Why” and the Art of Judgment

And then there’s the layer that’s hardest to define and perhaps most valuable of all: wisdom.

Wisdom is what you get from someone who has not only studied a domain but has lived in it, who has watched decisions play out over years, who has made mistakes and learned from them, who has developed an intuition for what matters and what doesn’t that goes beyond any framework.

Wisdom shows up in the asides.

It’s the professor who pauses mid-lecture to say, “You know, the real risk here isn’t the one on the spreadsheet, it’s the one nobody in the room wants to talk about.” It’s the offhand comment about how most people overweight the first opportunity they see because they’re afraid there won’t be another. It’s the story about the executive who made the technically correct decision and lost the company anyway, because she failed to read the room.

These moments don’t fit neatly into a lesson plan. They emerge from experience, from reflection, from years of watching smart people succeed and fail in ways that textbooks don’t fully explain. They are, almost by definition, the rabbit holes.

And they’re the parts of a course that students often remember ten years later, long after they’ve forgotten the specific frameworks.

I’ve had students come back years after graduation and say something like, “I still think about that thing you said about not confusing growth and scaling.” They don’t remember which class it came up in. They don’t know the case we were discussing. But the wisdom stuck because it arrived at the right moment, in a real context, delivered by someone who had earned the right to say it through experience.

Philosophers have long distinguished technical skill from practical wisdom. Aristotle described phronesis, practical wisdom, as the virtue that helps a person deliberate well about what is good and beneficial in life, not in the abstract, but in the concrete particulars that real decisions require.

Modern scholarship echoes this: Michael Polanyi pointed to tacit knowledge embedded in experience, imitation, and practiced perception, the idea that “we can know more than we can tell”. Donald Schön argued that professionals often operate in “indeterminate zones of practice” where textbook rules are insufficient, and where learning occurs through reflection-in-action. Expertise research suggests that high-level performance usually includes pattern recognition and situational judgment beyond explicit rule-following.

Here’s a workable definition: wisdom is the ability to choose well when there is no formula, because you’ve developed judgment about timing, tradeoffs, priorities, and human dynamics.

Consider the kinds of questions students bring, especially those headed into business, startups, consulting, product, investing, or leadership. When is it smart to join a startup, and when is it career theater? How do you prioritize when every team has a compelling argument? How do you decide with incomplete data without pretending you have certainty? How do you recognize when a metric is becoming a mask? When should you push through friction, and when is friction itself information? These questions are rarely answered well by domain facts alone, nor entirely by methods. They are responded to by judgment, shaped by experience, stories, mistakes, reflection, and mentorship.

Wisdom requires time.

It requires the kind of relationship between teacher and student where tangents are tolerated, where a professor can follow a thread because their instinct says it matters, even if it doesn’t map directly to the learning objectives on the syllabus.

Wisdom is not efficient. It cannot be optimized. And that is precisely what makes it so valuable, and so vulnerable.

What Actually Endures

The learning sciences offer insights that reinforce the argument above: different learning outcomes decay at different rates, and their durability depends heavily on how they are learned, practiced, and socially embedded.

Facts, definitions, terms, and formulas are the most perishable, showing rapid early forgetting that later flattens.

Procedures and skills can be more durable, but only when they are compiled through practice and refreshed periodically; without use, they show substantial decay, though learners retain “savings” that allow faster relearning. Methods endure when learners build schemas and practice selection across contexts, not when they memorize templates.

Mindsets can persist when reinforced by culture, norms, and identity.

And tacit knowledge, wisdom, and judgment, the hardest to teach and the hardest to measure, can endure the longest of all, accruing slowly through mentorship, reflective practice, and communities of practice.

Two interventions have robust evidence for improving long-term retention: spacing (distributed practice) and retrieval practice (the testing effect). Spacing improves retention across hundreds of comparisons, and the optimal interval increases as the target retention interval lengthens. Retrieval practice has been shown to outperform restudy on delayed tests; tests are not just assessments but learning events.

These findings have practical implications: if a course only delivers content and tests immediately, it optimizes for short-term performance. If it designs cumulative, spaced retrieval and requires students to reconstruct ideas, it optimizes for endurance.

But the most durable forms of learning, the kind I’m arguing for, cannot be produced by these techniques alone. John Bransford and Daniel Schwartz proposed a refinement: sometimes the most durable learning is not “I can do the same task later,” but “I am better prepared to learn new things later”.

This is why rabbit holes can be pedagogically rational: they create differentiated experiences that make later learning possible. You may not be able to use the insight immediately. But later, when you encounter the situation in the wild, you suddenly recognize it. That recognition is part of what education provides when it aims beyond the exam.

If I compress the durability literature into one sentence, it would be this: durable learning is constructed through repeated retrieval and application across varied contexts, with feedback and reflection, within relationships and communities that shape identity.

The Risk with AI: Expertise Without Wisdom?

This brings me to what is one of the most critical questions in education today.

Artificial intelligence is extraordinarily good at delivering domain knowledge and methods. Ask an LLM to explain network effects, and you’ll get a clear, comprehensive answer (usually). Ask it to walk you through a decision tree, and it will do so patiently and correctly.

For the “what” and the “how” of education, AI is not just adequate, it’s often excellent.

But wisdom? Mindsets? These are a different matter entirely.

AI can simulate wisdom. It can retrieve relevant anecdotes, offer balanced perspectives, and even generate nuanced advice that sounds wise. But there’s a crucial difference between pattern-matching on text about judgment and actually possessing judgment. An AI has never sat across from a founder who was about to make a catastrophic mistake and had to decide, in real time, how hard to push back. It has never watched an industry shift over a decade and developed an embodied sense of what early signals feel like. It has never been wrong in a way that reshaped how it thinks.

The wisdom a teacher brings comes from lived experience, and it’s transmitted not just through words but through presence, timing, and relationship. When I tell a student that now is not the right time for them to join a startup, that advice (hopefully) carries weight because of everything behind it, the founders I’ve worked with, the patterns I’ve seen, the mistakes I’ve watched play out.

The student may disagree. But the conversation itself, the friction, the back-and-forth, that’s where something valuable gets transferred.

I recognize the privilege inherent in this argument. Relying on deep, localized mentorship and small-room friction describes a boutique educational experience. It is inherently unscalable. If wisdom requires proximity to experts, we face a massive challenge in democratizing it. But accepting that this kind of education is hard to scale does not mean we should abandon it; it means we must fiercely protect it where it exists and find creative ways to expand access to it.

The implication is not that AI is bad.

It is that AI will increasingly commoditize information and many aspects of technique. That education must therefore protect and strengthen the areas AI cannot replicate well: mindset cultivation, judgment formation, and wisdom transfer through mentorship and practice.

If we reduce education to its most efficient components, if we let AI handle the domain knowledge and methods and compress the rest, we risk producing a generation of graduates who are technically proficient but lack judgment.

They’ll know how to build a model but not when to distrust one.

They’ll know the frameworks but not the felt sense of when a framework is leading them astray.

They’ll have expertise without wisdom.

And the gap between those two things is where the most consequential decisions get made.

Toward a Balanced Future

None of this is an argument against AI in education. AI can free up time, personalize instruction, and make domain knowledge more accessible than ever. These are genuine gains, and we should embrace them.

But we should be deliberate about how we use the time AI frees up.

If that time gets reinvested in more content delivery, more efficiency, and more optimization of the learning pipeline, we’ll have missed the point.

The time should go toward more wisdom, more conversation, more mentorship, more of the unstructured, inefficient, deeply human interactions where mindsets are shaped and judgment is cultivated.

Furthermore, we can actually use AI to aid this very process. AI does not possess lived wisdom, but it can serve as the ultimate “flight simulator” for judgment. Educators can use AI to instantly generate dozens of complex, ambiguous, high-stakes scenarios tailored to a student’s specific weaknesses. AI can provide the endless repetitions needed to practice decision-making; the human educator then steps in to guide the reflection.

The durability research points to a concrete redesign.

For instructors: make retrieval and spacing a course default; teach methods through contrasting cases; assess for transfer, not just repetition; build apprenticeship moments where you model your reasoning and narrate uncertainty; and institutionalize reflection through after-action reviews and decision journals.

For students: replace rereading with retrieval; space your learning to the time horizon you care about; practice method selection, not just execution; treat feedback as information; and seek mentorship and communities of practice, because you cannot outsource judgment formation to content alone.

Students, for their part, need to learn to value the rabbit holes.

They are the curriculum, the part that can’t be Googled, can’t be generated, and can’t be absorbed from a screen.

So here’s what I’d say to any student who thinks the rabbit holes are a waste of time:

The domain knowledge will fade or become obsolete.

The methods will evolve.

But the wisdom, the hard-won, experience-tested, only-transferable-through-human-connection wisdom, that’s what will serve you when the problems you face don’t come with a case packet and a set of discussion questions.

That’s what will help you when you’re sitting alone with a decision that no framework can fully resolve, and you need something more profound than knowledge to guide you.

The rabbit holes are not digressions.

They’re the point.