There’s a distinction I keep coming back to that I think matters more than most people give it credit for.
Knowledge and intelligence are different things. Aristotle drew this line precisely: episteme is systematic knowledge of how things are; deinotes is raw cleverness, the ability to look at a situation and figure out what to do. You can be deeply knowledgeable and not particularly clever: you’ve accumulated facts, frameworks, credentials, expertise, but when the situation changes, when the map no longer matches the territory, you don’t know what to do with any of it. You can also be highly clever and profoundly ignorant: quick to reason, good at pattern matching, but operating on a thin base of actual understanding. Both conditions are more common than we like to admit. Knowledge and intelligence are both necessary, but neither one alone is sufficient for making good decisions under genuine uncertainty.
The interesting question is what you need beyond both of them. Building a company requires decisions in this category: knowledge tells you what is known, cleverness helps you reason about what isn’t, and neither one tells you what to do.
Aristotle had a word for what sits beyond both. He called it phronesis, usually translated as practical wisdom, and he was careful to distinguish it not only from theoretical knowledge (episteme) but from cleverness (deinotes). The distinction matters. Cleverness is morally neutral: Aristotle pointed out that a thief and a judge can be equally clever. What separates them is not the quality of their reasoning but the character that directs it. Phronesis is what happens when intelligence is guided by honesty about what you know, what you don’t, and what is actually worth doing. It can’t be acquired from a textbook or a framework. It develops through years of acting under uncertainty and paying close attention to what you got wrong.
This is not just an intellectual virtue. Aristotle tied phronesis directly to moral character: you cannot have practical wisdom without integrity, because self-serving reasoning will corrupt every judgment, no matter how sophisticated the analysis behind it. And you cannot have integrity without practical wisdom, because good intentions without sound judgment produce well-meaning failures. The two are inseparable. The practically wise person isn’t just smart. They’re honest, and their honesty is structural, not performative.
I think about this a lot, because I’m building a venture studio, and the honest version of that sentence is: I’m building an institution in a space where the evidence for what works is thin, the feedback loops are slow, the attribution of outcomes to causes is genuinely unsolved, and the incentive to construct confident narratives far outweighs the incentive to acknowledge uncertainty. I’ve written about these problems at length.
I also know that phronesis, as Aristotle understood it, is not available to someone analyzing a problem from the outside. It is earned through years of operating in exactly the conditions that make it necessary: ambiguity, incomplete information, real consequences. I’ve spent twenty-five years building companies, managing teams, and making decisions that often turned out to be wrong, but I’ve tried to pay attention to the “why” along the way. That experience doesn’t guarantee wisdom. But phronesis without experience is, for Aristotle, an impossibility.
So why build?
The first wrong answer is certainty. “I’ve done the research, I’ve studied the model, and I know this works.” If I genuinely believe that studios face an unsolved attribution problem, that the feedback loops are too long and too noisy to generate reliable knowledge (so far!) about what drives outcomes, then claiming certainty about my own studio would be either dishonest or un-self-aware. The epistemic scrutiny has to apply to my own model first, not only to everyone else’s.
The second wrong answer is paralysis. Interrogate any model thoroughly enough and you can always find a reason not to act. The uncertainty is real, the confounders are real, the epistemological problems are genuinely hard. A person could spend a career mapping those problems and never build anything, and the mapping would be intellectually defensible at every stage. But phronesis isn’t contemplative. It doesn’t end with analysis. It ends with action, specifically the kind of action that incorporates the analysis rather than ignoring it.
The space between those two wrong answers is where I think the real work happens. Not certainty, not paralysis, but structured conviction: acting with full awareness of what you don’t know, and building that awareness into the architecture of what you’re building rather than papering over it with a confident pitch. In Aristotle’s terms: certainty is deinotes masquerading as phronesis, cleverness that has convinced itself it is wisdom. Paralysis is episteme refusing to become phronesis, knowledge that won’t accept the responsibility of action. The practically wise person lives in the space between, acting without certainty but not without judgment.
The first response to the attribution problem was methodological. If you can’t know with certainty whether your process produces outcomes, you can at least structure your decisions to be legible to yourself.
The epistemological problems I identified in earlier writing all share a common feature: the reasoning is invisible. VCs construct post-hoc narratives. Studios attribute outcomes to bundles they never decompose. The beliefs and the evidence get entangled in ways that make self-correction impossible. Structured evaluation doesn’t eliminate bias. But it makes bias observable, which is a prerequisite for correcting it. In a sense, the system substitutes institutional accountability for personal virtue. That’s an imperfect trade, but it’s the best available in a domain where the consequences of unexamined cleverness are measured in millions.
The second response was financial. Each stage of the studio’s process is structured as a real option: a bounded investment that purchases the right, but not the obligation, to invest further. A concept that passes ideation screening earns a small allocation for rapid validation. A concept that survives validation earns a larger allocation for MVP construction. At every gate, the question isn’t “will this succeed?” It’s “is the cost of learning what we need to learn at the next stage justified by the expected reduction in uncertainty?”
The logic here is directly informed by the uncertainty I’ve written about. If I can’t know at Stage 0 whether a concept will work, then the rational response isn’t to avoid commitment. It’s to structure commitment so that the cost of being wrong is bounded and the information gained from each stage improves the next decision. The option architecture doesn’t require confidence about outcomes. It requires confidence about the learning rate, that each stage produces enough information to make the next gate decision better than the last.
Frank Knight drew the distinction between risk, where probabilities can be estimated, and uncertainty, where they can’t. Early-stage venture creation sits firmly in Knightian uncertainty. The real options framework doesn’t pretend otherwise. It acknowledges that the probability estimates at each gate are subjective, imprecise, and likely wrong. What it offers is a structure for acting anyway, with the downside capped at each stage and the upside preserved through the option chain.
The third response was institutional, and it’s the one I think matters most.
The Quine problem I described in the attribution essay, beliefs forming a web where the core is protected by adjustable periphery, is a structural feature of organizations, not a character flaw of individuals. Studios, like all institutions, will naturally absorb confirming evidence and deflect disconfirming evidence. The question isn’t whether this will happen. It’s whether the institution is designed to counteract it.
This is where Aristotle’s insistence that phronesis requires moral character becomes operational. An individual might choose honesty over self-confirmation through personal integrity. An institution cannot rely on the character of any single person. It has to be designed so that the honest answer is structurally easier to reach than the comfortable one. The recalibration mechanism, the graveyard database, the recorded kill decisions: these are not just analytical tools. They are the institutional equivalent of the moral virtue that Aristotle said phronesis depends on. They make self-deception harder: if you advance a concept past honest scrutiny, the recorded scores and the eventual outcome will show it. If a dimension you overweight keeps failing to predict outcomes, the recalibration corrects for it whether you want it to or not.
The evaluation weights in the scoring system aren’t fixed. They’re recalibrated based on how well each dimension predicted actual outcomes. A dimension that consistently fails to distinguish successes from failures gets downweighted. One that proves predictive gets upweighted. Heck, the entire structure of the MAUT decomposition is even up for review. Version 1 had 48 indicators and 9 dimensions. We are now at 33 indicators in the 6 main dimensions with 4 more indicators describing an overlay for capital attractiveness. The system evolves. The system treats its own beliefs as hypotheses with expiration dates rather than convictions with permanent tenure.
The concepts that get killed, which should be the majority, don’t disappear. They’re preserved with their scores, their stage-gate evaluations, and the reasoning behind the kill decision. Over time, this graveyard becomes a dataset: a record of what the studio believed, what it decided, and what happened (or didn’t happen) as a result. It’s the raw material for the kind of self-examination that the attribution problem demands.
None of this is a guarantee. A studio can build all of this infrastructure and still be wrong about its core thesis. The self-calibration could converge on the wrong signal. The dimensions could be poorly specified in ways that the feedback loop can’t detect. The sample sizes in the early years may be too small for any of the recalibration to be statistically meaningful.
The conviction I hold isn’t first-order. I don’t have a justified belief that “this studio will produce superior returns” (although I sure as hell hold an unjustified belief in exactly that). The “justified” part is a claim that the evidence can’t yet support, and that the epistemological challenges I’ve outlined make genuinely difficult to evaluate even after the fact.
The conviction is second-order. I believe that a system designed to make its reasoning explicit, to bound its bets, to treat its own assumptions as testable hypotheses, and to accumulate structured evidence of its own performance has a better chance of learning what works than one built on confident narrative and post-hoc attribution. I’m not betting on a thesis. I’m betting on a learning architecture. I’m betting that an institution can be designed to cultivate something like phronesis: the capacity to act under uncertainty, learn from outcomes, and correct its own reasoning over time, even when the people inside it are as prone to self-deception as anyone else.
That’s a different kind of claim. It can still be wrong. The architecture could be over-engineered for problems that turn out to be less important than problems I haven’t anticipated. The feedback loops could be too slow to matter in a competitive environment that rewards speed over rigor. The whole endeavor could be an elaborate exercise in what I accused the industry of doing: constructing a more sophisticated version of the same confident self-narration, just with polysyllabic vocabulary.
I hold that possibility honestly. The difference, I hope, is that I’ve built the thing to tell me if that’s what’s happening.
I’m aware that publishing this kind of thinking carries a specific risk. An investor reading this might reasonably wonder whether I’m more in love with the epistemological framework than with the outcomes it’s supposed to produce. Whether the intellectual apparatus is the product, rather than the venture portfolio that’s supposed to be the product.
The answer is no. The portfolio is the product. The companies are the point. The epistemological infrastructure exists in service of better decisions about which companies to build, which to kill, and how to allocate resources between them. If the infrastructure doesn’t improve those decisions, it’s intellectual decoration, and I will shut it down.
But I think the opposite failure mode is more common and more dangerous: the studio that acts with full confidence, succeeds or fails, and learns nothing either way because the reasoning was never structured to permit learning. That studio may produce returns. It may even produce good returns. But it will never know why, and its “pattern recognition” will be indistinguishable from the lucky knowledge I wrote about weeks ago. Aristotle would call this deinotes: cleverness that produces outcomes but generates no wisdom. It looks like phronesis from the outside. It isn’t.
I would rather build intentionally and know what I’m learning than build fast and narrate my own success story without knowing whether it’s true.
That’s the bet. Not certainty. Not paralysis. Structured conviction, with the uncertainty built into the foundation rather than hidden behind a polished narrative.