Leveraging Human Ingenuity: A Meditation on Why AI Is No Different Than Any Other Technological Advancement

Throughout human history, using tools is what has in large part differentiated us from other species, and allowed us to evolve into being increasingly intelligent.

I'm no historian, but it makes logical sense that one of humanity's earliest regime-shifting discoveries was fire. One theory is that by leveraging fire to cook our foods, we evolved into a more intelligent species capable of allocating less of the energy collected from our nutritional sources to digestion and instead spending it on a much higher leverage organ system, our system of cognition. Simply by inventing the tool of fire, humanity did not progress. Humanity only progressed by using the tool of fire in its optimal application, not to burn down villages, not to cause harm, but to cook food and provide more habitable shelter so that energy and resources could be used in more sophisticated and productive ways, which in turn led to the flourishing of more human ingenuity.

That's the flywheel. Leverage previous learnings to create novel tools, then use those novel tools along with more human ingenuity and previous learnings to create even more novel tools. It's why many species on Earth have lived very statically for hundreds of millions of years, while in the last few hundred thousand, humans have completely revolutionized — a large number of times — what they focus on, spend their energy on, and care about. We are wired to build levels of abstraction in a way no other species can. Those abstractions have evolved from using sparks to cook food, to language that allowed cooperation at scale, to writing that transcended individual memory, to printing that democratized knowledge, to industry and digital technology — and now to systems that are commoditizing intelligence itself by compressing the entire corpus of human knowledge into an intuitive and convenient interface.

AI sits on that same exponential curve. It is not a break from the pattern. It is the pattern.

And because it's no different, the same rules apply. The question was never whether to use the tool — it’s whether the person holding it knows what they’re doing.

The lever of leverage.

We're all aware that those who tend to properly leverage technology throughout human history have a higher expectation of success. Economists cite this constantly, so we can assume it's well understood: those who properly leverage the most cutting edge technology are going to be more productive than those who do not. But "properly" is carrying all of the weight in that sentence. This is not the same as "those who use technology" or "those who trust technology," since naively wielding a powerful tool can often lead to disaster.

Many ambitious but inexperienced early humans surely burned down their shelter through the misuse of what was at the time their most powerful technology. The pattern hasn't changed — just the tools. Today it's the person who takes a pill because it exists, not because they understand what it does or whether they need it. It's the person who rents a tractor trailer to dig a hole in their backyard for a tomato plant. The teenager handed the keys to a sports car when they've never driven anything faster than a sedan. The destination may be the same, but the risk profiles are vastly different based on the driver. And it's today's aspiring "SaaS founder" who lets AI generate their entire codebase, ships it to production, and then has no idea why their users' data is leaking or their system is architecturally unsound.

Put simply, technology is leverage, but just because there exists access to leverage, doesn’t mean it’s going to be used properly or should be used at all. A longer wrench can always produce more torque per unit of input force than a shorter one. For the skilled craftsman, access to more torque is always likely to be something genuinely useful. However, for an unskilled laborer, the longer the wrench, the more likely a destructive outcome becomes.

One of the most elucidating phenomena I've been exposed to as an entrepreneur working with so many different types of people is how common it is for someone to be the biggest fish in their particular fishbowl and not realize how small the bowl is. The overconfidence that comes from local expertise — from being the most technical person in a room that lacks technical depth — is precisely what leads to destructive applications of technology. For whatever reason, somewhere along my education this really stuck with me, perhaps because I was fortunate enough to recognize that I could very much be susceptible to this behavior, given my general breadth-first approach to collecting skills and my enthusiasm for learning new things. That awareness — that you might not know what you don't know — is the primary thing that must be present in order to actually materialize the higher expectation of success that comes from applying cutting edge technology.

This does not mean someone should be barred from using a particular technology if they don't have an advanced degree in the underlying driver of what makes it possible. But it does mean that, for mission critical applications, one must be candid with oneself when evaluating whether the application of a particular technology is likely to generate a net benefit or not. Note the emphasis on mission critical — I'm a strong believer that if no one can be hurt, the best way to learn is by doing, failing fast, and iterating.

Competence multiplier or competence mask: a practical litmus test

A practical litmus test for whether a technology will produce net benefit is whether it functions as a competence multiplier or a competence mask.

When a tool makes someone faster at something they already understand — something they can evaluate with their own judgment and proven checks — it’s additive or multiplicative in the best way. It compresses time and raises quality without changing the first-principles-driven ground truth: they still know what “right” looks, feels, and smells like — and where to try to break it if they’re unsure.

But when the tool feels like magic, unlocking capabilities they can’t independently validate, the risk profile flips. In that case they’re not gaining power — they’re taking on leverage. And leverage doesn’t just amplify outcomes; it amplifies hidden errors. The danger isn’t that people use advanced tools. It’s that tools can raise confidence faster than they raise the ability to detect when they’re wrong.

If someone falls into the latter category for their desired application of AI (or, as of the advent of AI, any technology), all hope is not lost. The democratization of information in the digital era has given society access to the most accessible, tireless, and patient tutors there have ever been. The world today offers anyone a highly compressed path to truly understanding anything from first principles, as long as they are honest with themselves about how far down the knowledge graph they need to go in order to actually be comfortable applying a particular technology to a specific area of interest.

This is the difference between prompting AI to "do [this]" and prompting it to "teach me how to do [this]." The first only works if you can already evaluate the output. The second is where the real value lives — and if the AI's explanation surfaces topics you're not yet familiar with, those gaps are your next prompts. It's the same distinction that separates an effective manager, who leverages human capital to amplify what they already understand, from an ineffective one, who uses it to conceal what they don't.

The value is in systems orchestration.

I’ve never thought of myself primarily as a “coder.” Code is just one of the levers. It’s a commodity now, and increasingly so. The scarce thing was never the code itself. What I’ve always been drawn to is systems: taking inputs, transforming them through a set of constraints, and producing outputs that can be measured against reality.

Early in my career, right when I left my first real job to start my first company, I used to tell people my best skill was that I’m really good at Googling. What I meant was more specific. I was good at asking the right questions, and just as importantly, at knowing when an answer was signal versus noise — whether it came from an academic research paper, YouTube video, a blog post, an argument on Stack Overflow from 2006, or a person in front of me. That skill, applied relentlessly, is what let me build and scale a company to over $200M in revenue before I turned 27.

I’m not going to pretend there’s a mystical trick for that. There isn’t. There are just habits that reduce the chance of building something naive, brittle, or flat-out wrong.

The biggest one is this: I keep breaking problems down until they become checkable. When a problem is still too big to verify, it’s too big to “solve.” So, by definition, it’s decomposable — sometimes painfully — into smaller sub-problems that can be reasoned about from first principles, or at least translated into claims with observable consequences. If a claim can’t be tested, cross-checked, or falsified in any practical way, I treat it as a hypothesis, not a foundation.

That instinct comes from being a tinkerer long before I had language for it: curiosity as a reflex, the itch to pull things apart and see what makes them work. Later, I studied engineering for my undergrad at Rutgers, which formalized the same impulse into a disciplined way of thinking. Engineering wasn’t just equations and theory, it was a repeated lesson in constraint thinking: define the system, state assumptions, isolate variables, and build systems where all inputs remain consistent. I don’t do mechanical or aerospace engineering day-to-day, but I use the engineering posture constantly: clarity first, then structure, then verification.

Entrepreneurship rewards this because it forces you into unfamiliar domains on a weekly basis. You don’t get the luxury of becoming “the world’s leading expert” before you act. But you also can’t afford to move fast in a way that creates hidden landmines. The compromise is a kind of applied epistemology: learn fast, yes — but verify faster. Most failure isn’t caused by lack of intelligence; it’s caused by confusing plausible with true, and then building on top of it.

This is where AI becomes a superpower for me, not because it replaces judgment, but because it compresses the iteration loop. AI can generate options, explanations, drafts, and code at absurd speed. That’s incredibly valuable if you already have a feedback loop: if you know what constraints matter, what failure looks like, what tests should pass, and what “good” smells like when you try to break it.

Without that loop, AI is a different kind of leverage: it can produce confident nonsense faster than a human can, and it can tempt you into shipping something you don’t actually understand. So I treat AI the way I treat any high-leverage tool: I use it to accelerate decomposition, research, and implementation, but I don’t outsource correctness. When I’m unsure of the optimal approach, I ask it for multiple angles. I force assumptions into the open, and I keep bringing everything back to reality: unit tests, small experiments, sanity checks, and constraints that don’t care about what the statistically most likely next token is — only what remains internally consistent and coherent within the operating environment.

In other words, the scarce skill isn’t writing code. The scarce skill is orchestrating a system that stays true under pressure. Code is abundant. Compute is abundant. Even “answers” are abundant. What’s not abundant is the ability to define the right question, design the right iteration loop, and know what must be true before you trust the output.

AI is the ultimate tutor if you learn in a trustless way.

If there is anything meaningfully “new” about AI, it’s not that it can produce answers. The internet could already do that. It’s that it can behave like a personalized tutor: infinite patience, low friction, and an ability to rephrase an idea ten different ways until it clicks. That alone is a regime shift for anyone willing to learn from first principles.

But it comes with a trap: people confuse fluency for truth. In the old world, we outsourced trust to credentials: teacher, professor, “industry expert.” And even then, we all know credentials don’t guarantee correctness. Humans are wrong constantly, sometimes subtly, sometimes catastrophically. The difference now is that AI makes this failure mode scalable: it can be wrong with confidence, on demand, in any domain, at any speed.

So the skill that matters is not “trusting the tutor” or even knowing exactly what makes the tutor work (although understanding the basic principles of LLMs¹ is great). It’s learning to operate without needing trust at all. The goal is to turn authority-based knowledge into verification-based knowledge — claims you can stress test, triangulate, and falsify.

Practically, that means treating AI like a high-powered collaborator whose output must earn its place:

• Ask it to state assumptions explicitly.
• Ask it for counterexamples and failure modes.
• Ask it to propose tests or sanity checks.
• And keep iterating until the model’s story survives contact with constraints and reality.

Once you can do that, AI becomes what it should be: an accelerator for understanding and execution, not a substitute for judgment. The point isn’t to blindly believe the output. The point is to use the tool to get to something verifiable, then build on that with confidence. The flywheel of human ingenuity hasn’t stopped. Rather, due to its exponential nature, the tools have just gotten more powerful, and the same principles that determined success with every previous tool still apply. Those who thrive won’t be the most frantic adopters of the next trend. They’ll be the ones who apply sound principles to useful tools — carefully, correctly, and consistently. That’s the formula that turned a spark into civilization — and it still works.

Footnotes

1. A large language model (“LLM”) is a sophisticated mathematical function that predicts what word comes next from any piece of text. (3Blue1Brown)