The Hard Problems Don't Need Better AI. They Need Someone Who Knows How to Aim It.
Long-term, I want to aim my methods at the biggest problems. Here's why the methodology matters more than the technology.
Start Now. Build the Trajectory.
Many hard problems today don't need better AI—they need someone who understands the domain well enough to ask the right questions and match the tools you have to the decisions that matter. But as AI gets better, the same methodology works on harder problems. So start now. Build the process. Evolve it as capabilities improve.
That conviction came from two decades of building simulations where getting it wrong had consequences. I learned early on: you don't wait for perfect tools. You start with what exists, prove your approach on real problems, and scale it as things improve.
I watched AlphaFold crack protein structure in 2020 and thought: that's what happens when someone knows their domain well enough to aim—not for maximum fidelity, but for the right fidelity for the decision. The tools were good, but the aiming was exceptional. Meanwhile, most AI money was chasing consumer applications. The real impact came from someone with domain expertise and disciplined methodology.
The hard part isn't building better models. It's knowing what to model, at what fidelity, for what decision. That discipline doesn't come from AI breakthroughs—it comes from the person asking the question. And it scales. Whether I'm using statistics, physics simulations, machine learning, or whatever comes next, the methodology stays the same. The tools change. The process doesn't.
Right Fidelity AI is how I'm testing this. I'm building on real problems with real stakes—energy grids, environmental contamination, wildfire evacuation, human longevity. Each one proves the approach works with today's tools. Each one is built to evolve as tools improve. Start now. Build what matters. Adapt as the field gets better.
Heroes vs. Harness Builders
Wissner-Gross and Diamandis draw a useful distinction between two kinds of problem solvers.
A Hero solves one problem. Brilliant, dedicated, often legendary. They develop a breakthrough therapy, design a revolutionary engine, crack a specific scientific puzzle. The world needs heroes. But heroes don’t scale. When the hero moves on, the problem-solving capability leaves with them.
A Harness Builder creates the system that lets you solve every problem in a class. They don’t cure one disease — they build the framework that accelerates discovery of cures for all diseases in a family. Instead of optimizing one factory, they build the methodology that makes any process optimizable.
The work I’m doing is harness building. The ADM methodology is the harness: a repeatable process for taking a complex, high-stakes domain and making it tractable. Not by dumbing it down, but by asking the right questions at the right fidelity. The same process applies whether the domain is energy infrastructure, environmental contamination, epidemic response, or education. The domains are different. The process of making them tractable isn’t.
The insight isn’t that these problems are similar. They’re not. The insight is that the process of making them tractable is similar. That process is what transfers across domains.
Build the system that lets you solve all problems in a class, not just the one in front of you.
Where This Is Going
There are three domains I think about more than any others. Not because they’re the most technically interesting, but because they’re the most personal.
Environmental Modeling
This is where I started dreaming as an undergrad — building models that help people understand what’s happening to the systems we all depend on. Groundwater contamination, air quality, ecosystem dynamics. These are problems where the physics is well understood but the data is sparse, the stakes are local and immediate, and the decisions are made by people who aren’t modelers. The fidelity question is sharp: how detailed does the model need to be to help a community make the right call? I’m already working in this space with PFAS contamination modeling, and it’s where I want to go deeper.
Education
A student’s learning trajectory is a modeling problem. Their current state — what they know, what they’re confused about, what motivates them — is a hidden state that evolves over time in response to interventions: lessons, practice, feedback. The question isn’t “what’s the best curriculum?” It’s “what’s the right curriculum for this student at this moment, given what the tools we give teachers can observe about their state?”
That’s a fidelity question. The stakes are concrete: get the model wrong and you waste a semester of a child’s irreplaceable time. Get it right and you might be the difference between a student who gives up and one who discovers they love physics. Every parent knows this intuitively. The methodical part — turning that intuition into a measurable, improvable system — is what’s missing.
Healthcare
Same logic, different substrate. A patient’s physiology is a complex dynamical system. Genetics, environment, medications, comorbidities — all interacting. Precision medicine has been promising personalized treatment for years. What’s been missing isn’t genomic data or compute power. It’s the modeling rigor: asking what fidelity of patient model is needed for each clinical decision, validating against outcomes, and iterating. The fidelity question is sharp here: a diabetes screening model needs different resolution than a 10-year mortality prediction, which needs different resolution than a real-time ICU monitor. Matching method to decision is what’s been missing.
Building the Foundation
I’m not pretending to solve all of this today. These are long-game problems that require building credibility and deep domain relationships. Right now I’m proving out the methodology on problems like energy grid resilience and environmental contamination — domains where the data is real, the stakes are high, and the fidelity question is concrete.
But the vision is real. The studies prove the methodology works across different domains, and eventually the methodology goes where it can do the most good.
These are long-game targets, not daydreams. The methodology I use today is the same one that will get me there.
The Methodology Is the Same
Whether you share this vision or just have a hard problem that needs the right model — the methodology is the same. Let’s talk about your problem.