No, really, isn’t this absurd? As a person, the thing I do most is make decisions — and I mean a lot of them. We’re talking 100+ consequential choices every day: what to work on, when to rest, who to talk to, how to talk to them, where to live, what problems to focus my limited time. And yet, in all my years of school, not once did anyone teach me how to actually make a good decision.
And as I grew older the stakes only got higher. Right now I’m dealing with some big life decisions: job search and potential relocation, thinking through what kind of work I want to do next and where I want to move the family to. Since our daughter was born almost 2 years ago decisions like these have 100x the impact. The implications are profound, the variables are messy, and there’s no clear “right answer” waiting at the end of a spreadsheet. But that doesn’t mean I should be winging it. And a spreadsheet isn’t the answer.
I was lucky enough to eventually meet people who’d understood this and were willing to point me in the right direction. I’ve been lucky, and yet even when you know what you’re supposed to do it can be hard to do it. I’ve learned this over years of coaching, and with the advent of LLMs I’ve naturally gravitated toward exploring AI-assisted coaching. So recently I’ve been working on an AI agent to help me work through a structured decision-making process. And it’s been incredibly helpful. Not because it tells me what to do, but — like a good coach — because it keeps me honest about how I’m thinking and makes sure no stone stays unturned.
Mind the Gap
I figured out this gap late — probably in my early twenties. I’d gotten into the digital nomad thing, online business, traveling a bunch. For the first time, I was thinking about business beyond just software development and standard employment. Decisions got bigger: what products to build, where to live, how to structure my work. And one day it just hit me: I actually had no clue how to make good decisions.
I’ve never been a particularly good student. Didn’t like school. But I went looking for something — anything — that could teach me how to think properly. This was early days of MOOCs, before Coursera and Udemy took over everything. I found a couple of Stanford classes run by the Decision Education Foundation: one focused on business decisions (heavy on math and stats), the other on everyday life choices. I thought I’d try out the second one.
By far, that was the most impactful and useful class I’ve ever taken. By so much it’s hard for me to wrap my head around the impact it’s had on me. The biggest lesson? Don’t judge decisions by outcomes — which is instead the most common thing everybody in my life did. Something went well? Good decision. Bad? Bad decision. It seemed to make a lot of sense, but I quickly learned it was actually the worst thing you could do. Luck is always a factor. You can make a terrible decision and get lucky. You can make a brilliant decision and get unlucky. There’s so much we can’t control about how something turns out, but we’re unwilling to admit it. It makes us feel powerless, so we pretend that those winning owe it to their good decisions. The quality of a decision is in the process, not the result.
Data-Driven Decisions Are Often Just Cover-Your-Ass Theater
All this modern stuff about data-driven decisions? It’s largely bullshit. Sorry.
Don’t get me wrong — in the face of uncertainty, it’s only logical to look for ways to be more informed. But to think that data is objective, or worse, that it’s comprehensive in capturing everything that goes into a good decision? That’s delusional. You picked the data; the data didn’t pick itself. You think you’re not biased? You likely didn’t have the required volumes for anything you could call statistically significant and are calling a qualitative experiment quantitative because it sounds better. And what about that control group? Cohorts? Oh, and are you really just testing one thing? Most product changes aren’t single small adjustments — they often bring in more than one variable. Now how do you know it was really that one thing that did it? Attribution is hard.
Does that mean we should just be winging it? No, of course not. Like I said above, a solid controlled process can help make sense of complexity and uncertainty. But in my experience, what people call “data” is far too limited, and it’s mostly an attempt to create a safety net in case something goes wrong instead of trying to really make the best possible decision. It’s a way to feel better about a choice you’ve often already half-made.
A good decision in my book accounts for: facts, context, principles, constraints, biases, values, and yes, intuition. Instead of just running an experiment like “do people want feature A or feature B,” there’s a lot more we have to go into. What kind of company are we building? What are our principles? Is there a problem behind the problem? Where is this really coming from? What are our biases as we consider the problem? Any investment and pre-existing conditions that led us to frame things this way?
That’s first-principles thinking. That’s systems thinking. And it comes before any domain-specific thinking.
Why I Actually Built an AI Panel, Not a Coach
Here’s where AI gets interesting — but not in the way most people use it.
Andrej Karpathy a few weeks back made a point about too much anthropomorphism and I completely agree. I found this article from CommonCog to make the clearest point about it. In short, if you just tell an AI “you are a xyz,” you get high uncertainty about what you’re actually getting. Who’s “you”? What kind of coach? What’s their philosophy? Anthropomorphizing AI like that is a gamble.
But if you invoke specific personas — “what would Seneca say about this?” or “how would Charlie Munger think through this?” — you reduce that uncertainty. You’re anchoring the AI to known ways of thinking.
So I built a panel. Not a single AI guru, but a facilitator that brings multiple perspectives to the table:
Ancient thinkers (Aristotle, Seneca, Confucius, Moses) — to counter recency bias, because we over-index on whoever published a Medium post last week
Modern strategists (Peter Drucker, Charlie Munger, Warren Buffett, Shane Parrish from Farnam Street) — for mental models and business thinking
The sage lens — wisdom-focused perspectives that bring in values, principles, long-term thinking
The AI doesn’t become these people. It moderates a conversation. It’s not a coach telling me what to do. It’s a guide helping me work through a process with the help of known thinkers.
That process is the Decision Education Foundation’s 7-step framework, which I’ve incorporated into every coaching program and founders cohort I’ve ever run:
Frame the decision — What are we actually deciding? Is there a problem behind the problem?
Gather information — What do we know? What do we need to know? What can’t we know?
Identify alternatives — What are the real options here? Are we creating false binaries?
Evaluate trade-offs — What do we gain and lose with each path? What are the second-order effects?
Clarify values — What matters most to us? What kind of person/company do we want to be?
Check for biases — What assumptions are we making? What cognitive traps are we falling into?
Make the call and plan — Decide, commit, and define what success looks like independent of outcome
This isn’t a checklist. It’s not linear. It’s iterative. You loop back. You realize halfway through step 5 that you framed the decision wrong in step 1. That’s fine. That’s how thinking works.
Teaching AI to Actually Think
And here’s the hard part: AI doesn’t naturally do this.
I had to coach the coach. The AI made the same mistakes I used to make. It wanted to jump straight to solutions. It info-dumped. It went down rabbit holes. It was kind of weird to see the AI actually go down a rabbit hole, but it does. It definitely does.
Early versions of the panel were too noisy — having ten distinct voices trying to weigh in on every question was overwhelming. So I grouped thinkers into three lenses: skeptics, sages, and strategists. That helped, though it somewhat undid the benefit of invoking specific personas. I’ll try different approaches in the future.
The AI also struggled with treating the framework like a linear checklist instead of an iterative conversation. I had to add explicit instructions: adopt a coaching stance. Ask one question at a time. Check if it landed well. Adjust based on the response. Skepticism isn’t a step in a process; it’s a lens you apply to every new fact that enters the conversation. This “awareness” made a huge difference.
And time and dates? Still needs work. It turns out AI sucks at it, just like it sucked at math for a long time. I had to add a lot of stuff to my standard CLAUDE.md to make sure it was more time-aware.
But here’s the key: I could only fix these things because I’d made the same mistakes myself. I recognized when the AI was jumping to conclusions despite having a framework in front of it, because I’d done that. I caught it skipping the hard work of framing the problem, because I used to do that too.
If I hadn’t spent years learning how to think properly, I never would have been able to build a tool that could help me think better.
It Actually Works
I’m using this right now for job search and relocation decisions. It’s helping me think through what kind of work I want, what geography makes sense, how to weigh trade-offs between income, autonomy, and quality of life. Not giving me answers — helping me structure my own thinking, which is extremely helpful because the problem is complex and no matter how much I try to be self-aware, I’m human. I slip. I forget things. I get thrown off by my own biases.
My wife also used it recently to think through a master’s program proposal. She was exploring different directions, trying to figure out what really interested her and how to frame a compelling pitch. Her feedback: “Very effective. Very helpful. Asked very good questions.”
That’s the goal. Not an oracle. Not a guru. A thinking partner that helps you stick to a process when it’s hard to do it alone.
Decision-Making Is a Craft
Here’s what I’ve learned: decision-making is a craft, like any other. You can learn it. You can get better at it. And you can definitely teach it to an AI agent. It’s probably the most important craft you’ll ever develop, because everything else in your life flows from the choices you make.
But it’s hard to stay on target. Even when you know the steps, even when you’ve internalized the framework, it’s easy to slip into shortcuts. To skip the hard questions. To let biases quietly steer you without noticing.
Having a partner helps. A good coach is ideal, but there aren’t enough good ones, and plenty of bad ones will do worse than a well-built AI. The proliferation of coaches doesn’t mean the quality is there.
So I built this. And it’s working. Not perfectly — there are still rough edges, things I’m iterating on. But it’s doing an amazing job of keeping me honest, surfacing my assumptions, and making sure I actually do the thinking instead of just deciding and retrofitting justification.
That’s the real work. Not finding the right answer. Doing the thinking properly.