The Dreyfus Model

The Dreyfus model is one of those things you’ve probably seen or heard of in practice, but you didn’t know it had a name. It’s a model of how people develop skills, how we go from knowing nothing to becoming an expert, and past that to mastery.

The stages run novice, advanced beginner, competent, proficient, and expert, with a sixth stage, mastery, reserved for those who push past what already works [1], [3]. For any skill or competency you’re trying to develop, whether it’s a management competency here on Management Craft, cooking, or an athletic skill, you move through these stages in order.

It isn’t a corporate leveling system or an HR ladder, and it doesn’t set your title or your pay band. It’s a map of how good you actually are at a specific thing, and what getting better would look like.

The model’s core insight is that the levels change how you learn. At the beginning, as a novice, you rely on rules. Think of learning to cook: you follow a recipe and measure everything precisely, because you don’t know enough to depart from it without ruining the dish. The more skilled you become, the more you rely on a set of intuitions you have developed out of knowledge and experience, until an expert barely thinks about the rules at all.

The cost of not having this map is everywhere in management. A manager tells someone they need to get better at communication, or storytelling, or executive presence, and stops there. That feedback is too vague to act on. What the person needs is to know which rung they are on and what the next one looks like.

The first key is self-awareness: recognizing that for every skill you have, you sit somewhere on the spectrum, and being honest about what you still need to learn.

This matters most for founders and CEOs, because of how many new things you’re forced to learn. You start with some base of expertise, maybe engineering or design or finance or sales, and then you have to develop every other function of the business: engineering, sales, marketing, product, finance, HR. The chance that you’re an expert in all of them is zero. You come in as a novice to most of them, and the job is to know which skill you’re developing at any given time and roughly where you sit on it.

The second key is being realistic about how far you actually need to go. A lot of the time a founder only needs to move from a one to a two, from novice to advanced beginner. It feels good to push on to competent, but that’s often not the best use of your time. Getting to advanced beginner is usually enough to hire for a role, manage it, and give useful feedback on it. Over time you will need to get genuinely good at a handful of things. There is value in being T-shaped: deep expertise in one area, plus enough range across the others, a one-to-two on many skills, to work across the company instead of being boxed into your vertical.

The third key is using it to fix feedback. The common failure mode is a manager who is frustrated with someone’s performance and hands back something vague in a review. Instead: name the rung. You are at advanced beginner on executive presence, here is what that means, and by the next cycle I want you at a three, here is what a three looks like and roughly how you get there. The other failure mode is impatience. Managers get frustrated that someone is an advanced beginner and want them to be an expert, but people rarely jump from advanced beginner to expert quickly. You have to let someone move through the levels rather than expecting them to fast-forward.

The fourth, deeper, key is one a lot of people miss. The whole point of the early stages, the rules, the rubrics, the deliberate self-assessment, is to make themselves unnecessary. You reflect to build intuition, and then you let it go. The expert who stops to analyze what they are doing gets worse, not better. So the goal is not to think harder forever. It is to think hard, deliberately, now, so that later you do not have to. That settles an argument you hear a lot lately, that reflection and analysis are a waste of time. They are a waste of time for the expert. For the person still climbing, they are essential. You deliberate your way to the point where you no longer need to, finding yourself in flow.

A lot of companies build elaborate leveling systems, the engineer one through three, the associate to senior associate to supervisor ladders, because people want to see how they earn a raise, a promotion, and the status that comes with a better title. I’m not here to fully endorse complicated leveling systems, especially at small companies. What the Dreyfus model does well is simpler and more useful: it creates a shared language between a manager and the person they are coaching, so that “get better at this” means the same thing to both of them.

It also gives real agency to the person receiving the feedback. Any time your boss tells you to improve at something, a high-agency move is to take that feedback to an AI, plot yourself on the Dreyfus model, and then bring it back to your manager: here’s where I think I am, where do you think I am, here’s my plan to improve, and when I think I can get there by.

I use a version of this in my own work. When I’m working on a design, I’ll ask an AI to convene a council of expert designers, each strong in something like color, layout, or typography, and to tell me what principles they would use to critique my work, rather than just fixing it for me. Then I write those principles down by hand, because it helps me remember. I keep them on my desk, and refer back to them the next time. You can do this with almost any skill. It’s not a replacement for human mentorship, but it’s available at any hour without depending on getting other people in a room, and it can speed up how fast you learn.

There is a lot of fear that AI will replace people, or that we will cognitively surrender, offload everything to the machines, and stop developing [6]. The Dreyfus model offers a different way, toward what Doug Engelbart called augmenting human intellect [5]: you use the AI to map where you are and how to climb, but you still have to put in the reps and get better yourself. Let the AI do the work for you and you haven’t developed anything; you’re just prompting a machine and deluding yourself that you’re improving. Root yourself in the model and you keep building real skill.

I love to think of hiring and team building through the lens of The Dreyfus Model. If every competency is a ladder, then a team is a portfolio of positions on a lot of ladders at once, and hiring is the act of assembling that portfolio: which competencies you need someone at the top of, which you only need at competent, which you can cover at advanced beginner. Part of the founder’s job is knowing which rungs to climb yourself and which to hire for. You can’t be an expert at everything, so the real skill is deciding where you need to be, and where good enough is good enough.

Underneath all of this, the truth is that will matters as much as skill. A skill is built by reps, but reps only teach you when they carry an emotional charge: the sting of getting it wrong, the lift of getting it right. That feeling is the reward signal the brain learns from [4]. Take it away and the learning has nothing to run on. So you cannot develop someone who does not care. Engagement is upstream of everything: a person with no will to get better has no reward signal to learn from, and the best coaching in the world lands on nothing. Before you pour yourself into moving someone up the rungs, find out whether the will is there. If it is not, that is the first problem to solve.

One danger with any model like this is treating it as something you have to run all day. You shouldn’t and you won’t. A manager isn’t expected to juggle a million mental models at every moment. There’s a vice of overthinking and a vice of underthinking, and the move is to stay in the middle: kept lightweight, this is just knowing that every skill runs on a handful of levels and being willing to sketch what each one looks like. That used to be hard, but now you can ask an AI to lay out the levels from novice to expert for almost anything in seconds.

So reach for it at specific moments rather than constantly: in performance reviews, when you’re hiring, and when you notice you’re annoyed that someone on your team isn’t meeting your expectations. Irritation at a colleague or someone you manage is a fantastic signal to get specific about which rung they’re on and how you can coach them up.

There is a reason this is worth the effort. So much of the anxiety people carry is the feeling that the world, the system, the man, some outside villain is keeping them small. There is always a moment where you have some control and some agency, and a great place to start is to pick a skill you want to develop, plot yourself on the model, and put in the reps. We are the wardens of our own prisons, and we hold the keys. The keys, most of the time, are working on yourself and getting better at something. Genuine improvement just feels good, and in a world where you feel out of control, deciding to get better at something is a way to take some of that control back.

The model comes from two brothers at the University of California, Berkeley. It first appeared in a roughly eighteen-page report they wrote in 1980 for the Air Force Office of Scientific Research [1], and they developed it over the following decade, most fully in their book Mind Over Machine [2].

The pair is worth knowing for the split between them. Stuart Dreyfus, the older brother, is an operations researcher whose field, dynamic programming, is one of the mathematical foundations of the reinforcement learning that powers modern AI. Hubert Dreyfus, the philosopher, spent his career on Heidegger and phenomenology and became the most famous philosophical critic of artificial intelligence, the author of What Computers Can’t Do (1972), who argued that a machine could never capture the intuitive expertise a person develops. Hubert even picked up a small pop-culture afterlife: the Futurama writer Eric Kaplan studied under him at Berkeley and gave Professor Hubert J. Farnsworth his first name in his honor [7].

So the map of how humans climb toward expertise came from the two people standing on the exact fault line between human and machine intelligence. Stuart, years later, argued that expert intuition is the brain’s procedural system learning by trial, feedback, and reward, the same reinforcement-learning mechanism his own mathematics helped give the machines [4]. The brother who showed that computers can learn the way we do, and the brother who insisted they never really would, together handed us the best tool I know for this exact moment, when the easy thing is to let the machine do the work and call the result your own.