Nobody Writes Down the Reason (or, The Most Perishable Product of Getting Good at Something)

Someone asked me last week why I use a particular tool the way I do. Not a hostile question, not an interview. Just a colleague who'd noticed a habit and wanted to know. And I opened my mouth to answer and found, to my genuine surprise, that I couldn't. Not because the answer was complicated. Because the reasoning had simply evaporated.

I use the tool. I use it in a specific, idiosyncratic way that I've built up over a couple of years. The workflow is there, the muscle memory is there, the outputs are there. But the reasoning, the chain of small decisions that got me from "here's a thing that exists" to "here's how I use it": that's gone. What remains is the outcome, the artifact, the habit wearing itself like a skin. The reasoning that built the skin has dissolved into it, the way scaffolding comes down once the building can hold itself up.

I tried to reconstruct it in the moment. I said something like "well, I started using it this way because" and then trailed off into a story that sounded plausible but felt, even as I was telling it, a post-hoc invention. A myth I was generating in real time to explain a fact I already knew. Which is a strange thing to catch yourself doing.¹

Here's what I think is actually going on, which sounds simpler than it is: it takes a long time to discover what something is for. Not in the sense of reading the manual. I mean the specific, personal, particular sense of what a tool is for you, in the context of your actual work and your actual cognitive style and the actual problems you're trying to solve. And that discovery process is slow and weird and nonlinear, and it generates an enormous amount of reasoning. Intermediate conclusions, rejected framings, adjusted beliefs, small revelations. A running commentary on the process of figuring out.

And we don't write any of it down.

This isn't a productivity complaint. It's an observation about the structure of how capability and understanding arrive together, or more precisely, how they don't. Capability tends to arrive in a lump, at the beginning. You get access to the thing. You can, in some technical sense, use it. Understanding (the specific, personal, earned variety) dribbles in over months. And by the time the understanding has crystallized into something you could actually explain to someone, the reasoning that built it has been absorbed into the result and is no longer legible, even to you.

The psychologist James Gibson had a name for the relationship between creatures and the possibilities available to them in their environment: affordances. The basic idea is that what we perceive in any object or environment isn't really its objective properties. It's the action possibilities it offers to us specifically, given our particular capacities and needs. A chair isn't a chair in the abstract; it's a sitting-affordance for a person of a certain size who needs to rest. A rock that's useless to most people affords hammering to someone who needs to drive a stake. Affordances aren't in the object, and they're not in the perceiver. They're in the relation between them, and they're only visible to the perceiver who has the right relationship with the object to notice them.

The thing Gibson didn't emphasize enough, at least for my purposes, is that affordances can be latent. They can be there in the object-person relation and still invisible to you, because you haven't yet done the work of perceiving them. A tool can sit on your desk for months in a kind of affordance-poverty. You see only the obvious uses, the intended uses, the uses spelled out in the documentation. And then one afternoon a particular problem arrives and you pick the tool up in a slightly different way and something clicks into place. You've discovered an affordance you didn't know was there. The tool didn't change. You changed, slightly, in some way that was hard to specify, and suddenly the tool is different.²

And what I want to suggest is that the reasoning surrounding that click (the particular problem that was present, the particular failure mode that you'd run into with your previous approach, the specific thing you noticed that made you pick the tool up differently) is incredibly valuable and almost never recorded. It's the most perishable product of the discovery process, and it is almost certainly generating faster than it ever has before.

This is where Daniel Wegner's work on transactive memory becomes worth dwelling on, or at least worth dwelling on for me, which isn't quite the same thing. Wegner's insight, developed in the 1980s from watching couples navigate daily life, was that human beings are fundamentally bad at remembering things in isolation but surprisingly effective at remembering things in relation to other people who also remember things. We don't store everything ourselves; we store a kind of directory: who knows what, and how to access it. Couples routinely outsource whole categories of memory to each other. One person knows where the important documents are; the other knows the relevant passwords; both know that the other knows, and the system works until one of them isn't there anymore.

What Wegner observed at the level of couples scales up in uncomfortable ways to teams, departments, and organizations. Any group of people working together long enough develops a kind of collective cognitive infrastructure. Not just shared knowledge, but shared access pathways to knowledge. Person A doesn't know the answer, but knows that Person B does, and knows approximately how to ask Person B in a way that will produce a useful result. The system functions as a whole even when no individual contains all of it. Which sounds efficient, and it is, until the people start leaving.

What I find quietly devastating about this is the implication for reasoning, as opposed to facts. The transactive memory system is great at preserving what. Where the keys are, how to get to the restaurant, which vendor to call, which approach worked last time. What it's much less reliable about is preserving why. The reasoning behind a choice, the logic that supported a decision, the history of adjustments that got you from the original plan to the thing you're actually doing. That stuff tends to live in conversation, in the dynamic back-and-forth of working something out with other people. And when that conversation ends, and the people disperse, a lot of the why goes with them.³

I've become increasingly aware of how much of my own reasoning is stored, in some distributed and unstable form, in people I've worked with. Not facts about projects (those are written down, usually, or at least findable). I mean something more specific: the reasons behind the choices that shaped those projects. The argument that didn't make it into the final memo. The constraint that was understood by everyone who was in the room and never explained to anyone who wasn't. The decision that looks, from the outside, arbitrary or wrong, because the reasoning that justified it is now inaccessible.

This is uncomfortable to sit with, because it suggests that a lot of what we think of as our own thinking is actually distributed across people and relationships that are themselves temporary. When the people leave, or the relationships change, or we simply move to a different context, some of that thinking leaves too. What remains is the output: the decision, the habit, the workflow, stripped of the reasoning that made it make sense.

William Stanley Jevons noticed, in 1865, that making coal use more efficient didn't reduce coal consumption; it increased it, because cheaper coal made more coal-powered applications viable, which created more demand, which burned more coal. The Jevons paradox, as it came to be called, is a good example of a case where the obvious prediction (more efficiency means less consumption) turns out to get the direction exactly backward. Efficiency creates more demand, not less, because it unlocks applications that weren't viable at the previous price.

I want to propose a Jevons paradox of reasoning. As capability gets cheaper and more accessible (any capability, not just AI specifically), the cost of discovering what it's for goes down. More people can experiment with more things. The rate of discovery accelerates. The number of those click moments, those afternoon revelations, multiplies. And the reasoning surrounding each of those moments (the specific cognitive path that led from confusion to understanding) accumulates faster than it can be recorded. The cheaper and more accessible the capability, the more reasoning evaporates. Not because we're being careless. Because the rate of discovery outpaces the rate of documentation.⁴

There's an old argument, in Plato's Phaedrus, about whether writing is actually good for memory. The Egyptian god Thamus argues against writing on the grounds that it will produce forgetfulness, not because written records are bad, but because the existence of external records trains people to stop doing the internal work of retention. Why remember something when you can look it up? The argument has resurfaced approximately every time a new external memory technology arrives: the printing press, the internet, the smartphone. And the standard response is that external memory frees up cognitive capacity for higher-order thinking.

I think both sides are right and both sides are missing something. What external memory technologies are genuinely good at is preserving the what. Dates, facts, outputs, decisions. What they're not good at, and what we systematically fail to use them for, is preserving the because. The reasoning. The intermediate states. The stuff that made the what make sense in context.

Socrates, for all his curmudgeonly resistance to the written word, was onto something when he worried about the appearance of wisdom without its substance. You can have a perfectly complete record of what was decided and still be entirely lost when the context shifts and you need to understand why it was decided, because the why was never written. It was in the room. It was in the people. It lived in conversation and dissolved when the conversation ended.⁵

None of this is an argument against writing things down, or against building better external memory systems, or against the various AI-powered tools that are now offering to synthesize your history and keep your context current. The case for those tools is real. What I want to add to it (not argue against it, just add to it) is the observation that the thing most worth preserving is also the thing least likely to be preserved, because it's the thing that disappears before you realize it was worth writing down.

The reasoning that explains your current habit was most visible the week before the habit solidified. That's when it was still active, still working, still in play. By the time the habit has become comfortable and automatic, the reasoning has been absorbed into the behavior and is no longer accessible. You know what you do. You've forgotten why you do it. And if someone asks you, if a colleague puzzled by a habit they've noticed asks you a perfectly ordinary question about your workflow, you find yourself generating a plausible story in real time, which sounds like an explanation but is actually a reconstruction. An educated guess about your own past thinking.

This would be less important if the reasoning only mattered for explaining yourself to others. But it matters for something more practical: making the next decision.

When circumstances change (and they always change, eventually), what you need to navigate the change is not just a record of what you chose, but an understanding of why you chose it. You need to know what conditions the original choice was designed for, what trade-offs it was making, what it would have been wrong to do instead. You need the reasoning, not just the result. And if the reasoning isn't there, if it dissolved into habit and conversation and the particular Tuesday afternoon when something clicked, you're left making the new decision with incomplete information about the old one. You're renovating without knowing which walls are load-bearing. The project proceeds, more or less, and the gaps only reveal themselves when something unexpected goes wrong and you need to understand the original logic to fix it.

The other consequence, which is subtler, is that you lose the ability to transfer the understanding. If you know what you do but not why you do it, you can demonstrate the practice but you can't explain it. Not in a way that equips someone else to adapt it to their different circumstances, their different cognitive style, their different problem shape. What gets transferred is the behavior, extracted from the reasoning that would make the behavior sensible. And behavior without reasoning is a recipe that you copy without knowing which substitutions are safe.⁶

Which is, I think, roughly the situation most of us are in most of the time. Not because we're bad at documentation. Because the reasoning is always disappearing faster than we can catch it, because capability keeps arriving in lumps and understanding keeps dribbling in after, because the discovery process is messy and nonlinear and generates its most valuable byproducts in the moments before the habit forms and the reasoning goes quiet.

I don't have a clean solution to this. The honest answer is that I record far less of my reasoning than I should, and the gap shows up in exactly the way you'd expect: when someone asks me why I do something the way I do it, and I find myself making up a story that sounds right but might not be.

What I've started doing, imperfectly, is trying to write down not just what I decided but the thing that made me decide it. Not the outcome. The click. The particular problem that revealed the latent affordance. The specific failure mode that made me pick the tool up differently. The conversation in which I understood something I'd been using wrong for six months.

It's a strange practice, writing down your reasons. It feels overserious in the moment, like you're treating an ordinary Tuesday like a historical event worth documenting. But I've started to think that's actually the right register, because the ordinary Tuesday is a historical event, in the sense that the reasoning it contains won't be available later. It will dissolve into the habit it creates, and the habit will work, and you'll know how to do the thing, and you will have forgotten, genuinely and irretrievably, why.

My colleague is still waiting, technically, for my answer about the workflow. I've been thinking about it since she asked.

The best I can say, honestly, is: I think I figured it out in about month eight. There was a specific afternoon when something clicked. I can feel the shape of the moment, the sense of things settling into place, the small cognitive satisfaction of a previously wrong model becoming right.

I just can't tell you what I was thinking.

¹ There's a version of this that is probably just ordinary confabulation, the well-documented tendency of human beings to generate post-hoc explanations for decisions actually made on other grounds. The unsettling thing is that confabulation is nearly impossible to distinguish from accurate reconstruction, including from the inside. The story sounds right because you're the one telling it. ↩︎

² I should note that the click doesn't always come. Some tools never yield their latent affordances, either because the affordances aren't there for you specifically, or because you never quite approach the tool from the right angle. The sunk cost of having paid for something does not guarantee that you'll ever figure out what it's actually useful for. ↩︎

³ The extreme version of this is when an institution loses the people who understood a particular system and is left maintaining something nobody can fully explain. The system works, technically. Everyone follows the procedures. Nobody knows why the procedures are the way they are, because the people who made those decisions have left, and the reasoning left with them. This is sometimes called "organizational memory loss," which is a clinical way of describing something that is, experientially, closer to amnesia. ↩︎

⁴ This is related to, but not identical with, the Jevons paradox. Jevons is about resource consumption; what I'm describing is about cognitive byproduct. The acceleration of discovery doesn't just produce more useful understanding. It also produces more evaporated reasoning. The ratio of what we learn to what we document stays unfavorable, or gets more unfavorable, as the rate of discovery increases. ↩︎

⁵ The other thing Socrates was right about, which is less often cited: writing can give the appearance of knowing something you actually only have a record of. You look up the fact, you see it on the screen, and for a moment you feel like you understand something you've actually just remembered exists. The record and the understanding are different things, and external memory systems are very good at providing the first while leaving the second entirely to you. ↩︎

⁶ There's a useful analogy in software: when a system breaks in an unexpected way, the most useful thing is often not the code itself but the comments, the human reasoning about why the code does what it does. A codebase without comments is a set of decisions without reasoning. You can run it. You can't easily change it without understanding what it was designed for. And yet most code is written with far more attention to the decision than to the documented logic behind it. ↩︎