Express the General

In Fear and Trembling, Kierkegaard draws a line between two kinds of people. The tragic hero, who acts within what he calls the general -- the shared, the communicable, the part of us that others can understand. And Abraham, the knight of faith, who cannot explain himself at all. Abraham's reasons live entirely inside him. He is silent.

Kierkegaard adds a line I keep coming back to: "The tragic hero renounces himself in order to express the general." For most of my life I read that as philosophy. Lately I read it as the most practical advice I know for working with AI.

We are the single individual

An AI model is trained on close to everything we have written. The whole of it. That corpus is the general: the shared, the communicable, the sum of what people have managed to put into words.

When you sit down to talk to it, you are the opposite. You are one person, with one context, one set of habits, one way of seeing the problem in front of you. You are the single individual.

And the entire quality of what you get back depends on a single skill: how well you, the single individual, can express your will in the general.

Say the will, and nothing else

The task is smaller than people make it. You are not trying to teach the AI, or cover every angle, or prove you know how it should be done. You are trying to do one thing: express your will as clearly and completely as you can, and add nothing else.

Everything past your will is noise. A method you half-remember. A guess at the solution. A detail you threw in because you were not sure how to say what you actually wanted. The AI already holds the general. It does not need your guesses, and every one you hand it pulls it away from your real intent and toward information that was never yours.

State the will too thinly and the model fills the gaps with its own defaults: ask for "a website" and you get a generic landing page. Blur the will with extra information and you send the model searching, and what it finds is rarely what you meant. The aim is the narrow middle: all of your will, none of your noise.

When you hit it, the model has nothing to second-guess. It acts precisely, and fast, because the will is right there and it never has to look elsewhere to reconstruct what you wanted.

Expressing the general is a skill

This is what Kierkegaard's line finally meant to me. The tragic hero can express the general. That is not a given. It is an achievement, a competence. Abraham cannot, and so he is silent.

Most of us, talking to an AI, are somewhere between the two. We have a will. We lack the craft to express it without our guesses leaking in alongside it. We lead because we cannot yet articulate, and we mistake the leading for being clear.

So here is the discipline, the only definition of prompting I actually trust:

Express your will as clearly and completely as you can, and add nothing else.

What it looks like

Take the simplest task. You want a function to be more readable.

Will buried in guesses:

"Rename the variables to x1 and x2, pull the loop into a helper, put a comment on every line, and avoid early returns, I read somewhere they are bad."

Every clause is a guess wearing the costume of a requirement. None of it is your will. Your will was "make this readable," and you buried it under four methods you are not even sure about. The model cannot tell you that early returns would make the function clearer here, because you ruled out the better answer before it could be offered.

The will, stated plainly:

"Refactor this for readability. It is a hot path, so keep it fast."

Readable is the goal. Fast is the one real limit. Nothing else. The method is left to the general, and the general is far better at method than you are.

It is the difference between a junior who tells a senior engineer which keys to press, and one who states the outcome and the real limits and lets expertise do the rest. The first is leading. The second is expressing the general.

Why this is the whole game

Everything I build with gcontext comes back to this one act: taking what is particular, scattered, and silent, and expressing it as the general, so an agent can meet it with its full competence.

That is true of the context you hand an agent about your system, and it is true of every sentence you type into it. The bottleneck was never the model's intelligence. It is our ability to express. The gap between what we want and what we have learned to say.

The work, the actual work, is learning to express the general. To renounce the part of ourselves that is only noise, and to state the part that is will, cleanly enough that something other than us can act on it.

That is what gcontext is for. And it is what the name has meant all along.