Alignment

Opening signal

The word alignment came into ordinary language from the laboratories where models are built. There it means roughly this: fitting a system to human intentions and values. The question is a serious one and it is good that we are asking it. But a certain gap runs through it — we ask very carefully whether the machine is aligned with the human, and far more rarely whether the human is aligned with themselves.

That gap has a very concrete shape. Imagine a week in which everything adds up. Tasks ticked off, meetings held, the project moved along, nobody let down. On paper, an exemplary week. And on Friday evening you are sitting with a feeling you cannot name: not tiredness, because you slept; not sadness, because nothing bad happened. Something more like absence from your own life. You performed it correctly and you were not in it.

This is not burnout and it is not a midlife crisis. It is misalignment — and it has the unpleasant property of not hurting straight away. It produces no symptom you can respond to. It simply draws energy in the background, quietly, for years, until one day you notice you have too little of it for things that once came by themselves.

Work culture offers one answer to this: more optimisation. A better system, a better tool, a better method of planning. Sometimes that helps. But if the source of the problem is a divergence between what you do and who you are, then every improvement only speeds up the journey in the wrong direction. A better engine does not correct a badly set course — it deepens the error.

This chapter is about order. And about the fact that there is only one right one.

Source

Let us begin with what the texts actually say, and with what we do not know about them. The distinction matters here more than usual, because the formula in question is at once the best known and the most frequently garbled in the whole tradition.

The sentence “as above, so below” comes from the Emerald Tablet — a short text whose earliest attestation is the Arabic work Kitab sirr al-khaliqa, ascribed to pseudo-Apollonius of Tyana and dated roughly to the eighth or ninth century. Nobody has ever seen a Greek original. Nor the emerald. The popular version of the sentence is itself a nineteenth-century abbreviation; the Latin text puts it more cautiously: that what is below is like what is above, and the reverse, so that the miracles of the one thing may be accomplished. That caution is significant — it speaks of likeness and of purpose, not of identity and not of the influence of stars on people.

The second source is the Corpus Hermeticum, a collection of Greek treatises written down roughly between the first and third centuries AD, translated into Latin by Marsilio Ficino in 1471 and for a century and a half taken to be older than Moses — until Isaac Casaubon showed on linguistic grounds in 1614 that they are considerably later. That correction did not invalidate the texts; it only stripped them of the myth of antiquity.

One motif recurs in those treatises: the divided human being, scattered across levels, and knowledge understood not as the gathering of information but as the transformation of the knower. Poimandres, the first treatise in the collection, does not end with a list of conclusions — it ends with someone who sees differently. And the state opposite to knowledge, in these texts, is not ignorance but sleep, drunkenness and forgetting.

The modern source of this chapter is of a completely different kind, and it is worth stating just as openly: it is the vocabulary of safety research on learning systems, from which we took the word alignment. We use it here by analogy and only by analogy. We are not claiming that a human being is a model.

Our reading

We read correspondence as a diagnostic instrument, not as a description of the cosmos. Your values are the level “above”. Your Tuesday is the level “below”. If the two do not correspond, one of them is untrue — and statistically it is not the Tuesday that lies. The Tuesday is mercilessly honest. What usually lies are the values: the ones we would like to hold because they reflect well on us, and which we have never chosen under conditions of cost.

That reversal is the core of our reading. You do not begin with a declaration and then check whether life keeps up with it. You begin with the trace — with what you actually did, who you replied to fastest, what you spent money on without hesitating — and only from that do you read off the pattern that really governs above.

Hence the principle we propose: alignment begins before the tools. Before you align models, workflows, applications and systems, you have to align what lies underneath: values, energy, intention, the way you work, your relationships and your direction. The reverse order works too — and that is precisely the problem. You will build an excellently functioning machine for executing a life you did not want.

It is worth stating what alignment is: a correspondence between what I think; what I do; how I work; how I rest; whom I give access to myself; what systems I build — and what future I allow to happen. The last term is the one most often left out, and without it the whole thing has no direction: the future happens from default settings unless you decide otherwise.

And two things alignment is not. It is not balance — balance assumes that every area gets an equal share, which is a fantasy; there are years of work, years of a child, years of illness, years of a single project. Alignment permits extreme inequality, on one condition: that it is chosen, named and has an end date. Nor is it declared consistency. The point is not values that sound good, but a calendar and a bank account that confirm what you say about yourself in conversation.

The test is one sentence long and uncomfortable: if a stranger read your last month — only the calendar, the spending and the message history — would they guess what matters to you? Alignment is not a perfect state. It is legibility.

Operational translation

Misalignment rarely announces itself directly. It usually comes in by a side door, as a symptom we mistake for something else. The symptoms are worth knowing, because they arrive earlier than any crisis.

• Tiredness after a good day. You did everything you planned and you are wrung out. That is a signal that the work was correct but not yours.
• Procrastination on the important things, efficiency on the unimportant ones. What gets put off is often not the project that is difficult, but the one in which something does not add up and you have not named it yet.
• Irritability with no addressee. Irritation spread evenly across everyone usually concerns none of them.
• Relief at a cancelled meeting. The cleanest reading there is. Notice which meetings cancelled by the other side bring you relief — that is a ready-made list.

Once you have noticed the signal, you need somewhere to put it. Six axes are enough and there is no point multiplying them further.

Values. What I hold to be important — and what I last did when it cost something. A value that has never cost anything has not yet been chosen.

Energy. Whether my actions match my body and my rhythm, or an idea of a person who will never be tired. A plan drawn up for that person is not ambitious, merely untrue.

Work. Whether the thing most of my hours go into moves anything I care about — or only sustains a structure that once made sense.

Relationships. Whom I give access to myself, and whether those are the same people I would call important. The divergence between those two lists is one of the most expensive I know.

Tools. Whether technology works for me or I work for it. Whether this system protects my attention or scatters it. Whether it speeds up something important or only produces more.

Future self. Whether the person I am becoming through these choices is the person I want to be. That is a question about direction, not about goals — goals can change, direction changes more rarely and at a higher price.

The last axis explains why alignment is not a luxury of introspection. In the world after AI, intelligence itself gets cheaper. What gets more expensive: taste, intention, judgement, energy, trust and the capacity to choose. All six rest on alignment — a person at odds with themselves has no stable taste, does not trust their own judgement and chooses whatever is loudest.

Embodied check

Misalignment can be read in the body before it can be put into a sentence. Not because the body is wiser — because it is slower and cannot be talked round with arguments. A calendar can be explained away. A jaw cannot.

Three places are worth checking, because they are cheap to observe and surprisingly reliable. The in-breath. Notice at what point in the day the breath goes shallow and high, just under the collarbones. It usually attaches to a particular person, a particular inbox or a particular kind of task — and that is an address, not a mood. Tension in the jaw and shoulders after a conversation. Not during it, but ten minutes after. What is left after a conversation says more than what you felt in it. Sunday evening. The most uncomfortable indicator we have, and the hardest to argue with — the body knows what starts tomorrow before you have had time to justify it.

There is also a practical rule, more important than the rest of this paragraph: identify the state before the decision. Before you answer a difficult message or close an agreement, ask yourself three questions: how much did I sleep, when did I last eat, is this tension from this matter or carried over from the last one. Decisions made in sleep debt shift systematically in one direction — towards agreeing to terms you would not accept rested.

There is no recovery protocol here and no supplement. There is one assumption: the body is the first system that requires alignment, and checking it takes thirty seconds, not a separate life project.

Practice

The alignment check — a seven-day experiment

One week, one sheet of paper, about ten minutes a day and twenty at the end. You change nothing this week apart from one thing at the end — this is a survey, not a renovation.

Days 1–6: recording the evidence. Once a day, at a fixed hour, write two sentences. The first: what took the most energy today, and was it mine. The second: what I did today although it cost something, because I judged it important. If the second sentence stays empty for the third day running, that is already a result and you do not need the rest of the week to read it.

Day 7: the six axes. Go through values, energy, work, relationships, tools, future self. No scores out of ten — a numerical scale gives the illusion of precision and lets you hide in the middle. Three states instead:

• aligned — it works, needs no move, leave it alone;
• strained — still holding, but costing; watch it and name the price;
• diverged — it has not matched what you declare for a long time; it requires a decision, not an improvement.

Next to each axis add one sentence of evidence from the past six days. The evidence sentence is compulsory and it does all the work. “Energy: strained — third week running I give up my mornings to keep up.” Without the evidence the exercise turns instantly into a mood.

The one-move rule. At the end you pick exactly one axis — not three — and one specific move for the coming week. Choose from the diverged, and if there are none, from the strained. The move has to be small enough to be done in a week in which you already have too much — because every week will be like that. One commitment removed. One conversation. One notification turned off. One hour moved from producing to deciding.

After a quarter. If you repeat this week three times, look for one thing: which axis has stood in the same state for twelve weeks. A repeated misalignment is not a bad patch. It is a structure — and structures do not change with one move, only with a decision.

What this experiment does not promise. It will not tell you the meaning of your life, it does not replace therapy or a conversation with somebody who knows you, and it will not protect you from periods when things are simply hard. It does one thing — it stops a misalignment running in the background for years, unnoticed, quietly drawing the energy you wanted to spend elsewhere.

Limits

The principle of alignment has its broken version and it is worth describing precisely, because it is very tempting. The pursuit of consistency becomes rigidity the moment it starts forbidding you to change your mind. Someone who has said of themselves that they are a person who does X can go on doing X for five years after it stopped making sense — not because they care about it, but because stopping would look like inconsistency. Consistency with an old version of yourself is the most expensive kind of loyalty I know.

The second limit: misalignment can be justified and temporary. The year after a child is born, a year of illness in the family, a year of paying off something that has to be paid off — these are not periods to be fixed with an alignment check. They are periods to be survived. The right move then is to name the price and the end date, not to audit six axes every week.

The third: this practice works badly under heavy exhaustion. When you are tired everything looks diverged, because there is no energy for anything, not because something is wrong with the direction. If you have been sleeping five hours a night for three weeks, the alignment check will mainly tell you that you have been sleeping five hours a night.

The fourth: alignment can be used as an instrument of violence against yourself. You will recognise it by the tone of the notes — if every entry is an accusation, this is no longer diagnostics but a trial in which you are prosecutor and defendant at once. A practice that produces shame does not produce change.

And the fifth, the most practical: not everything can be reconciled and not everything needs to be. Some tensions are structural — you want time for people and you want to build something large. That is not removed. That is held, deliberately, with the price written into the budget.

Ethical question

The question is: who pays for my alignment?

Because usually somebody does. Putting your own life in order almost always means moving something onto somebody else — a project, a shift, a conversation I do not want to have, an obligation that has stopped suiting me. That is permissible and often necessary. What is not permissible is doing it without counting the cost and without saying so plainly.

Three distinctions worth holding here.

A boundary and emotional closure. A boundary is stated openly, it has an addressee and a reason, and it leaves the relationship open. Closure is silent — it consists in ceasing to reply and calling it hygiene. The first builds trust; the second simply transfers the cost to the other side, who does not even know they have borne it.

Protecting your energy and indifference. Refusing to take part in somebody else’s chaos is protecting your energy. Refusing to take part in somebody else’s difficult moment because “it is not my responsibility” is indifference with a better vocabulary. The difference shows in whether you are prepared to pay at any point — or whether it always turns out that just now you are not.

Autonomy and isolation. A life lived in accordance with yourself very easily becomes a life in which nobody has the right to disturb you any more. That is comfortable and empty. An alignment with nobody in it who can come to you and ruin your week is not alignment — it is impermeability.

There is also the reverse question, asked less often: do I have the right to demand alignment of others? I do not. I can describe my own terms, I can leave, I can decline to enter. I cannot turn my own orderliness into a standard by which I measure people in a worse period of their lives. Another person’s Tuesday is not a document I am entitled to read.

Closing formula

We return to the word that came out of the laboratories. Whether the systems we are building are aligned with human values is one of the more important questions of this decade. But it has a second half, spoken of more quietly: aligning technology with the human assumes that there is something on the human side to be aligned with. That there is an intention there, and not a set of reflexes. A direction, and not default settings.

This chapter closes nothing — it only sets the order. First the lower level: your Tuesday, your energy, your people, your tools. Then the upper level: who you are becoming through them. And the check of whether the one corresponds to the other, carried out regularly and without pathos.

Alignment is a condition, not a result. Once the course is set, the question remains what to do with all the energy that has been freed up along the way — and with the energy that arrives uninvited, together with tension. That is what the next chapters are about.