Leaving at four

Episode 1 — Leaving at four

He told me this on the phone, after the broadcast, with an unease in his voice that I only understood once I had hung up.

Chief financial officer of a mid-sized group. For eighteen months, part of what used to fill his evenings — the consolidation, the reconciliations, the formatting of the accounts, the preparation of the board packs — has been produced by a system, reviewed by him, validated in two hours instead of two days.

He said: I am going to pick up my daughter from school on Tuesdays. She is eight. It is the first time.

Then he added, quieter: three people are not being replaced in my department.

Both sentences come from the same deployment, the same budget, the same executive decision. They are both true. It is that simultaneity I want to address over seven texts, because I do not know how to hold it with the tools public debate offers me.

Two convenient ways of not thinking about it

The first is to keep only the four o’clock hour. It is called augmentation, assistance, liberation from low-value time. The vocabulary already says it all: what disappeared had no value, so nothing disappeared.

The second is to keep only the three jobs. It is called destruction, replacement, wrecking. The vocabulary is just as complete: what appeared has no value, so nothing was gained.

Each of these positions holds up perfectly, as long as you look at only one half of the phone call.

I am not looking for a compromise between them, which would be mere prudence. I am looking for a criterion that lets us say, in a given case, whether what just happened is progress for the human person or degradation — regardless of the net job count.

A text that asks exactly this question

On 25 May 2026, Pope Leo XIV published his first encyclical, Magnifica humanitas, devoted to the protection of the human person in the age of artificial intelligence. It was signed ten days earlier, on 15 May, the one hundred and thirty-fifth anniversary of Rerum novarum — and that date is not decorative: it places the text in the lineage of the encyclical through which the Church entered the workers’ question in 1891.

Let me state at once where I read from. I am not reading this text as a theologian, and I will not discuss here what belongs to faith. I read it as a document of social doctrine — that is, as one of the few recent intellectual productions that approaches automation from the person rather than from productivity. One may find it insufficient. One cannot find it banal: almost every report published on the subject over the past three years starts from the labour market and arrives, at best, at the person.

The encyclical does the reverse, and it draws from it a criterion of disarming simplicity: the human being must remain at the centre, never at the periphery.

Put that way, it sounds like a moral truism with no operative content. I believe on the contrary that it is the only criterion in this matter that does not go stale, and I will try to show that it settles concrete cases — including cases where it settles against intuition.

The text also sets out, from its first lines, an alternative between two possible constructions: a new tower of Babel, or a city that can be inhabited together. At first I found the image too beautiful to be useful. I now believe it says something precise, and I will return to it in episode 6, because the difference between the two constructions is not in their height. It is in who decides to build them.

An argument I refuse to use

There is a convenient way of dissolving concern about jobs, and it circulates widely, including among serious people.

It consists in recalling that agricultural mechanisation, electricity, the shipping container, business computing and the automated teller machine were all announced as job destroyers, and that the number of jobs nevertheless kept rising. The loom did not abolish work: it moved the weavers.

The argument is sound on the facts, and I am going to do without it. For three reasons.

The first is logical, and it is the same one I made two weeks ago to those who waved away anxiety about school by citing Socrates against writing. A string of excessive alarms does not establish a rule. It establishes that we were wrong before. Whether this one belongs to the same order remains an open question, and it is not settled by analogy.

The second is a question of scale of observation. The net balance is an aggregate quantity, at the scale of an economy and over decades. The loss, by contrast, is concentrated on identifiable people, within a timeframe counted in months. Telling a forty-eight-year-old accountant that the balance will be positive in 2040 is true and useless: her working life is not long enough to benefit from the average.

The third seems to me the most important. This argument answers a question — how many jobs — that the encyclical does not ask, and that I do not want to ask either. One can perfectly well imagine an economy that creates as many jobs as it destroys, in which the remaining work would be entirely organised by systems no one understands, for people who have nothing to say about it. The accounts would balance. The person would be at the periphery.

That possibility is what interests me, because it appears in no employment statistic.

What I am looking for

A criterion for distinguishing, in any given deployment, the automation that enlarges the human being from the one that shrinks them — and that depends on neither a technological forecast nor a net accounting.

I believe there are four, that they hold together, and that the last is a condition without which the first three produce nothing.

I warn you at once: the conclusion is not reassuring. The Tuesday-at-four story is true, it is beautiful, and it is not a prediction. It is an exception produced by someone who had the power to produce it. Everything that follows consists in understanding what it would take for it to stop being one.

The itinerary of the seven episodes

This series is a single line of reasoning, cut into seven parts.

1. Leaving at four — this text: setting out the simultaneity, refusing both consolations, and adopting the criterion of the person at the centre rather than the job balance.

2. What the machine actually took from him — first criterion: it did not take a profession, it took tasks. Which ones, and how to tell apart those whose disappearance enlarges the person who performed them.

3. The CFO could delegate, the cashier cannot — second criterion: freed time is not distributed according to technology, but according to position. Whoever decides what gets automated decides who benefits from it.

4. The hour given back does not go where you think — third criterion: productivity gains have never been spontaneously converted into time. Time that is freed but not institutionalised is reabsorbed, and this has already happened several times.

5. You do not become better by doing less — the most serious objection to the three previous episodes, taken seriously: nothing guarantees that the hour given back is used to grow, and competence is built precisely through what you delegate.

6. What must be decided, and by whom — the instruments are not commutative. The order in which they are activated decides the outcome, and the usual order is the reverse of the right one.

7. What I would do with a company and three years — the synthesis, what I would not do, and who bears the burden of proof.

One clarification, to avoid the usual misunderstanding.

I am putting no one on trial. Not this chief financial officer, who did what any parent would do. Not the executives who arbitrate under the constraint of margins and competition. Not the engineers who build these systems, many of whom honestly believe they are working on something useful.

I simply think we have got into the habit of asking the wrong question — how many — where we should ask another — who, what, and decided by whom. And that it is still early.

Open question: what is the hour in your week that you do not take and would like to take?

Episode 2 — What the machine actually took from him

Let us go back over the case closely, because the vocabulary of public debate distorts it immediately.

People say: artificial intelligence replaces professions. That is false, and the falseness is not a detail of language.

It does not replace professions. It performs tasks. A profession is an assembly of tasks, and automation never takes the whole assembly: it takes the most formalisable tasks, leaves the others, and redistributes what remains.

This has been true since the first spreadsheet, and it remains true here. What has changed is the extent of what has become formalisable, and the speed at which that extent grows.

So let us look precisely at what was taken.

The inventory

At this chief financial officer’s company, the machine absorbed account reconciliation, the aggregation of files coming from four subsidiaries, the formatting of monthly statements, the production of the first variance commentaries, the drafting of the board materials.

It did not absorb: deciding whether a provision is sufficient, telling the chief executive that an acquisition is too expensive, holding a bank negotiation, arbitrating between two investments one of which will cost someone their job, signing off the accounts.

The difference between the two lists is not a difference of difficulty. Some tasks in the first list are technically painful; some in the second are very simple to state. Signing is a gesture.

The difference lies elsewhere, and it is the first criterion of this series.

Work that builds the person who does it, and work that wears them down

Magnifica humanitas devotes a passage to the dignity of work, and it takes up there an idea that runs through social doctrine since Rerum novarum: work is not only a means of producing, nor only a means of earning a living. It is also one of the places where a person is realised.

That sentence is easy to quote and difficult to use. I am trying to make it operative.

If work is a place of realisation, then not all tasks have the same value for the person who performs them. Some build a skill, a representation of the field, a relationship, a judgement. Others build none: they consume time and attention, they tire, and after ten years they have left nothing.

Posting a thousand account reconciliations teaches you a great deal. Posting the hundred thousandth teaches you nothing.

Hence the first criterion, which I formulate as follows:

Automation enlarges a person when it takes from them the work that no longer builds them.

And the test that goes with it, which I find astonishingly discriminating when applied honestly:

Would you want your child to do this task, eight hours a day, for thirty years?

If the answer is no — and for invoice data entry, statement checking, transcription of recordings, formatting of tables, the answer is no — then its automation is a gain in dignity, whatever happens meanwhile to the job that contained it.

If the answer is yes — for teaching, care, conducting an interview, making an object, negotiating, diagnosing — then its automation is a loss, even if it is profitable, and even if it is well done.

Why this criterion does not move

It has a property I was looking for: it does not depend on the state of the technology.

The usual criterion — automate what the machine can do — is revised with every version, and it condemns you to running. The criterion of building is placed on the side of the person. It only moves if our idea of what makes a human being grow moves, which happens on the scale of centuries, not of quarters.

It has a second property, less obvious: it cuts both ways, and it sometimes cuts against the employer.

A company that automated customer relations — because it is measurable, costly and technically feasible — would remove tasks that genuinely build the people who do them: listening to someone who is upset, understanding a situation, finding an arrangement. It would be profitable and it would be, in the sense of this criterion, a regression.

Conversely, a company that refused to automate data entry out of attachment to the job would keep people in a task that no longer teaches them anything. It is generous and it is, in the sense of the same criterion, a disservice.

The criterion is therefore uncomfortable on both sides. That is rather a good sign.

What this criterion does not settle, and I will not hide it

There is a hole, and it is wide open.

The accountant who did those reconciliations did not experience them as alienation. She experienced them as her salary, her profession, her recognised skill, her place in a department, the reason people said good morning to her.

Explaining to her that the task that disappeared no longer built her is true, condescending, and without effect on her rent.

The dignity of work, in social doctrine, does not designate only the quality of the task. It also designates the fact of having a job, living from it, and being recognised in it. Both dimensions are in the same text, and they can come into conflict in the same concrete case.

This criterion therefore settles one question — which work is worth defending — and leaves another entirely open: what do we do with those whose task we have just established no longer built them.

I have no answer at this stage. Part of it will come in episode 6, and it will be neither elegant nor free.

A remark about our economies

A clarification that applies to Congo, Algeria and most of the continent.

Among us there is a temptation that is the symmetrical opposite of the one in the economies of the North. There, they fear losing office jobs. Here, we congratulate ourselves on gaining them: call centres, offshored data entry, accounting back offices, content moderation, data annotation for the laboratories of the North.

These jobs are real, they support families, and I am not going to look down on them from a radio studio.

But we have to look at what they are: they are very exactly the tasks in my first list. Formalisable, repetitive, measurable, and therefore the first to be exposed. A national strategy that consists in attracting data-entry jobs because the hourly cost is competitive is a strategy that sells an advantage that is disappearing.

That is not a reason to refuse these jobs. It is a reason never to treat them as a destination.

Open question: in your work, what is the task that has taught you nothing for a long time? And why are you still doing it?

Episode 3 — The CFO could delegate, the cashier cannot

I come back to the phone call, to a detail I had not noticed at the time.

He said: I delegated certain tasks. He used the active voice. It was he who chose what would go to the machine and what he would keep.

That detail contains, I believe, half of this whole affair.

Who holds the handle

He kept the bank negotiation, the investment arbitration, the relationship with the board. He delegated the reconciliation and the formatting.

That is to say, very exactly, that he kept what interests him and gave away what bores him.

Anyone would do the same. The point is not moral, it is structural: he could do it because he was the chief financial officer. He had authority over the scope of the tool, over its configuration, over what is entrusted to it. The accountant working under him had none of those three things.

She did not delegate. Things were taken from her.

And that is the second criterion of this series:

The time freed by an automation goes to the person who has the power to define what gets automated.

This is not a property of the technology. It is a property of the hierarchy, which the technology reveals and amplifies.

The same tool, two opposite effects

Take a document recognition and processing system, exactly the same one, and deploy it in two places.

At the chief financial officer’s: it removes a chore, leaves the heart of the profession intact, and gives back an evening. The human is at the centre, the tool at the periphery.

At the data-entry operator’s, whose whole profession was made of that chore: it removes the job, or it turns it into the surveillance of a screen where proposals to validate scroll past, at the machine’s pace, with a productivity indicator. The tool is at the centre, the human at the periphery — and literally so: they have become the catch-up device placed at the end of the line.

Same technology. Opposite effects. The variable is not technical.

What subsidiarity says here

Magnifica humanitas takes up, among the principles of social doctrine it mobilises, that of subsidiarity: decisions must be taken at the level closest to those they concern, and higher levels intervene only to support what the lower level cannot do.

The principle is usually cited about states and local authorities. Here it has a very concrete translation, and I have not seen it anywhere in discussions about the deployment of these systems.

If subsidiarity means anything, then the person who performs a task is best placed to say whether that task still builds them.

They know, better than any consulting firm, what in their day is pure waste, and what requires a judgement no one has formalised. They know it because they do it.

Yet it is precisely this person who is never consulted. The automation scope is decided in the executive committee, on the basis of a process map produced by people who never executed those processes, with the unit cost of the task as the main criterion.

Mechanical result: we automate what is expensive and measurable, not what is useless. These are two very different sets, and their intersection is smaller than one thinks.

What happens when you reverse it

I am not proposing a participatory utopia. I am proposing a method that already exists elsewhere and that we have simply not applied to this subject.

Every serious continuous-improvement approach, in industry, starts with the people who hold the job. They are asked where they waste time, what they do twice, what forces them to wait. This is not workplace democracy: it is production common sense, and it has been taught in every engineering school for forty years.

Applied to automation, it gives a simple device:

Before any tool purchase, ask each team concerned to produce two lists. What I would like never to do again. What I do not want taken from me.

Both lists are instructive, and the second more than the first. In very different settings, I have seen people designate as untouchable a task their management considered an obvious candidate for automation — because it was the only moment of their day when they spoke to a client, or the only place where they understood what their company made.

These lists do not decide. The manager decides, that is their role and their responsibility. But a scope defined without them is defined by the unit cost alone, and it will mechanically produce the second scenario from a moment ago.

The objection: and if no one is consulted?

It comes at once: no company under margin pressure has time for this, and the competitor who does not do it will move faster.

That is true, and it is why I will never present this device as a voluntary good practice. Voluntary good practices do not survive a difficult quarter.

Episode 6 examines what can make them mandatory, and at what price. I only note here that the faster-competitor argument is the same one that was used, for a century and a half, against limiting working hours, against the weekly rest and against machine safety. It was not false. It simply was never a sufficient reason.

The African case, the hardest case

We have to face this point, because it makes the criterion much less comforting.

Over a large part of the continent, the question of who has the power to define the scope does not even arise, for a simple reason: neither the chief financial officer, nor the accountant, nor the manager defines anything. The system is bought off the shelf, configured elsewhere, trained on data that do not come from here, and it arrives with its categories already made.

A management package designed for a European company encodes a European organisation of work. A language model trained massively on English and French handles badly what is said in Lingala, Wolof, Kikongo or Derja — and the company using it does not notice, because it has no way of measuring what it does not understand.

At the scale of a continent, it is the same asymmetry as in the accounting department, one rung higher: those who bear the consequences are not those who design the systems.

Magnifica humanitas formulates it at that level, and it is one of the passages where the text is sharpest: artificial intelligence must serve the whole human family, and not the power of a few. The Pope insists there on the need for those who design and those who bear to build together, failing which the future will be built for a handful.

That is a geopolitical thesis, and it has a very concrete industrial consequence for us: digital sovereignty is not measured in square metres of clean room, nor in the percentage of data hosted locally. It is measured by the capacity to say no to a configuration, and to impose another.

Open question: in your organisation, who decided what the machine would do? And how far is that person from the job concerned?

Episode 4 — The hour given back does not go where you think

Let us grant the two previous episodes. The machine took a task that no longer taught anything, and the person concerned had a say in the scope.

What remains is the question that decides everything: what becomes of the hour?

The Tuesday-at-four story assumes a conversion. The productivity gain was turned into available time for one person. This conversion seems obvious. In reality it is the rarest event in this whole affair, and economic history gives an incriminating account of it.

Three precedents that should make us cautious

The first is domestic. Running water, the washing machine, the refrigerator and the vacuum cleaner reduced by a considerable factor the time needed for any given household task. Total time spent on housework, by contrast, did not fall in the same proportions — and for long periods it did not fall at all. Standards rose: we wash more often, we iron what we did not iron, we clean what we did not clean. The unit gain was absorbed by an increase in volume.

The second is office work. The word processor, the spreadsheet and email divided the cost of producing a document by ten and the cost of sending a message by a hundred. No one worked less. We produce more documents, we send more messages, and email into the bargain stretched the working day beyond its bounds — on the train, in the evening, on Sunday.

The third is industrial and older. The productivity gains of the twentieth century were converted into free time — the forty-hour week, paid holidays, retirement — but never spontaneously. Each conversion was the result of a balance of power, a negotiation, a law. Where the law did not exist, the gain went entirely to output volume.

Three cases, a single lesson:

Time that is freed but not institutionalised is reabsorbed.

That is the third criterion, and it is the one that hurts the thesis I defend most, because it prevents it from being a spontaneous piece of good news.

What this says about the Tuesday at four

This chief financial officer goes to pick up his daughter. How?

Not because the machine produced an hour. A machine does not produce an hour: it produces an available capacity, which is immediately claimed by everything that was waiting.

He picks up his daughter because he decided that Tuesday from four to six would not be available, and because he is in a position to make that decision hold. No one in his company is going to ask him why he is unreachable — he is the chief financial officer.

The accountant in his department who wanted to do the same would have to request it, justify it, and obtain it. She probably would not obtain it, and above all: even if she obtained it, she knows what it costs, in a career, to be the one who leaves at four.

The four o’clock hour is therefore not an effect of artificial intelligence. It is an effect of status, made possible by artificial intelligence.

That is not nothing — without the productivity gain, the decision would have been impossible even for him. But it is very different from a general promise, and mistaking it for a general promise is exactly the mechanism by which a company sells its teams a moral good it will not give them.

How time gets institutionalised

There are not thirty-six ways, and none of them is new. They are all unpleasant for whoever puts them in place, which is why people prefer to talk about company culture.

Through the written rule. A meeting-free slot, a maximum working span, a right to disconnect that is actually equipped, a workload capped by volume rather than by hours of presence. What is not written down does not exist.

Through the contractual counterpart. This is the most honest and the rarest path: when an automation deployment is decided, an identified share of the gain is assigned in advance to a reduction of working time, to higher pay, or to a training budget. Written before, not after.

Through the collective. A company agreement, an industry-wide convention. This is the historical mechanism for converting productivity gains, and it is the one today most weakened, especially in the sectors most exposed to automation.

By default, when none of this is done: the gain goes to the margin. Not out of cynicism, but because it is the only channel open. A capacity freed without an assigned destination returns to the operating account, as water flows to the lowest point.

What this changes in public debate

I draw from this episode a consequence that will displease almost everyone.

The question “is artificial intelligence going to destroy jobs?” is badly posed, and it takes up all the room. The question that really decides people’s fate is: what is the productivity gain allocated to, and who decided it.

That is a question of distribution, not of technology. It has been asked for two centuries under other forms, and we know quite well the instruments that allow answering it.

In one sense it is reassuring: we are not facing an unprecedented problem. In another sense it is worrying: these instruments — collective bargaining, social legislation, employee representation — are in a state of historical weakness at the very moment the shock arrives. And they are almost nonexistent in the sectors that will take the shock first among us: platform work, back-office subcontracting, the call centre.

Open question: the last time your team gained time on a task, where did that time go? Would you know if someone asked you?

Episode 5 — You do not become better by doing less

I have put forward three criteria. Now I have to deal with the objection that targets all three, because it is better than they are.

I will begin by strengthening it.

The objection, in its strong version

It has three parts.

One. This whole series rests on a presupposition that has never been demonstrated: that an hour removed from constrained work becomes an hour of human growth. Nothing establishes it. The hour given back to the chief financial officer could just as well have gone to two extra meetings, to a third board mandate, or to his phone. Statistically, that is even what happens most often. A whole discourse of liberation has been built on a particular case where someone decided to go pick up his daughter, and that case is cited precisely because it is rare.

Two. Worse: the competence that allows him today to supervise the machine was built by the tasks he has just handed over to it. He knows a reconciliation is wrong because he has done ten thousand of them. He senses that a consolidation does not hold together before verifying it. That intuition did not fall from the sky: it is the deposit of fifteen years of thankless work. The chief financial officer who will be thirty in 2040 will never have done those ten thousand reconciliations. He will supervise a system he has never been able to match, and he will validate what seems plausible to him.

Three. Finally, the vocabulary shift in this series is suspect. “Becoming better” has so far been used to designate two very different things: being a better professional, and being a better father. The second is a considerable moral good and it has nothing to do with productivity. Confusing them allows exactly one thing: selling a moral gain to justify an economic loss. The department lost three jobs, but the executive sees his daughter — and that is presented as a balance sheet.

This objection is the best thing in this series. I am going to concede the first two points to it.

The second point is already demonstrated, and I contributed to it

On the building of competence, I have nothing to add, because I spent seven texts on this subject a few weeks ago, about school, and the conclusion holds here word for word.

Supervision is not learned through explanation. It is learned through the repeated practice of the gesture one will later supervise. And that practice, as soon as it ceases to be necessary, disappears — because no one spontaneously chooses the slow version of a task when the fast version produces an acceptable result.

Civil aviation drew from this the only serious conclusion I know: when friction ceases to be necessary, it must become mandatory, planned and costly, otherwise it evaporates. That is the whole purpose of the simulator, of imposed failure scenarios, and of periodic manual flying requirements.

Transposed to a finance department, that means something very concrete that no one does: a young management controller should, at regular intervals, redo by hand a consolidation the system produces in three seconds. Not to produce it. To remain able to sense that it is wrong.

It is expensive, it looks absurd, and it is the only thing that will prevent, in fifteen years, entire control chains from controlling nothing anymore.

The third point is false, and it matters that it is

I do contest the accusation of confusion, because it seems to me to rest on an implicit hierarchy I do not share.

The objection assumes that “better professional” is the serious register, and “better father” the sentimental one — and that using the second to talk about economics is manipulation.

That is exactly the reverse of what Magnifica humanitas says, and it is also, I believe, the reverse of what is true.

The encyclical’s criterion is not the person’s productivity. It is the person. A doctrine that places the human being at the centre cannot consider that time spent with a child is a sentimental adjustment variable while time spent consolidating financial statements is the serious substance of existence. It holds the reverse: work is ordered to the person, not the person to work.

If that proposition is true, then an hour given back to family life is not an added soul in an economic balance sheet. It is the only thing in the balance sheet that really counts, and everything else is the means to it.

That does not whitewash the three jobs suppressed. It only forbids treating the Tuesday at four as a touching anecdote in an accounting demonstration. The two sides of the scale are in the same unit, and that unit is the person.

The first point, and where the objection stops being right

The strongest point remains: nothing guarantees that the hour is used to grow.

I grant it entirely. And I maintain that the objection proves something other than what it thinks it proves.

It proves that the gain is not automatic. It does not prove that it is impossible.

That is a frequent confusion and it is paralysing. From “nothing guarantees it”, one slides towards “so it is an illusion”, which leads to building nothing — and to obtaining, indeed, the illusion.

History nevertheless offers a massive counter-example. The forty-hour week, paid holidays, the end of child labour are not spontaneous effects of the steam engine. They are conversions decided, wrested, written down. Nothing guaranteed them, and they happened.

The right conclusion is therefore not that the thesis of this series is false. It is that it describes a result to produce and not a trend to observe. Which completely changes what needs to be written next: not a forecast, but a list of decisions.

What must be accepted

Three things, which I prefer to write in black and white.

Useful friction becomes a cost. Making someone redo by hand a task the machine produces in three seconds costs paid hours, for an immediate result of zero. It is only justifiable if one has understood what one is buying: the capacity to spot an error in ten years. No quarterly dashboard will ever show that benefit.

It will have to be justified to those who bear it. A young analyst imposed the slow version while the tool is open on the table deserves better than “it is the procedure”. “You are learning to judge what you will no longer produce” can hold, provided you believe it yourself.

Not everything will be saved. Certain skills will leave the general population to survive only among specialists. That is not necessarily a drama — no one knows how to navigate by sextant anymore. It becomes one when the disappearance was neither decided, nor even noticed.

Open question: what is the painful thing in your training that you now measure as having built you? Would you still impose it on someone starting out?

Episode 6 — What must be decided, and by whom

Five episodes of diagnosis. It is time to propose, and I would like to avoid the usual form.

Documents about artificial intelligence in companies all contain more or less the same list: train the teams, raise managers’ awareness, experiment on use cases, put governance in place, equip, measure. The list is not false. Its flaw is being a menu, where each line would have its own effect that would add to the others.

Yet these instruments do not have the same power, and above all — this is the point of this episode — they are not commutative. The order in which they are activated decides the outcome.

I am going to argue that the usual order is almost exactly the reverse of the right one.

Layer 1 — The scope, and who sets it (maximum effect)

This is the decision that contains all the others: which tasks leave, which stay, and on the basis of what criterion.

Today it is taken on a single criterion — the unit cost of the task — by people who do not perform it. Episodes 2 and 3 showed why this mechanically produces the wrong result: we automate what is expensive and measurable, and not what no longer teaches anything.

Acting on this layer means two things, and they are inexpensive.

Establish the scope with the two lists from episode 3, produced by the teams concerned: what I would like never to do again, what I do not want taken from me.

And apply the test from episode 2, task by task: would I want my child to do this for thirty years? That test is not a kindness. It is a filter that removes automations that are profitable and that impoverish.

An organisation that does only this, and nothing else on the list, already gets the essential.

Layer 2 — The counterpart, written beforehand (strong effect, never done)

Episode 4 established it: a gain that is not allocated returns to the margin, because that is the only channel open.

The counterpart consists in opening another channel, and writing it before the deployment — not after, because after there is nothing left to distribute.

Three possible forms, combinable: a share of the gain allocated to reducing working time; a share allocated to pay; a share allocated to a retraining budget for the jobs identified as exposed.

The important word is identified. A company that deploys knows, to within six months, which jobs will become obsolete. Knowing and not saying it is a choice, and it is the majority choice, because saying it triggers early departures and conflicts.

I think it is a bad calculation, including for the company. A retraining that starts eighteen months before the suppression succeeds. A retraining that starts on the day of the announcement is an indemnity disguised as training.

Layer 3 — Maintaining competence (deferred effect, invisible)

This is what episode 5 made mandatory: if formative friction no longer maintains itself, it must be planned.

Concretely: periodic exercises without the tool, on the gestures that ground the judgement of the profession. And, to measure that supervision still works, the deliberate injection of known errors into the workflow — a document containing a false figure in a file review, a deliberately faulty calculation note in a verification.

It is not a trap set for the teams and it must absolutely not be presented as one. It is a measuring instrument, and it is the only one I know that gives information before the accident rather than after. Without it, a control chain may have ceased to control anything for years without anything signalling it: a competent supervisor and an incompetent supervisor validate exactly the same things as long as the machine is right.

Layer 4 — The tool and its purchase (real effect, single effect)

I put it last, and it is not a value judgement: without a tool, none of the above has an object.

But alone, it produces nothing.

Buying without having decided the scope gives the same work in the same place, faster, with the same people more tired. Buying without a written counterpart gives a gain that goes entirely to the margin and a team that has understood it. Buying without competence maintenance gives a validation chain that empties itself in silence.

The tool is a multiplier. Multiplied by zero, it makes zero.

The order, and why it is always the reverse

Scope → counterpart → competence → tool.

The real order is exactly the reverse: you buy the tool, you announce a training plan, you discover the scope as you go, and the counterpart is never discussed.

This is not incompetence. It is a rational consequence of management constraints. The tool is visible, it is decided in a quarter, it appears in the annual report, and it reassures a board worried about missing something. The scope is invisible, it takes months, it forces you to listen to people you do not usually listen to, and it produces its effects after the mandate of whoever decided it.

In other words: the most effective instrument is the one that pays least, politically, inside an organisation.

Naming it is not enough to solve it. But as long as it is not named, we will keep explaining deployment failures by a defect of change management, when they come from a problem of sequence.

Babel, and why the image is more precise than it looks

Magnifica humanitas opens with an alternative: erecting a new tower of Babel, or building a city where people live together.

At first I found the image too convenient. I now believe it is very exactly in its place, provided one remembers what, in the story, makes the tower fail.

It is not its height. It is not the technique — the bricks and the bitumen work perfectly, the text says so. It is that the builders stop understanding one another.

An organisation that deploys a system without those who execute it having been able to say what still builds them; a country that imports tools whose categories were fixed elsewhere; a continent whose languages are not in the data: these are three versions of the same defect, and that defect is not technical. It is a defect of common language between those who design and those who bear.

That is very precisely what the encyclical points to when it affirms that only a construction associating both — the designers and those who bear the consequences, the rich countries and the poor countries, the centres and the peripheries — can produce a future that is not reserved for a few.

I am not obliged to share the theology to recognise that the diagnosis is good.

Three remarks about our economies

The shock will arrive through subcontracting, not through our head offices. The most exposed jobs on the continent are not in finance departments: they are in call centres, offshored data entry, data annotation, moderation. That is, in the activities where collective representation is weakest, contracts are shortest, and the contracting party farthest away. None of the four layers above applies to them spontaneously.

The informal economy changes the nature of the problem. Where most employment is outside salaried work, the contractual counterpart has no legal support. I have no instrument to propose for this case, and I say so in episode 7, among what this series has not covered.

Language is an economic variable. A system that works badly in the language in which one actually works produces two effects: it frees no hour, and it forces one to go through a second language to be understood by the machine. The productivity gain is then captured by those who already speak the language of the tool. That is the subject I pursue elsewhere, and it intersects with this one more than it first appears.

Open question: in your organisation, if you could act on only one of the four layers for three years, which one, and why?

Episode 7 — What I would do with a company and three years

I began this series with a phone call and with an unease.

A man glad to go pick up his daughter, and embarrassed to be. I believe I now understand where the unease came from: he felt that the two facts were linked without knowing how, and that he did not have the vocabulary to hold both in the same sentence.

Here is what I believe I have found in six texts.

The four criteria

The work that builds. Automation enlarges a person when it takes from them a task that no longer teaches them anything. The test: would I want my child to do this for thirty years. This criterion depends on no technical forecast, and it sometimes cuts against the employer.

The power to define. The benefit of an automation goes to whoever decides its scope. That is not a property of the technology but of the hierarchy, and it is the very concrete translation of the principle of subsidiarity: the person who performs a task is best placed to say whether it still builds them.

The institution of time. A productivity gain not allocated in advance returns to the margin, because that is the only channel open. Three historical precedents show it. Free time was never a spontaneous effect of the machine: it was always a written conversion.

The maintenance of competence. These three criteria assume people capable of judging what the machine produces. That capacity is built by the tasks one has just delegated. It will therefore no longer maintain itself, and it becomes a cost to be accepted.

Above the four, the encyclical’s criterion, which orders them: the human being at the centre, never at the periphery. It is not a policy. It is a criterion, and it is exactly what was missing.

What I would do, with a company and three years

I do not run a large group and I do not claim an expertise I do not have. But a series that does not end in decisions is only a commentary.

Year 1 — the scope, and nothing else.

Choose two departments, not twelve. Have the two lists produced there by the teams themselves, job by job. Apply the child test. Derive from it a written scope, published internally, with the explicit mention of the jobs that will become obsolete and of the horizon at which that will happen.

Deploy no new tool that year. The scope will do the work, and it will do it better than a pilot.

Year 2 — the counterpart, and the retraining.

Write, before any deployment, the destination of the gain: what share for time, what share for pay, what share for training. Negotiate that writing with those concerned, and do not merely announce it.

Start the retraining of the jobs identified the previous year, eighteen months before their disappearance. It is the least visible expense of the plan, and the one that decides whether it will be lived as progress or as betrayal.

Year 3 — the tool, and sorting out fatigue.

Deploy, at last — on a decided scope, with a written counterpart and a retraining under way.

And do the thankless work of episode 5, trade by trade, with those who practise it: distinguish the fatigue that builds from the fatigue that wears down. Plan the first, remove the second without hesitation. The second frees the hours that fund the first.

What I would not do

Banning. Not on principle, but because it is unenforceable, that it can be bypassed in one gesture, and that it moves use out of the hierarchy’s sight — where it is no longer correctable, and where no one answers for anything.

Promising that there will be no job cuts. It is the most frequent and the most destructive promise, because it is almost always belied, and it destroys in one stroke the credibility of everything else in the plan. Better to say which jobs, and on what horizon.

Selling the freed time as a moral argument. That is the objection from episode 5, and it is right on this precise point: a company that showcases fathers who see their children while it cuts jobs without a counterpart is not doing humanism, it is doing communication. The difference can be seen in one place only: is it written in an agreement, or in a brochure.

Waiting for the technology to stabilise. That argument assumes it will stabilise. Nothing indicates it. The four criteria of this series are not technical: they have no reason to wait.

Who bears the burden of proof

The debate is asymmetrical, and it should not be.

Those who promise that the job balance will be positive must say where, when, and for whom — nominatively, in the company concerned, on a horizon compatible with a working life. An aggregate balance over twenty years is not an answer to a forty-eight-year-old person.

Those who want to ban or freeze must demonstrate that the ban is enforceable, and that it does not merely make use invisible.

And I, who propose criteria, must say what would refute them. So I will say it. If, in ten years, one observed that the hours freed by automation had been distributed more or less equally across hierarchies, without any written rule having imposed it, my third criterion would be false and a good part of this series with it. I do not believe it — the three precedents in episode 4 all point the other way. But it is an empirical, measurable question, and not a matter of opinion.

Three things this series has not covered

An honest series says what it left aside. There are three holes, and I know them.

Weapons. It is the harshest passage of Magnifica humanitas, and I deliberately set it aside: the encyclical asks that artificial intelligence be disarmed, and it devotes a passage to the transformation of the conduct of war. I chose to keep this series on work, because a text cannot carry everything. But I note that the criterion is the same there, only sharper: a life-or-death decision delegated to a system is the limit case of the human being moved to the periphery. This subject deserves better than a paragraph.

The informal economy. All my instruments — negotiated scope, written counterpart, funded retraining — assume a salaried framework, an identifiable employer and a form of collective representation. Over a large part of the continent, none of this exists: work is informal, the contract is oral, and the employer is sometimes the client himself. What the reasoning becomes in that framework, I do not know, and I will not fabricate an answer to close a series properly.

The theology. I read this text as a document of social doctrine and I took from it what was useful to me: a criterion, principles, a way of asking the question. It is a partial reading and I own it, but it has a cost that must be named. The reason the encyclical places the person at the centre is not a methodological preference: it rests on what this text affirms about the human person, and which I have left aside. Can a reader who does not share that affirmation keep the criterion all the same? I believe yes, and I cannot demonstrate it here.

The Tuesday, at last

I called this chief financial officer back before publishing. I wanted to know whether the Tuesday still held, six months later.

It holds. He told me it had become the thing he defended most fiercely in his calendar, more than any committee — and that no one in his company had done the same since.

There, I believe, is this whole series in one sentence. The hour exists. It is real, it is not a figure of speech, and a machine made it possible.

It went to a single person, the one who could take it.

Nothing in the technology says that this must remain so. Nothing says the opposite either. That is precisely why it is a decision, and why presenting it as a trend is the best way never to take it.

Open question, the last one: if your work gave you back an hour next week, who would you give it to? And what, today, stops you from taking it?