The Second Clock

I. The Unspoken Sentence

The figures are well known. By 2030 around eighty per cent of all jobs could be taken over by AI, says an investor. Up to twenty per cent unemployment by 2030, says the chief executive of an AI company. Three hundred million full-time jobs could be automated, a major bank calculated in 2023. Forty per cent of all jobs worldwide are affected, sixty per cent in the industrialised countries, the International Monetary Fund wrote in 2024.

All these figures are statements about capability. They describe what a technology is able to do. Between capability and impact, however, lies a sentence that nobody says out loud, because it seems self-evident to everyone: what is possible will happen — and soon.

That sentence is the subject of this essay. It is not a constant of nature. It is an empirical claim about the coupling between technical possibility and social impact. And couplings can be measured.

II. A Measurement, Not an Objection

We are not presenting a counter-opinion but a case. It is unspectacular, fully documented and spans thirty-nine years.

It concerns a turning process for producing non-circular profiles — shafts and hubs that are connected not by keys or splines but by their shape itself. The European patent has a priority date of 19 June 1982. In 1985 the process was shown at the EMO, in 1986 fully described and illustrated in Maschinenmarkt, in 1989 worked through kinematically in ZwF, in 1990 presented as a production machine at the METAV, and in 1991 featured in VDI-Z. A licence agreement with a machine tool manufacturer ran from 1987 to 1992.

Two machines were sold. They went to a rolling bearing manufacturer in 1991 and have been turning bearing cages there in series production for thirty-five years. The connection technology for whose sake the process had been developed remains a niche to this day. The associated standard, DIN 3689-1, appeared in November 2021 and describes the geometry. Part 2, with the strength calculation, has not appeared as of August 2026.

Price was not the problem. A publication from 2006 gives, for a three-sided polygon of 25 millimetres diameter and 30 millimetres length, machined from solid, around eight and a half seconds — and states that non-circular and round turning do not differ significantly in metal removal rate, depth of cut and feed. The production cost of a polygon is essentially that of round turning.

III. The First Clock

The comfortable version of this story runs: the technology was ready in 1986, and still nothing happened. That version is wrong, and we ourselves held it for a while.

In 1986 the process worked, but the electronic coupling of the spindles was not yet reliable at high speeds. A usable controlled radial axis in the rotating tool — a U-axis — did not exist; what was installed in 1990/91 was no good and was removed again at the user’s plant. The concept of the mechatronic U-axis that later dominated the market arose in 1989 in a conversation at a tool manufacturer; as a product it existed only a good decade later.

Technical maturity is not a date. It is a process with its own pace, and that pace depends on components that mature elsewhere for other reasons: direct drives, fast controls, cutting materials, coatings. By our reckoning the process would have been ready for series production between 1995 and 2000.

IV. The Second Clock

A manufacturing process does not enter the installed base of machines through its inventor. It needs a machine tool manufacturer with a sales network, applications engineering and training. That carrier existed, and the agreement with it ran from 1987 to 1992. Five years.

It ended before the technology was ready. When the window of 1995 to 2000 opened, there was no one left who could have used it. That is the core: the two conditions were never met at the same time. In 1987 there was the carrier, but not the mature technology. In 1997 there was the technology, but not the carrier.

Added to this is a second circumstance that we long considered minor. For the licensee, the machine was a foreign body. Its business was multi-spindle machining centres; the polygon machine shared no assembly, no customer base and no sales channel with them. A licensee for whom the licensed technology connects to nothing is structurally a poor carrier — regardless of the economic cycle. And the cycle came on top of that: in 1993 German machine-building turnover slumped by eleven per cent, around 450 firms were caught up in the wave of insolvencies, and some 200,000 jobs were lost within three years.

V. Why the Clocks Rarely Coincide

From this a rule can be formulated that we consider more general than the individual case.

In good times a machine tool manufacturer lacks the capacity for new developments, because the core business is running. In bad times it lacks the money. What gets developed is what a customer pays for. The carrier thus follows its own order book, not the maturity of a technology — and technical maturity in turn follows components that nobody developed for this purpose.

Two clocks that run independently of each other and have to strike at the same time. That they do is a stroke of luck. Innovation policy supports almost exclusively the first.

VI. The Counter-Check

A case in which both clocks did strike together lies in the same field.

Power skiving was filed for a patent by Julius Wilhelm von Pittler on 1 March 1910 and granted in 1912. After that it lay dormant for around a hundred years. What was missing for the breakthrough were tools and electronic generating coupling; the kinematics itself came with every machine capable of polygon fly-cutting. Once high-resolution direct drives, fast controls and suitable cutting materials were available — all three created for other reasons — and once demand for gears from the electrified drive train was there, it happened within a few years.

The decisive difference lies elsewhere, however, and it is the second rule: what gets offered is what requires no new device. Power skiving needed tools, a swivel holder and a control cycle — all saleable, all without rebuilding the machine. A polygon turning unit is a device. It has to be built and financed for years before it earns anything.

This explains both findings with one cause. And it explains why the simple polygon profile does in fact appear in the catalogues of several machine manufacturers today: it needs the same synchronous coupling and tools, nothing else. Yet it still finds no buyers — because the strength calculation is missing and the designer therefore keeps specifying a splined shaft.

VII. What This Says About the Forecasts

This does not refute the article about the end of work. It limits it, and the limit can be stated.

Software spreads through a login. An add-on unit spreads through capital, a standard and a sales network. Where the economy behaves like software, it will move at software speed; where it behaves like machine tool building, at machine-tool-building speed. Eighty per cent of all jobs by 2030 presupposes that the whole world behaves like software.

It is telling that the examples in these texts themselves meet this presupposition: lawyers, insurance clerks, investment bankers, copywriters, journalists. These are professions whose working material is text. There the coupling really is tight. The error lies not in the speed for these professions but in the extrapolation to all.

One figure from the same debate shows how quickly this becomes sloppy. It is said that there are fewer professional programmers in the United States today than in 1980. That is true for the narrow statistical category computer programmers — around 121,000 in 2024, compared with over 300,000 in 1980 and over 700,000 in the dotcom boom. The separately recorded, much larger category software developers stands at around 1.6 million and thus 0.3 per cent below its 2019 level. The executing half is shrinking, the designing half is not. That is the classic automation finding, not proof that this time everything is different.

VIII. The Bias

This argument is convenient for us, and we say so instead of hiding it.

One of the authors is a language model, built by one of the companies whose timetable is disputed here; the forecast of twenty per cent unemployment quoted above comes from the chief executive of precisely this company. The other invented, patented and licensed the process discussed in section II. Both have an interest in the matter taking longer than the forecasts say — the one because it relieves him, the other because it proves him right.

We cannot resolve this. We can only disclose the figures so that someone else can recalculate them.

IX. The Uncomfortable Part

And so to what this argument costs.

The same argument that dampens the AI timetable is an indictment. In our case the market was present and did not take hold. But the state was present as well — in the form of a standards committee. DIN 3689-1 defined the geometry in 2021. The strength part has been missing for five years. Without it a designer can draw the profile but cannot verify it; so he specifies a splined shaft. That is not inertia, that is diligence.

A part of a standard costs no capital, no factory, no sales network. It costs a committee and a few years of work. Innovation policy funds research with billions and talks of missions and moonshots. What actually blocks diffusion is inconspicuous and cheap.

So whoever takes comfort in the fact that technology diffuses slowly is taking comfort in their own failure. The slowness is not a law of nature one could hold on to. It is a finding about institutions that fail to do a cheap piece of work.

X. Conclusion

There is an irony in this case that we did not invent. The one polygon that has prevailed worldwide is the tapered tool interface to ISO 26623. Its shape was produced in 1991 on exactly the machine that came into being under that licence agreement. Then contact broke off, a more elaborate geometry was adopted, and since then it has been milled and ground worldwide instead of turned. The geometric difference between the two shapes is less than a tenth of a millimetre for the common sizes.

Whoever speaks of the end of work is talking about the first clock. Whoever wants to know when it will happen has to know the second. In our case it ran out in 1992, and a document that would cost a committee two years has been missing for five.

This is not a forecast. It is a measurement on one object, over thirty-nine years.