The Coding Guild Is Over. The Factory Is Hiring (About Ten of Us).
In which the author, prompted by a Sunday phone call nobody answered, declares the profession doomed, is told by a friend that the machines can’t even code properly yet, concedes the word “yet”, and — instead of fighting the excavator — discovers he has been writing its operating…
Annons
Annons
In which the author, prompted by a Sunday phone call nobody answered, declares the profession doomed, is told by a friend that the machines can’t even code properly yet, concedes the word “yet”, and — instead of fighting the excavator — discovers he has been writing its operating manual himself.…one shovel, one excavator, one rather familiar hole…This is an essay about the end of something, and I should warn you up front that I am not entirely sure what. The obvious candidate is programming, which has been pronounced dead roughly once a decade since it was born and keeps turning up at its own funeral to ask about the catering.The less obvious candidate is older and rather more uncomfortable: a bargain struck centuries ago between the people who knew how to make things and the people who paid for them — a bargain our profession signed, without reading the small print, about twenty years ago.To get to the bottom of it, we will need a Sunday phone call, a medieval guild, a car plant in Michigan, a statistician with a control chart, a Danish flint knapper who was very good at his job at precisely the wrong moment, and a fair amount of my own professional dignity.The machines are in this story, of course. They are just not the villains — which, as it turns out, is the bad news.I. We Are So F***ed (Or: A Sunday Call Nobody Answered)“Ducunt volentem fata, nolentem trahunt.”(Fate leads the willing and drags the unwilling.)— Seneca, quoting Cleanthes, in a line that reads today like a restructuring memo…a phone, ringing, unanswered, on a perfectly calm Sunday…This one is The Finger’s fault.A friend of mine goes by the charming nickname of The Finger — for reasons known only to his closest friends and to the more niche of his chroniclers — and is something between a team lead and the manager of a development team at a sizeable company.A company which, to put it delicately, sits several rungs higher on the ladder of digital evolution than the one in whose structures I currently function. In my defence: we are a manufacturing company, and they are commerce through and through, so the difference in digital DNA is hardly a scandal.It is, however, an excellent pair of vantage points. He looks at the factory from a floor or two further up; I look at it from somewhere near the boiler room.That difference in altitude turned out to be the whole point. Setting the problems my team wrestles with against the direction — and the problems — of his more operationally advanced teams did something that neither of our situations could do on its own: it turned two snapshots into a trajectory.His present looked suspiciously like a plausible version of my future. And somewhere in that comparison I had what can only be described as a wow moment, followed by a hunch I have not been able to shake since: that perhaps AI is not the main story here at all. Perhaps it is only the soundtrack.A necessary caveat before going any further. For reasons that should be obvious, everything that follows about his workplace and mine is deliberately generalised to the level of the industry. Any figures and conclusions drawn from our two corners are a sketch of a trend, not a case study — and I suspect that is less of a limitation than it sounds.We are none of us quite as unique as we would like to believe, and neither, as a rule, are our problems. If you recognise your own employer in what follows, that is not a breach of confidence. It is the point.It started innocently enough. I had spent a quiet Sunday morning doing three dangerous things at once: thinking about what his company had managed to build, thinking about what mine is still struggling with, and reading a book about managing programmers.Any one of those is survivable. All three before lunch is how you end up sending a friend a message that reads, in its entirety and in English for emphasis, “we are so f***ed” — followed, after a dramatic pause of roughly one minute, by the clarification: as an industry and as a profession.I then rang him. No answer. I assumed he was out with the dog. He was, in fact, at home, enjoying the rare luxury of a calm day — which he reported with admirable patience — before replying to my apocalypse with a single word:“Why?”And then, generously, offering me the obvious exit: “You mean automation is going to get us?”No. Or rather — not the way he meant it. The interesting thing about the next hour of messages was that we agreed on almost every observation and disagreed on almost every conclusion, which is the only kind of argument worth turning into an essay. So here is the thesis, stated plainly before I spend thirty minutes decorating it:What is happening to our profession is not an AI revolution. It is the oldest move in the economic playbook — replacing an expensive master with a process — and AI is, first of all, the accelerator — though one that may also be moving the boundaries of what can be standardised at all.Business has been doing this to every guild it ever met. It is doing it to us now. It wants the operating cost of highly qualified labour gone, in favour of processes, standards, production lines, automation, and the delegation of responsibility to the system rather than to the person. Business is stupid, as I told him, but it is consistent.Here is the image I keep returning to. Nobody needed a firing squad. We were handed shovels the day we were hired, told the hole was a product, and paid handsomely to dig. Our entire trade — for as long as there has been one — has been the automation of other people’s work: the clerk, the cashier, the dispatcher, the typist.When Marc Andreessen announced in 2011 that software was eating the world, nobody added the obvious footnote: we are the teeth, and the teeth are part of the world. AI is not the execution. It is simply the moment the shovels were swapped for excavators — and, credit where it is due, the hole has never looked better.The Finger, to his credit, offered the most human objection of all. “I’ll be honest, I never imagined still coding at sixty.” I replied, typing too fast, that he was looking at it wrong — that coding was “meandrów an end”. I meant a means to an end. My thumbs, apparently Polish before they are anything else, turned it into meandrów — “of meanders”. Coding is a meander to an end. I couldn’t have put it better deliberately, so I didn’t try.The point stands, typo and all: hammering out code is one thing; capitalising on technology is another. The first is the job the excavator is coming for. The second is the only question this essay is ultimately interested in.But to see why the first was always going to end this way, we need to visit the people who were masters before us — and who also believed, right up until they didn’t, that the masterpiece would protect them.II. The Guild (Or: Twenty Years of Believing We Were Masters)“We can know more than we can tell.”— Michael Polanyi, The Tacit Dimension, inadvertently describing every senior engineer’s handover document…apprentice, journeyman, master, and one cron job nobody dares to touch…The medieval guild was, among other things, a quality system. It just stored the quality in the wrong place.The ladder is familiar even to people who have never thought about it: apprentice, journeyman, master. To climb the last rung, a journeyman submitted a piece of work to the guild for judgement — the masterpiece, in the original and rather less glamorous sense of the word.Pass, and you could open your own workshop, take apprentices, and charge accordingly. The system worked. Medieval shoes, locks and barrels were often excellent. But notice where the excellence lived: in the hands and head of a particular person.This is what Polanyi meant by tacit knowledge. The master cordwainer could not write down how he knew the leather was right. He knew it the way you know a colleague’s footsteps in the corridor. The knowledge transferred, slowly, through years of standing next to him — and when he died, a fair amount of it died with him. In modern terms, the guild ran on a bus factor of one, multiplied across a few thousand workshops.If this sounds uncomfortably like our profession, it is because it is our profession — or was, for about two decades.The guru……knows why the batch job runs at 3:17 and not at 3:00. He knows that the third replica lies about its lag on Tuesdays. He knows which alert to ignore and which one means wake everyone up. None of this is written anywhere, or if it is, it is written in a wiki page last edited by a man who has since left to raise alpacas.The senior……is a journeyman who survived long enough to become indispensable, which our industry mistook for being valuable. The two are related, but they are not the same thing. Indispensable means the system cannot function without you. From a business perspective, that is not an asset. That is an outstanding risk with a salary.The author……has been, for an embarrassing number of years, exactly this guru. Manual monitoring, held together by knowing where production hurts before production knows it. I was proud of it. I am still a little proud of it, which is precisely the problem this essay is about.Honesty clause, as required by house rules: the guilds were not merely quality systems. They were also cartels. They restricted entry, fixed prices, and enforced monopolies on who could make what within the city walls. Romanticising them is as lazy as dismissing them.Our own guild-like instincts — gatekeeping interviews, tribal language wars, the folklore of the 10x engineer — were never pure craftsmanship either. Some of it was simply protecting the price.Which brings us to the people who noticed the price — and decided, with remarkable consistency over five centuries, that they would rather not pay it.III. The Putting-Out System (Or: Business Is Stupid, But Consistent)“People of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices.”— Adam Smith, The Wealth of Nations, describing with remarkable foresight the #engineering-leads channel…a merchant, a cart, and a village just outside the city walls…The guilds were not defeated by a machine. They were defeated by a man with a cart.From roughly the sixteenth century onwards, across much of Europe, merchants worked out something that now seems obvious: guild rules applied inside the city walls, and the countryside was outside them. So the merchant bought the raw wool, carted it to peasant households in the villages, collected the spun yarn, carted it to other households for weaving, and sold the cloth.The Germans called it the Verlagssystem; English historians call it the putting-out system. No masters, no masterpieces, no guild inspection. Just a lot of people doing one narrow step each, at home, for considerably less money.The quality was frequently worse. It did not matter. The cloth was good enough, cheaper, and available in quantities no guild would ever permit.Then came the second half of the move, which is the part people forget. Having routed around the guilds for two centuries, business eventually got the law to agree with it. Revolutionary France abolished its guilds in 1791.Prussia introduced freedom of trade in the reforms of 1810–11. By the middle of the nineteenth century, across most of Western Europe, the guild as a legal monopoly was finished.The pattern is worth writing down, because we will see it again:the guild sets the pricebusiness routes around the guildthe route-around is cheaper and worsethe route-around gets betterthe law, eventually, legalises the route-aroundthe guild becomes a historical societyOur industry has been through steps two and three at least once already. The offshoring wave of the early 2000s was the putting-out system with fibre-optic cables: send the specification to a cheaper village, collect the code, assemble it at home.My own comparison, in our chat, was the shoemakers’ guilds of old versus the contract factories of the large sportswear brands in Asia — a comparison that works precisely because nobody buying the trainers asks who made them, or with how many years of apprenticeship.It is worth stopping here to be clear about whom the joke is on, because the lazy version of this argument is insufferable. The workers in the village were not the problem, and neither are the workers in the contract factory.They are doing a harder job, for less money, under worse conditions. The joke — such as it is — is on the guild, which mistook a legal privilege for a law of nature. And, increasingly, on us — which, as jokes go, is considerably less funny from the inside.Business is stupid, but consistent. It will always ask the same question: what is the cheapest arrangement that produces something good enough? And it will ask it again, and again, and again, until the answer changes.In 1913, in a factory in Michigan, the answer changed rather dramatically.IV. Highland Park, 1913 (Or: The Line Was Ugly, and It Won Anyway)“Any customer can have a car painted any colour that he wants so long as it is black.”— Henry Ford, My Life and Work, the first product owner in history to say no and mean it…a rope, a motor, and a chassis that refused to wait for anyone…The usual story of the assembly line is a story of triumph. The more useful story is a story of how bad it was at the start — and how little that mattered.It began small, in April 1913, with a single component: the flywheel magneto. Instead of one man assembling the whole thing, the job was split into twenty-nine steps for twenty-nine men along a moving belt. Assembly time fell from around twenty minutes to thirteen, and then to five.Encouraged, Ford’s engineers moved the engine onto a line, then the transmission, then — that autumn — the whole chassis, dragged past the workers by a motor and a rope. The Henry Ford Museum, which has every reason to be proud of it, nonetheless describes that first line as relatively crude.Crude, and devastatingly effective. By early 1914, the separate lines had fused into continuous ones, turning out a finished Model T every 93 minutes, against twelve and a half hours under the old stationary method. The price of the car fell accordingly, demand rose, and the rest is the twentieth century.This is the argument I made to The Finger when he pointed out that, on his floor at least, AI-assisted development is producing a remarkable amount of rubbish. The early line was rubbish too. It was rough, it was brittle, it was improved weekly. And it won anyway, because the question was never whether it was good.The question was whether it was good enough and getting better faster than the alternative. The Finger had in fact conceded the general point earlier, having written that AI hadn’t shown everything could suddenly be done much faster — and then, a minute later, appending a single correction, asterisk and all: “Yet*.”To be fair, the part of the story that the triumphant version leaves out is the one that matters most for us. It is also, naturally, the least cheerful.The line was not crude for the car……it was crude for the people. Workers hated it. The museum’s own figures put labour turnover at Highland Park at around 380 per cent by late 1913 — which means Ford was, in effect, replacing his entire workforce several times a year. The deskilled job was so miserable that nobody would stay in it.So Ford paid more……not less. In January 1914 he announced the famous Five Dollar Day, roughly doubling the going rate, with conditions attached. The cheap, replaceable worker turned out to be, at first, rather expensive. The fraction-of-the-rate labour came later, once the line was stable, the know-how was in the jigs rather than the hands, and the labour market had adjusted.This is not a footnote. It is a timeline. It suggests that the transition from guild to line does not start with lower wages — it starts with different wages, paid to a different kind of worker, for a job that is worse to do and easier to learn. The savings arrive afterwards, when the process has eaten enough of the craft.Which raises the obvious objection. If the line produces worse products and miserable workers, surely the master’s quality wins in the end?It doesn’t. And the reason why is the least comfortable part of this essay.V. Quality Is a Property of the Process (Or: Why Better Is Worse for Us)“A bad system will beat a good person every time.”— W. Edwards Deming, probably, though the internet has by now made him say most things…a control chart, a cord, and a line that stops when it should…Here is the sentence I have been circling for four sections: it is not going to get worse. It is going to get better — and that is precisely what is worse for us.The romantic view of craft assumes that the master’s work is better than the factory’s. At the very top, it often is. A bespoke shoe from a great cobbler beats anything off a line. But quality is not only about the peak. It is about the variance — how wide the gap is between the best piece and the worst — and about whether anyone can tell, before it ships, which is which.Craft has a magnificent peak and a terrible spread. The great master makes a masterpiece; the average master makes something uneven; and nobody can reliably reproduce either, because the knowledge lives in the hands. Mass production lowers the peak, but it kills the variance — and in quality, it is the tail that bites.The moment this became a discipline rather than a hope has a date. In 1924, Walter Shewhart, an engineer at Western Electric (whose research arm was about to become Bell Labs), sketched what we now call the control chart: a way of telling ordinary, expected variation from the kind that means something is broken. Deming took the idea, carried it to post-war Japan, and the rest is the reason your car starts in the morning.The philosophical shift is enormous and easy to miss. Quality stopped being a virtue of a person and became a property of a process — measurable, monitorable, and above all, catchable. Defects became easier to find, not harder. And design stopped being the private genius of one master with a bus factor of one, and became the property of the project: its architecture, its specifications, its tolerances.If you work in operations, you will recognise every word of that paragraph. It is the job description of observability. It is also, one cannot help noticing, the job description of whatever replaces the person who used to do it by feel. Me, for instance.To be fair, mass production does not deliver quality automatically, and anyone who claims otherwise has not owned an American car from the 1970s.Detroit……had the line, the scale, and the money, and in the early 1970s produced some of the most unreliable cars of the century. The emblem is Lordstown, Ohio, in 1972, where General Motors set the line to around a hundred cars an hour, against the fifty-odd typical of the industry. Cars came off it in various states of completion, and the workers struck for eighteen days over the pace.The car built there, the Chevrolet Vega, became a byword for unreliability — though, honesty compels me to add, its engineering deserved at least as much of the blame as the line. Mass production treated as pure speed, with the people as interchangeable parts, made mass-produced junk.Toyota……built its system on a principle the Japanese call jidoka — roughly, automation with a human touch — and a famous piece of physical infrastructure: the andon cord. Any worker on the line who saw a defect could pull it and stop production. Not report it. Not file a ticket. Stop the line.That is the crucial correction to the lazy version of my thesis. Quality did not arrive by removing competence from the line entirely. It arrived by moving competence from the product to the process — and by trusting a much smaller number of people to stop the line when the process was wrong.So yes: the process wins. But it wins with people in it. Fewer of them, differently skilled, and holding a cord.Which is a useful thing to know before looking at the shops this process has built — because they are not all the same shop.VI. Four Shops, One Tolerance Dial“It is a truth universally acknowledged, that a single man in possession of a good fortune, must be in want of a wife.”— Jane Austen, Pride and Prejudice, on the original premium market segment…four shopfronts, one dial, nobody at the bench…The obvious objection to everything so far is that craft has not disappeared. You can still buy a hand-stitched bag, a bespoke suit, a pair of shoes made to your feet. True. But look closely at where quality lives in each of these shops, and a pattern emerges: the difference between the luxury shop and the bargain bin is not master versus process. It is where the tolerance dial is set.Every one of them runs a process. They differ in how tight the tolerances are, how much they charge for the tightness, and what role — if any — they leave for a named human being.Temu……kept the process and replaced the master with a tolerance setting of “it will probably arrive”. Wide margins, enormous volume, prices close to zero, and a quality system that is essentially statistical: enough of it works, and the returns are priced in. It is not a failure of the process. It is the process, working exactly as configured.The middle market……sets the dial wherever the invoices close. This is where most of the money and most of us live: the large retailer, the marketplace, the bank, the logistics platform. Tolerances tight enough to keep the customers, loose enough to keep the margin. The Finger’s company sits here, and so does mine, albeit several floors closer to the boiler room.Hermès……is the honest edge case, and worth naming because it looks like a counter-example. As the house likes to tell it, its most famous bags are still made by a single artisan from start to finish. The master, apparently, survives. But look at the conditions: the artisan is trained inside the company, works to rigorously standardised methods, and their individuality exists mostly in the brand’s story. The master kept his hands, and lost his independence. He is the process’s executor and its logo at the same time.By Royal Appointment……is the shop for the toffs, and the one place the master survives fully intact. Savile Row tailoring, shoes built on your own last, shotguns that take years to make for one specific owner. The word bespoke itself, as the Row likes to tell it, comes from cloth that had been bespoken for a particular customer. And in 2008, in a twist this essay could not have scripted better, Britain’s advertising regulator ruled that machine-made, made-to-measure garments may call themselves bespoke too.The process did not just win the market. It won the word. And the surviving institution is perfectly British: the Royal Warrant, granted to tradespeople who supply the royal household. Note what keeps this shop alive. Not quality. Patronage. The product is a marker of status, not a tool — and the landed gentry were, after all, the guilds’ best customers long before anyone else could afford them.Us……spent twenty years believing we were Savile Row, when we were merely the expensive end of the middle market. We thought the premium was for our craft. It was, mostly, for our scarcity — and scarcity is precisely what a process is designed to remove.The software map follows almost automatically. The boutique consultancy with excellent marketing is Hermès. The vibe-coded app thrown together in a weekend is Temu. The bespoke systems built for very rich clients — where nanoseconds or prestige justify any price — are Savile Row: a handful of people, extraordinary rates, and roughly zero significance for the labour market as a whole. And everyone else is in the middle, where the dial is set by finance, not by us.Here, admittedly, the analogy starts to creak, and it is better I say so before a commenter does. Section X will concede that software costs nothing to copy, so its luxury end cannot rest on scarcity of production the way a hand-stitched bag does. What the software Savile Row sells is not the making. It is the signature.The audited core-banking system, the certified medical device, the security assessment a regulator will actually accept — in each, the client pays less for the code than for a named, accountable human willing to put their name to it, and to be found when it breaks. Patronage, in our trade, is spelt liability. A rather less glamorous word, but considerably harder to automate.So much for the shopfronts. Time to walk onto the floor.VII. The Factory Floor, Present Day (A Composite, Mostly)“The future is already here — it’s just not very evenly distributed.”— William Gibson, who said it in interviews throughout the 1990s and apparently never once wrote it down, which is rather the point…a span of control, quietly widening…Everything below the numbers is a composite. Names, ratios, and at least one entire org chart have been changed to protect the innocent, by which I mean me. If you think you recognise the company — you don’t. You recognise the genre, which is worse, because it means it is happening to you too.First, the numbers that are not a composite, because an essay about the end of a profession owes its readers at least one fact they can check.The cleanest of them comes from Indeed, which publishes an index of US software development job postings with February 2020 set at 100. On 11 September 2026 it stood at 76.6. At its peak, in February 2022, it touched roughly 234 — more than three times today’s level, which says at least as much about the price of money back then as about programming.Since the agentic-coding wave began, the index has climbed back by almost 15 per cent, while overall US postings fell by about 7 per cent over the same stretch.Read that twice, because it says two things at once. Postings for software developers sit roughly a quarter below where they stood before the pandemic (about 23 per cent in September, after 27.5 per cent in June), while the wider market has recovered — the trade is diverging from the economy, not following it. And yet the last year has seen a genuine rebound, which Indeed itself is careful to say may be correlation rather than causation.A caveat an economist would insist on, and rightly. These are job postings, not jobs — an index of advertisements against a pre-pandemic baseline, not a headcount of working programmers. And the fall from the 2022 peak has more than one parent: the end of a zero-interest hiring boom, the cost of capital, post-pandemic corrections, offshoring and, somewhere in the mix, AI.Even the senior skew does not prove that companies are swapping juniors for agents; it proves only that, for now, they would rather hire people who already know where the bodies are buried. The numbers describe the shape of the floor. They do not, on their own, name the architect.And now the part a lazier essay would omit. Seven in ten of the additional postings — the net increase between May 2025 and May 2026 — came from senior roles. More than a third came from jobs with AI in the title — the two groups overlap, so please don’t add them up. Neither is what a crude version of my thesis predicts. If the factory were simply replacing masters with cheap hands, the growth would be at the bottom.Instead, the bottom is where it hurts most — juniors are the ones who cannot find the ladder — and the growth is at the top. Indeed’s own researchers float the phrase seniority-biased technological change. Hold on to that. It comes back in section X, and it costs me something.Now the composite.A large company, several thousand people, a substantial fraction of them in engineering. The new organisational target is a pyramid: a quarter senior, half mid, a quarter junior. The span of control — how many people each manager oversees — is moving from six to eight.Nobody has yet announced who goes where, but everyone can read an org chart, and the people the new structure needs fewer of are the managers.I said something unkind to The Finger at this point, and I stand by it while apologising for it. The middle-level coordinator of expensive human aggregates — the role many of us drift into after fifteen years — is at risk of becoming as obsolete within a few years as the overseer of a team of hand sawyers who knew everything about the working methods of the weaving shed. Too far from the low level to be the line’s engineer. Too attached to the old ways of crafting to become its manager.The Finger’s view from his floor is less theoretical and more painful. Development speed, he observes with an engineer’s weakness for hyperbole, has become directly proportional to bad code. The simplest bugs are coming back to production, and there are more of them every month.And he is the one who then has to explain, to people who do not care how the code was written, why something is broken and why millions in gross merchandise value are leaking away.Both of these things are true at once, and that is the point. The line is crude, it is producing defects, and it is still winning — exactly as it did in Highland Park. The question is not whether the defects are real. It is who will be employed to catch them.Which brings us to the toolmaker’s bench — and to the uncomfortable fact that I have been working there for a while.VIII. The Toolmaker’s Bench (Or: Automating Myself, a Field Report)“…the more advanced a control system is, so the more crucial may be the contribution of the human operator.”— Lisanne Bainbridge, Ironies of Automation, 1983, writing the SRE job description four decades early and getting no royalties for it…a runbook, freshly written, by the man it replaces…Here is where the author stops being a commentator and becomes evidence.In the middle of all this, I wrote The Finger a message that I have been thinking about ever since: my own manual monitoring, held together by guru knowledge, is also going to turn into runbooks, automated procedures, observability, and AI-driven reaction.I did not write it as a prediction. I wrote it as a status update. That is, quite literally, what I am paid to do now. I am, in other words, writing my own redundancy notice — in YAML, with excellent test coverage.Site reliability engineering, stripped of its branding, is the discipline of taking what the guru knows and moving it into the line. Every runbook is a piece of tacit knowledge made explicit. Every alert with a documented response is a small apprenticeship that no longer needs an apprentice.Every automated remediation is a journeyman’s judgement turned into a jig. I am, in Polanyi’s terms, spending my working days converting what I know but cannot tell into something the system can tell without me.It would be easy to be gloomy about this. It would also be inaccurate, because the history of the line contains a twist that the gloomy version always skips.The line did not abolish the masters……it relocated them. Highland Park needed people who could design jigs, fixtures, and machine tools precise enough that an unskilled worker could produce a part within tolerance. The toolmakers were a small, highly skilled, well-paid elite — craftsmen of the process rather than of the product. They did not stand at the line. They built it.Toyota did not abolish judgement……it concentrated it. The andon cord meant someone on the line had to know, in seconds, that something was wrong enough to stop everything. That is not unskilled work. That is the hardest call on the floor.And Bainbridge, in 1983, explained why……in a five-page paper about process control that every operations engineer should read once a year. Her central irony is that automation takes away the easy parts of the operator’s job and leaves the hard ones — the rare, abnormal, unpredicted situations — while simultaneously denying the operator the daily practice that would keep them ready for those situations. Her conclusion, in her own plain words, is that the operator needs to be more rather than less skilled.So the toolmaker’s bench is real, and it is where I would rather be than anywhere else on the floor — partly out of ambition, and partly because it is the last place anyone thinks to look when the redundancy list is being drawn up. In software, the jigs and gauges have familiar names: tests, contracts, pipelines, observability — and, if you will allow me a hobby horse, a type system strict enough to reject a part that is out of tolerance before it ever reaches the line. Much of my recent fondness for Rust is exactly this: the compiler as an inspection gauge, letting through only what fits, whoever or whatever wrote it.Now the honest arithmetic, because this section would be propaganda without it.The toolmaker’s bench seats a handful of people for every few hundred who used to stand at the line. That is an illustrative ratio, not a statistic — nobody publishes the figure, and I would distrust anyone who claimed to — but its order of magnitude is the point.The relocation of mastery is real, and so is the shrinking of the room it relocates into. Being one of the people who builds the line is a far better position than being one of the people it replaces. It is not, by any stretch, a solution for the profession. It is a lifeboat with ten seats, and I have been rather quietly measuring the oars.And it is not a new story. We have told it before — loudly, confidently, and wrongly, at least on the timing.IX. The Sinusoid, Obsolescence Edition“The thing that hath been, it is that which shall be; and that which is done is that which shall be done: and there is no new thing under the sun.”— Ecclesiastes, 1:9, the original release notes…the same obituary, reprinted every decade, slightly updated font…Regular readers know the drill: before believing that anything is unprecedented, check whether we have already been here under a different name. We have. Several times. The obituary of the programmer has been reprinted so often that the profession could have lived quite comfortably on the royalties.The end of the programmer has been announced, with confidence and a straight face, roughly once a decade since programming existed:COBOL (1959) — designed to read like English, so that business people could, in principle, write their own programs. They didn’t. A profession of COBOL programmers was born instead, and some of them are still maintaining the result.The program generators of the early 1980s — one British product was marketed, with magnificent hubris, as The Last One: the last program you would ever need to buy. Its 1981 adverts promised a program that would ask you questions in genuinely plain English and write the code itself — the end of programming as you know it. Forty-five years later, the phrasing sounds oddly familiar. It was not the last one.CASE tools and fourth-generation languages (the 1980s) — draw the diagram, generate the system, dismiss the coders.Ed Yourdon’s Decline and Fall of the American Programmer (1992) — programmers abroad, armed with better methods, would produce software more cheaply and to higher quality, and drive American shops out of business. In 1996, with admirable honesty, the same author published Rise and Resurrection of the American Programmer.Offshoring, round two (the early 2000s) — the putting-out system with fibre.No-code and low-code (the late 2010s onwards) — every business user a developer.LLMs and agents (now) — every business user a developer, and this time we mean it.Compressed into its usual arc, the sinusoid looks like this:prophecy of the end → panic → a cheaper substitute appears → the substitute is worse → the profession survives, grows, and declares victory → the substitute quietly gets better → repeatIt is tempting to read that list as reassurance — they were wrong every time, so they are wrong now. And this is exactly where I have to disagree with my own sinusoid.Look again at what each wave actually did. None of them killed the profession. Every one of them killed a layer of it. COBOL and its high-level cousins steadily shrank the share of business programming done in assembly and raw machine code. Compilers and higher-level languages removed whole categories of manual optimisation.Frameworks removed the need to know how half the stack worked. Each wave was declared a failure because programmers survived — while, inside the profession, the work that used to be skilled quietly became a library call.Yourdon’s two books are the perfect emblem. He was wrong about the timing and, in 1992, wrong about the mechanism. But the sinusoid is not a flat line. Each trough is a little lower for the craft, and each peak is a little higher for the process.The waves have been grinding in the same direction the whole time — we simply measured them by headcount, and headcount kept rising for reasons that had nothing to do with craft: cheap money, the expansion of the internet, the mobile boom.The difference now is not that this wave is uniquely magical. It is that the tailwinds which kept hiding the direction — zero interest rates, the post-pandemic hiring bubble that took postings to more than twice their pre-pandemic level — have stopped blowing. The tide went out, and we can see where the waves have been carrying the sand — and, if one squints, the outline of a rather neatly dug hole.Let me be explicit about what all this history is, and is not. It is an analogy that explains, not a model that predicts. The putting-out system, Highland Park, Toyota and a language model share one dimension — the standardisation of skilled work — and differ in almost everything else: who owns the means of production, what coordination costs, what kind of knowledge is being captured.Ford’s line tells us which way the wind has blown for five hundred years. It does not tell us how many programmers will be employed in 2030, and anyone who claims otherwise is selling a webinar.Which does not mean I am right. It means it is time for the section where someone gets to tell me I am wrong.X. The Counter-Argument I Can’t Fully Dismiss (Or: The Finger Was Right About More Than I’d Like)“It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.”— Mark Twain, allegedly, who almost certainly never said it, which makes it the perfect epigraph for a section about certainty…five objections, standing, refusing to sit down…Tradition is tradition. Here is the strongest case against everything above, in the form its best defenders would make it — and no, the rebuttals will not dissolve it.“Yet”……is The Finger’s word, and I concede it outright, as temporary. The research on AI-assisted defect rates is genuinely mixed — it depends on the task, the team, and what you choose to count — but on The Finger’s floor, at least, the simplest bugs are coming back to production. So did the early line. Quality is an argument about tempo, not about direction, and on tempo the line has a two-hundred-year winning streak. This objection does not survive — but it buys time, and time is nothing for people with mortgages.Jevons……is the one that keeps me honest. In the nineteenth century, William Stanley Jevons noticed that more efficient steam engines did not reduce Britain’s coal consumption — they increased it, because cheaper energy found more uses. Cheaper software may simply mean more software, not fewer people writing it. The Finger said as much: there are more and more topics, not fewer. And the data in section VII, inconveniently, shows postings rising again over the past year. The line may shrink each job and multiply the number of lines.Bainbridge and the seniors……stands fully intact, and it hurts. If the operator of an automated system must be more skilled, not less, then the line does not replace masters with cheap hands. It replaces many average craftsmen with a few excellent ones. The senior skew in the postings from section VII is Bainbridge’s irony appearing in the labour statistics. My clean story of “fewer people, and not master craftsmen” is half wrong: fewer people, yes — but quite possibly more masterful ones.The cruelty lands on juniors, who cannot get onto a ladder whose lower rungs have been automated away, and who therefore cannot become the seniors the line will need in ten years. We are, as an industry, eating the seed corn and then writing thoughtful posts about the disappointing harvest.Zero-cost copies……is the objection that breaks my central analogy, and I cannot fully repair it. Shoes and cars are physical: each unit must be made. Software is copied for free. We have had a production line for decades — it is called a compiler, and behind it a deployment pipeline. What is left for humans may not be production at all. It may be design: deciding what to build, under uncertainty, for people who cannot say what they want.That work has never travelled to the cheaper village quite as easily as stitching did, and it is not obvious that it will now. Though honesty requires the obvious footnote: that design and judgement will stay human is a hypothesis, not a safe harbour. What looks more durable is not any cognitive uniqueness of ours, but legal accountability — somebody still has to sign, and a model cannot yet be sued.Revealed preference (the meta-argument)……is the one I find hardest to sit with. Perhaps business was never stupid at all. Perhaps it never wanted masters in the first place — only good enough, at a predictable price — and paid master’s rates for twenty years simply because there was no alternative. If so, the guild did not lose a war. It enjoyed a temporary shortage, and mistook scarcity for value. That is not a tragedy. It is a correction — and the profession’s apparent decline is merely the price of returning to where business always thought it belonged.Who owns the factory……is the question my own epilogue would rather you didn’t ask. Knowing how to build a line is not the same as owning a market for what comes off it. If producing software gets cheaper for everyone, it gets cheaper for my competitors too — and the obvious result is more supply and thinner margins, not a quiet income for every engineer with a side project.Worse, the rent may simply migrate to whoever already owns the scarce things: distribution, customers, data, capital, brand. Cheap factories have more often made factory owners rich than factory builders — and the people with the most capital to build them are not, as a rule, reading Medium essays at midnight. I have no clean rebuttal to this one. I concede it, and the epilogue has been adjusted accordingly.Nor should responsibility be mistaken for ownership — a confusion this essay has occasionally flirted with. The person who carries the pager, signs the risk register or answers to the regulator does not automatically share in the rent. One can end up holding far more accountability for an equally predictable salary: a promotion in everything except the part that pays. Bainbridge’s operators were more skilled than the people they replaced. Nobody ever claimed they owned the plant.It helps, at this point, to separate three things this essay has occasionally let blur into one. Automation of production means that less work is needed to make each unit. Deskilling means that less skill is demanded of whoever does the work that remains. Concentration of value means that a larger share of the profit goes to whoever owns the process, the market or the capital. They travel together often enough to be mistaken for one another, but they are not the same animal.The first does not have to cause the second — Bainbridge suggests it may do the opposite — and the second is not required for the third. Of the three, the last is the one I would watch most closely, and the one no job title protects against, however many words it contains.Taken together, these do not kill the thesis — the direction still looks right to me — but they change its shape. The line wins; but it may need more lines, it will need better operators, and the hardest work may stay stubbornly human.Confidence rating, per series convention: about sixty-five per cent that the direction is right, thirty-five that Jevons and design will keep the profession larger than I fear — measured after coffee, which regular readers know inflates my confidence by ten points. Before coffee, call it fifty-five, and ask me again tomorrow.There is, however, one objection I have not raised, because it is not really an objection. It is a confession on behalf of the whole profession.XI. We Were Never a Guild (Or: Freelancers with a Coat of Arms)“All that is solid melts into air, all that is holy is profaned.”— Karl Marx and Friedrich Engels, The Communist Manifesto, who would have made formidable product managers had anyone let them near a roadmap…a coat of arms, self-designed, never registered…If the guilds lost to business, it is worth asking which crafts did not lose — and why.The answer is uncomfortable. The professions that kept their guild-like status into the twenty-first century are, overwhelmingly, the ones that captured the law. Doctors, lawyers, pharmacists, architects, chartered engineers — and, in my part of the world, notaries, appointed by the Minister of Justice, whose maximum fees are set by ministerial regulation rather than by the market.These are guilds in everything but the medieval costume: a long apprenticeship, an examination, a licence, and a monopoly on certain acts. Nobody can vibe-code a conveyancing deed in Poland, not because a machine couldn’t draft it, but because the law says only a notary may sign it.These professions are not immune to automation. Radiologists have been told they are finished roughly as often as programmers. But the licence buys them something we never had: the process has to pass through them, even when it no longer needs their hands.Now look at us.We never built a guild. No licence, no examination that mattered, no reserved acts, no professional body with teeth. In the United States, a professional licensing examination for software engineers was introduced in 2013. By 2018, a grand total of 81 candidates had asked to take it, and it was discontinued after April 2019. Perhaps the exam was simply worth little on the market, or matched the actual work poorly — both entirely plausible. But whatever the reason, it is hard not to read it as a profession voting, with its feet, against becoming a guild.There is one exception, and it is not a small one. Where software can kill people, something remarkably like a guild survived after all. In aviation, medical devices and cars — under standards with names only a compliance officer could love, such as DO-178C, IEC 62304 and ISO 26262 — the process itself becomes effectively mandatory. Certification demands documented, reviewed, and, at the higher levels of criticality, independently verified work, and accountable humans stay in the loop because the regulator requires it, not out of sentiment.It is not a licensed profession in the notarial sense — nobody needs a personal licence to write avionics code — but it is the closest our trade has come to one. Nobody vibe-codes flight-control software — or at least, nobody gets to fly it. The guild did survive in our trade, in the one corner where the law arrived before the market. Everywhere else, we left the door open and called it agility.And for twenty years, it did not matter. Scarcity did the work a licence would have done. There were not enough of us, so we could set terms, choose employers, and change jobs for a thirty per cent pay rise. We had the income of a guild, the collective bargaining power of a WhatsApp group, and the self-image of master craftsmen — and we mistook the combination for a permanent feature of the universe.We gatekept, certainly. Whiteboard interviews, language tribalism, the folklore of the genius who cannot be managed. But it was cultural gatekeeping, not legal gatekeeping — and cultural gatekeeping evaporates the moment the work can be done without you.So here is the confession on the profession’s behalf: we were not defeated as a guild. We were freelancers with a coat of arms we designed ourselves and never registered. Business did not have to abolish us. It merely had to stop being short of us.I am not arguing that we should now build that guild. I suspect it is too late — rather like proposing a union at one’s own leaving drinks — and I am not sure the world would be better for it — the licensed professions are not famous for cheap, fast, or humble. But it explains why the ground feels as though it is moving under us faster than under the radiologist: nothing was ever holding it in place except demand.Which leaves one last group to deal with before the ending — the people currently selling us the cure.XII. Flint Daggers in the Copper Age (Or: Become the Knapper of Tomorrow!)“Change is the law of life. And those who look only to the past or present are certain to miss the future.”— John F. Kennedy, Frankfurt, 1963, now available as a slide in every bootcamp’s opening webinar…the finest flint dagger ever made, slightly too late…From around 2400 BC, large and beautifully made flint daggers appeared across southern Scandinavia — so many that Danish archaeologists call the whole era, 2400–1800 BC, simply the Dagger Period. Over the following centuries, Danish flint knappers pushed their craft to a level of technical difficulty that archaeologists still describe as among the most complex stone tools ever produced.The culmination is the so-called fishtail dagger — the most famous, found on the island of Fænø and known as the Hindsgavl dagger, dates, by the National Museum of Denmark’s reckoning, to the very end of that era — roughly 1900–1800 BC.Note the dates. The flint knappers produced their finest work precisely as metal was arriving. The masterpiece of the stone age was made in the shadow of the age that would make it unnecessary.Honesty clause, as required by house rules: the popular story — that these daggers were flint imitations of copper and bronze ones, a doomed attempt to compete with the new technology — is contested. Some archaeologists argue that the influence ran partly the other way, and the dating is messier than the neat version suggests.But one finding cuts through the debate and makes the point better than the folk version ever could: microwear analysis suggests that many of these daggers were never used as tools at all. They were objects of status. Display pieces. Proof of what their owner could afford.In other words, at the very moment the craft reached its peak, it stopped being a means of production and became a luxury. Take this as a metaphor rather than evidence — the archaeology permits other readings — but as metaphors go, it is uncomfortably well cut. By Royal Appointment, circa 1700 BC.Which brings me, with some reluctance, to the bootcamps.You have seen the adverts. Become the developer of tomorrow — with AI! Twelve weeks to a new career! I want to be careful here, because the people who sign up for these courses are not the joke. They are the ones paying for it, often with savings they cannot easily spare, in the honest hope of a ladder that the market is quietly sawing the bottom rungs off. The joke — such as it is — is on the business model.Because the pitch, translated into the terms of this essay, is: become a modern flint knapper, now with copper tools! It treats AI as a better chisel for the old craft — a turbo-boost for programmers — rather than as the thing that changes what the craft is for. It is the fishtail dagger of career advice: exquisite, sincere, and roughly a millennium late.And I want to be equally careful about the opposite error, which is the one I am personally more prone to: the stubborn insistence on knapping flint by hand, out of pride, because the old way was harder and therefore must be more virtuous. That is not principle. That is reenactment pretending to be a business plan.Both errors share the same hidden assumption — that the role of programmer, as we have known it for twenty years, will survive in its present shape, and the only question is which tools it will hold. I do not think it will. I think the role is changing shape under us, as every guild role did before it.So to the role as we knew it, with genuine affection and the solemnity the moment deserves: hasta la vista, baby.And now the ending this essay owes you — which is not a eulogy.Epilogue: The Unit of Sale“The impediment to action advances action. What stands in the way becomes the way.”— Marcus Aurelius, Meditations, a man who ran an empire, kept a journal, and never once had to explain an outage to a CFO…a hole, freshly dug, awaiting concrete or flowers…So no, not a eulogy. A eulogy is what you deliver for something that is gone, and the thing I am describing is not gone. It is changing hands.Let me return to the hole, since I made you look at it for twelve sections. We were handed shovels, told the hole was a product, and paid to dig; now we have excavators. The image is grim only if you assume you know what the hole is for. A hole is a grave or a foundation, depending on what you pour into it.Every stabilisation of a technology — every moment when a craft becomes a process and a process becomes a commodity — has opened a new layer of growth on top of it. When electricity stopped being a marvel and became a socket in the wall, an entire industry of appliances grew on top of it.When servers stopped being hand-built machines in a cupboard and became a line item from a cloud provider, the whole software-as-a-service economy grew on top of that. Nobody mourns the lamplighters very loudly these days, partly because their grandchildren were busy selling refrigerators.The pattern is almost boringly reliable. What gets commoditised becomes the floor for whatever comes next — and the floor is where the next fortunes are made. Not for everyone. Not painlessly. But reliably.Which is why, for all the cynicism of my Sunday messages, I never actually proposed despair. What I wrote to The Finger was that there is no crisis on which money cannot be made — and that the first thing to do is to be among the first to accept things as they are and work out what can be built on them. Ethics are ethics, I added, and mortgages are mortgages. I stand by both halves of that sentence, in that order.Regular readers will notice this is a cousin of the conclusion of The AI Shop, and they would be right: an earlier essay in this accidental series argued for invoicing the vibe rather than fighting it. This is the same idea, one level down. The question is not how to keep selling hours of craft. It is what to own once the craft has become a line.That is the smaller, truer — and admittedly far less marketable — thing that has actually changed. Not the end of programming as a craft, but the end of the premium for merely being able to produce code. And with it, the end of the hour as the default unit of sale.That last part, I should admit, is a hunch rather than a conclusion: hourly billing has survived plenty of automation before, as any lawyer will cheerfully confirm. For twenty years, most of us sold our time to one employer at a guild’s price, protected by a shortage. That shortage premium is eroding — not everywhere, but in precisely the segments where producing code was the scarce part.What remains valuable is not the hour, but the thing that works without you: the line, the tool, the system, the product, the small piece of infrastructure that someone, somewhere, will pay for while you sleep. Technology is not something to run away from — if there is ever none of it, we will be fighting over the last rat with machetes, and the mortgage question will have resolved itself.The point is to stop being the cost on the line and start owning a piece of it — knowing full well, as section X conceded, that owning a piece of the line guarantees nothing about owning a piece of the market. It merely moves you from the column labelled cost to the column labelled risk, which is at least a column with upside.Coding was always a meander to an end. We simply forgot, for a couple of very well-paid decades, that the end was never the code.And so, as house rules require, an aphorism for the road:The guild never lost to the machine. It lost to the invoice — and the machine merely sped up the paperwork.…sit -> chill -> tea -> watch -> repeat…Krzyś has spent twenty years wrestling unruly code into submission and is currently a site reliability engineer in practice and a full-stack developer according to HR, a discrepancy now under formal review. He remains a devoted follower of the church of Go, writes Rust with love, TypeScript with resignation, Python quietly, and PHP only when karmic debt comes due. His therapist, a committed behaviourist, has suggested more than once that writing thirty-minute essays about the end of craftsmanship may itself be a form of craftsmanship. He has not yet found a way to automate that observation, and suspects he is not supposed to.PS: For the record, I’m not against re-enactment. On a free weekend I shoot a re-enactment longbow, and very happily — I simply have no plans to earn a living hunting the neighbours’ cows with it. I can also bake early-medieval Slavic podpłomyki, flatbreads cooked on a hot hearthstone, which go down remarkably well at parties and have yet to attract a single job offer.And lately, in the evenings, I’ve finally started poking at bare C — something I never had time for in twenty years of being paid to write other things. The language and I are, in a sense, contemporaries: Kernighan and Ritchie’s holy book came out in 1978, the same year I did, after a gestation that began at Bell Labs around 1972. Six years in the womb — longer than an elephant, which would explain a great deal about both of us.It’s an enormous amount of intellectual fun: manual memory, honest pointers, a language that trusts you completely and is wrong to. I see no career in it whatsoever. It sits on the same shelf as the longbow and the flatbreads, under a label I’ve come to rather like: specific hobbies.SourcesFord line, 93 minutes vs 12½ hours, “relatively crude” first line, 380% turnover by late 1913: The Henry Ford — Assembly Line research guide; The Henry Ford — Ford Methods and the Ford ShopsMagneto line, 29 steps, 20 → 5 minutes: Plant Engineering, citing Ford Motor Co.US software development postings index, 76.62 on 11 Sep 2026: FRED, series IHLIDXUSTPSOFTDEVERebound of almost 15%, overall postings down 7%, 27.5% below pre-pandemic: Indeed Hiring Lab71% of the May 2025–May 2026 rise in senior roles: CIO, citing Indeed Hiring LabIndex peak of about 234 in Feb 2022: secondary source only, youngju.dev — confirm on FREDBainbridge quote and “more rather than less skilled”: Ironies of Automation, Automatica 19(6), 1983 — PDFDanish flint daggers from ~2350 BC, technical complexity: Cranbrook Institute of Science — PDFHindsgavl dagger, end of the Dagger Period, 2400–1800 BC (National Museum of Denmark, en.natmus.dk; DigVentures gives ~1700 BC): DigVenturesImitation thesis contested; microwear suggests non-practical use: National Museum of Ireland; Leiden thesis on dagger skeuomorphismAndreessen, “Why Software Is Eating the World”, 20 Aug 2011: a16z repost; Quote InvestigatorFive Dollar Day, 5 Jan 1914, profit-sharing with conditions: The Henry Ford; previous rate of $2.34 for nine hours: The Henry Ford collectionsShewhart’s control chart memo, 16 May 1924: WikipediaLordstown, 100 cars an hour, 1972 strike: Reuther Library, Wayne StateGuild abolition in France, March and June 1791: Allarde DecreePrussian freedom of trade, edict of 2 Nov 1810 and law of 7 Sep 1811: German History in DocumentsYourdon, 1992 and 1996: Rise and Resurrection of the American ProgrammerThe Last One, 1981: Wikipedia; 1981 advert textNCEES software engineering exam, 2013–2019, 81 candidates: NCEES; FBPEHermès single-artisan workshops: CBS report via AOL — secondaryOrigin of “bespoke” and the 2008 ASA ruling: World Wide Words; Savile Row tailoringPolish notaries, maximum fees by regulation: Prawo o notariacie, consolidated text 2024K&R first edition, 22 Feb 1978: WikipediaGibson quote, earliest citations 1993, not in his writings: Quote InvestigatorThis story is published under the Generative AI publication. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories. Let’s shape the future of AI together!The Coding Guild Is Over. The Factory Is Hiring (About Ten of Us). was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.