the developer discovers they were a worker

labor day finds the software developer discovering that, in the end, they were a worker all along.

for two decades, the developer lived under a statute of exception. not quite capital, not quite labor. equity stakes, signing bonuses, 30% raises for switching companies, free food, four-day weeks, remote contracts at manhattan rates from cerro chato. a tacit agreement between the industry and whoever inhabited it: there is scarcity, your conditions are exceptional, the market needs you. that agreement is ending, and this may 1st is probably the first one on which it pays to look at the industry with economic honesty.

the cycle is not just a post-pandemic correction

the tech layoff numbers are well known: 165,000 in 2022, 263,000 in 2023, 153,000 in 2024. but the interpretation that accompanies them tends to be comfortable: overhiring during the pandemic, necessary adjustment, the market returning to normal. the first quarter of 2026 registered 52,050 announced cuts in the tech sector according to challenger, gray & christmas, 40% more than the same period of 2025 and the worst start to a year since 2023. that is not normalization. it is acceleration.

what changes in this cycle is the corporate narrative. in october 2025, amazon announced 14,000 corporate job cuts. beth galetti, its vp of people, explicitly cited "this generation of artificial intelligence" as a factor reorganizing the company's structure. andy jassy had written in an internal memo in june of that year that, as amazon deployed ai and agents, it would need fewer people doing some of the jobs that exist today. salesforce, for its part, cut its customer support team from 9,000 to 5,000 people. marc benioff confirmed it in september 2025 without euphemisms: "i need fewer people." atlassian eliminated 1,600 positions in march 2026, 10% of its workforce, with more than 900 cuts in research and development. its ceo mike cannon-brookes was direct: "it would be dishonest to pretend that ai doesn't change the mix of skills we need or the number of roles required in certain areas."

the most revealing data point comes from the same challenger firm: since 2023, companies have explicitly attributed almost 100,000 eliminated positions in the united states to artificial intelligence. in march 2026 alone, ai was the top cited cause of layoffs in the tech sector.

microsoft broke a 51-year precedent by launching its first voluntary retirement program in april 2026, with approximately 8,750 eligible employees. to put it in perspective: microsoft did not use this instrument during the browser wars, nor when it lost the smartphone market, nor during any of its previous major restructurings. the company reported revenue of 81.3 billion dollars with 17% year-over-year growth in the immediately preceding quarter. it is not a program of financial distress. it is a deliberate reallocation of human capital.

there is an argument that remains pending: the drop in job postings for developers began before the mass deployment of artificial intelligence. according to indeed hiring lab, software developer openings fell more than 36% from the 2022 peak to mid-2025, and almost half of that decline occurred before the public launch of chatgpt. this suggests that the first move was post-pandemic cost discipline, not technological substitution. the second move, the current one, happens when the macro would have justified a recovery that never arrived, and the corporate narrative shifted to "reallocation toward ai". both forces are real. separating them matters for understanding what comes next.

the broken ladder

up to this point, the analysis is cyclical. what makes it structural is the second data point, the one that matters beyond the cycle: the apprenticeship model collapsed.

in august 2025, erik brynjolfsson, bharat chandar and ruyu chen of the stanford digital economy lab published "canaries in the coal mine? six facts about the recent employment effects of artificial intelligence", using adp payroll records covering more than 25 million workers. the central finding: employment of software developers aged 22 to 25 fell almost 20% from the peak of late 2022. over the same period, employment of developers over 35 grew between 6% and 12%. at companies that adopted artificial intelligence, hiring of inexperienced profiles in exposed occupations fell 13%.

a parallel paper by hosseini and lichtinger, published on ssrn in august 2025 with data on 62 million workers across 285,000 us firms, reaches convergent conclusions: after the adoption of generative ai, employment of inexperienced profiles falls between 7% and 12% at adopting firms relative to non-adopters, while employment of senior profiles remains stable or grows. the mechanism is not mass layoffs but a slowdown in hiring. the entry door closes before anyone has to leave.

the graduate numbers confirm the squeeze. according to the new york federal reserve, unemployment among recent computer science graduates reached 6.1% and computer engineering 7.5%, in both cases above the average for college graduates and for occupations historically considered less secure. the irony is not minor.

on the demand side, 54% of engineering leaders surveyed by leaddev in 2025 plan to hire fewer inexperienced profiles in the long term due to the adoption of ai development tools, and 38% report that these tools have reduced the direct mentoring that experienced developers give to those just entering. the jetbrains state of developer ecosystem 2025, based on 24,534 developers in 194 countries, found that 85% use ai tools regularly, that 41% of the code written in 2025 was ai-generated according to the respondents themselves, and that 61% of inexperienced developers consider the job market difficult, versus 34% of those with more seniority.

the economic mechanism is clear: for decades, the industry's model rested on a transfer cycle. companies hired inexperienced profiles, veterans taught them, two years later the developer was productive and the training cost was amortized into product. that cycle was what kept the statute of exception open. when a senior developer with artificial intelligence tools can do the work of several 2020 developers, the financial logic of hiring the inexperienced profile disappears. marc benioff anticipated it in the december 2024 earnings call: salesforce observed engineering productivity gains of more than 30% thanks to ai agents, and announced that in 2025 it would not hire new software engineers.

the immediate consequence is efficiency. the deferred consequence is that in five years there are no new senior developers, and in ten there are no architects. if companies do not hire inexperienced profiles for three to five years, in 2031 there will be a hole in the talent pipeline that no artificial intelligence tool can fill, because the tacit competencies (understanding business logic, making architecture decisions under uncertainty, managing technical debt) do not transfer from model to model.

the pattern is old. the speed is new.

this is not the first trade to lose its ladder before losing its practitioners. typesetters built over decades a five-year apprenticeship system inside the international typographical union. when computerized typesetting arrived in the seventies and eighties, the first thing to disappear was not the veteran typesetters but the entry into the trade: the union lost half its members between 1984 and 1987. morse telegraph operators followed a similar path. the first commercial teleprinter of 1908 was explicitly designed to eliminate the need for operators trained in morse code, not to displace existing operators immediately. technical draftsmen with 6,000-hour apprenticeship programs watched computer-aided design in the eighties and nineties reshape the trade, closing the drafting-board entry before anyone fired the veterans en masse.

the empirical pattern is consistent: automation first compresses the codifiable tasks in which apprentices build expertise. the tasks of judgment, coordination and architecture, the ones the most experienced master, resist longer. the trade eventually reconfigures itself, but the interval between the closing of the ladder and its reopening in a new form can last a decade. daron acemoglu and pascual restrepo documented in econometrica (2022) that between 50% and 70% of the changes in the us wage structure over the last four decades are explained by the relative deterioration of groups specialized in routine tasks in industries with rapid automation.

what distinguishes the current cycle is the speed of diffusion. jetbrains documents 85% adoption of ai tools among developers in less than 24 months, a curve that in previous transitions took decades. the margin for institutional adaptation is structurally narrower.

the gains are real. the problem is who captures them.

artificial intelligence is producing measurable economic gains, and that is relevant for understanding why the ladder is closing: companies are not choosing to give up productivity. they are choosing to capture it in another way.

the pwc global ai jobs barometer 2025, based on the analysis of almost 1 billion job postings across six continents, found that the industries most exposed to ai went from productivity growth of 7% (2018-2022) to 27% (2018-2024), with revenue per employee three times higher than that of the least exposed industries. jobs requiring artificial intelligence skills offer a 56% wage premium over equivalent roles without those skills. the developer who knows how to orchestrate ai systems, evaluate results and design architectures is better positioned than two years ago.

but total job postings fell 11.3%, while those requiring ai skills grew 7.5%. and the stanford ai index 2025 documents that the inference cost for a model equivalent to gpt-3.5 fell more than 280 times between november 2022 and october 2024. when the economic unit of routine cognitive work tends to zero, the surplus is not distributed: it concentrates in whoever already has the human capital to orchestrate the system. that is exactly what closes the ladder from below. it is not that business is going badly. it is that business is going well without needing the bottom of the pyramid.

three questions without a corporate answer

the developer is in that interval, between the closing of one ladder and the opening of the next. the interesting questions are not whether ai "replaces programmers", that is a headline question. they are these:

who absorbs the training cost that companies are externalizing? if companies capture today's productivity and stop investing in the training of whoever will be senior in 2031, that cost does not disappear. it is paid by universities, by accelerated training programs, by individuals who self-finance, or eventually by the same companies when the talent pipeline runs dry.

what does a software worker do when the statute of exception closes? the historical tools to answer that question are not in tech career manuals but in twentieth-century economics: portable capital, intellectual property of one's own, cooperation networks between individuals, negotiation of conditions without assuming that the company returns the loyalty it receives.

what is lost if the interval lasts too long? a developer without a ladder is not a new proletarian. it is something more fragmented: a permanent contractor with their own capital costs and individual bargaining power. that already has a name in other sectors. historically, it works badly.

labor day, in software, is not for celebrating identity or longing for privileges. it is for looking coldly at the evidence that the sector is being recategorized, and asking, without nostalgia and without alarmism, what institutional arrangement comes next. the answers are not going to come from the companies that are writing the problem into their earnings presentations as "strategic reallocation toward ai."

sources: stanford digital economy lab (brynjolfsson, chandar & chen, aug/nov 2025) · hosseini maasoum & lichtinger, ssrn 5425555 (aug 2025) · challenger, gray & christmas (monthly reports 2025-2026) · indeed hiring lab (jul 2025, nov 2025) · pwc global ai jobs barometer 2025 · jetbrains state of developer ecosystem 2025 · leaddev engineering leadership report 2025 · peng et al., arxiv:2302.06590 · becker et al. (metr), arxiv:2507.09089 · stanford hai ai index 2025 · acemoglu & restrepo, econometrica (2022) · new york federal reserve, labor market for recent college graduates (2025-2026)