In 1986, I argued that by 2050 programming would become the new blue-collar work. The forecast was about more than code. It was about the direction of business, government, the economy, and politics: how competition and institutional power would shape the relationship between people and the technologies they built.
That transition appears to be arriving well ahead of schedule. AI is not the sole cause. It is the latest—and unusually powerful—stage in a longer process of moving work from people into increasingly abstracted and automated systems.
I have watched that process from inside technology for more than 35 years: from JCL and database procedures to scripting, higher-level languages, and now systems that can generate code from natural-language instructions. The pattern is not that every new tool immediately eliminates an occupation. It is that successive layers of work become automatable, production is reorganized around those capabilities, and the human contribution—and the labor market built around it—changes.
The question is not whether a machine can produce code. The question is what happens to the people whose work organizations can now produce with fewer people, at lower cost, or under tighter management.
The productivity story is not the whole labor story
There is a familiar response to automation: productivity increases, costs fall, demand grows, and new jobs emerge. That can happen. It does not tell us who benefits, when they benefit, or whether displaced workers can reach the new jobs.
Research on automation distinguishes between two forces: displacement, as machines take over tasks previously done by workers, and reinstatement, as new tasks create fresh demand for human labor. In their analysis of U.S. employment trends, Daron Acemoglu and Pascual Restrepo found that displacement accelerated while the creation of new tasks weakened over the period they examined. Automation can raise productivity and still reduce labor demand or labor’s share of value added.
The timing matters. Productivity gains may accrue to firms and owners before they generate new work elsewhere. Meanwhile, a worker who loses a position or sees the value of their expertise fall faces an immediate consequence—not a future aggregate benefit. A 2025 survey of firms in the New York–Northern New Jersey region found that AI-related layoffs were still uncommon, but some businesses had already reduced hiring, particularly for jobs requiring a college degree. The distinction between layoffs and fewer opportunities to enter a profession is important: a field can narrow its entry point before headline employment totals show a decline.
That is why projections of employment growth do not settle the question. U.S. projections published in 2025 anticipated growth in software-developer employment through 2033, while also identifying software development as an occupation potentially affected by AI. Those projections are useful snapshots, but they cannot by themselves resolve how job quality, hiring access, wages, autonomy, and the distribution of work may change as adoption deepens.
A profession can be diminished without disappearing
“New blue collar” does not mean that programming will vanish. It means the work can be reorganized into a more standardized, measured, and cost-pressured production system—one in which fewer people exercise the full range of professional judgment, while more output is generated through automated processes.
That is a labor argument, not a claim that programming and traditional factory work are identical. The history behind “blue collar” is relevant precisely because it concerns how work is structured, valued, and controlled—not simply whether it requires skill. A technically sophisticated occupation can still lose discretion, bargaining power, and the connection between individual expertise and the value produced.
The evidence on AI-assisted coding already shows why output gains and labor outcomes must be considered separately. In one controlled experiment, developers using an AI coding assistant completed a defined programming task faster. A separate set of field experiments involving 4,867 developers found an increase in completed tasks when AI tools were available. Such findings establish that AI can increase output in particular settings; they do not establish that the productivity gain will be shared with workers, that staffing will remain unchanged, or that the profession will preserve its current structure.
AI can also change the work before it eliminates a job title. A 2025 International Labour Organization analysis estimated that one in four workers globally are in occupations with some exposure to generative AI, while identifying job transformation—not wholesale redundancy—as the most likely immediate effect for most occupations. Transformation is not a neutral promise. It can mean better tools and more meaningful work; it can also mean fewer people doing more work, narrower roles, lower entry-level demand, or diminished control over how work is performed.
The longer pattern matters
This argument does not begin with generative AI. It begins with automation’s successive removal of human effort from the production process. JCL, database procedures, and scripting were signs of that direction: tasks that once required direct human instruction became encoded, repeatable, and increasingly abstracted. AI extends the pattern by moving further into the interpretation, generation, and adaptation of instructions themselves.
The economic question is whether new work will be created quickly enough, in sufficient volume and quality, to offset displacement—and whether workers will have access to it. The broader automation literature offers no guarantee that the gains will be evenly distributed. Research on productivity across advanced economies has found that productivity growth can coincide with a falling labor share, with the wage response weakening over time.
Nor does a future increase in total demand answer the near-term question. An industry can ultimately expand while individual workers, occupational pathways, and wage structures are being undermined during the transition. The aggregate economy may adjust; that does not mean every profession remains a viable path to satisfying, well-compensated work.
Where to build a career in technology
For people who love technology, my advice remains: do not assume that programming itself is a reliable route to long-term professional satisfaction. Look instead toward the work that connects technology to people, institutions, and the material world around it.
Design and usability matter because technology must fit human needs and real-world contexts. Manufacturing and supply-chain work matter because digital systems depend on physical production, materials, logistics, and maintenance. Training and education matter because the world remains profoundly unskilled in the use of technology, and new systems create an ongoing need for people who can help others understand and apply them.
These are not claims that such fields are immune to automation. They are comparatively durable places to work through the next 15 to 30 years while business, government, and society work out how they intend to reorganize around AI. The ILO’s global analysis points to the need for training and social dialogue as the transition unfolds; whether those measures are provided at the scale required is a matter of institutional choice, not technological destiny.
The question is who the transition serves
AI is not necessarily the end of the world. It may, however, mark a turning point in the history of human labor: a period when the machine moves into the production role and people are pushed further away from the machinery—or retained mainly to supervise, guide, and absorb responsibility for automated output.
The outcome will be shaped by business incentives, public policy, governance, and power. It does not require every executive to consciously desire an anti-human future. Systems can produce anti-human results through ordinary competitive pressures and institutional choices: reduce costs, increase output, capture the gains, and leave workers to adapt.
Programming becoming the new blue collar is therefore not a claim that skilled technology work is about to disappear overnight. It is a warning about the direction of travel: technical labor becoming more automatable, more standardized, and more exposed to the same pressures that have long shaped other forms of labor. The earlier we recognize that trajectory, the more seriously we can decide where human judgment belongs—and who should share in the value that technology creates.
Data reviewed for this article:
- https://www.aeaweb.org/articles?id=10.1257%2Fjep.33.2.3
- https://libertystreeteconomics.newyorkfed.org/2025/09/are-businesses-scaling-back-hiring-due-to-ai/
- https://www.bls.gov/opub/ted/2025/ai-impacts-in-bls-employment-projections.htm
- https://arxiv.org/abs/2302.06590
- https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4945566
- https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure
- https://www.brookings.edu/articles/is-automation-labor-displacing-productivity-growth-employment-and-the-labor-share/
- https://www.ilo.org/publications/generative-ai-and-jobs-2025-update
