Matthew Ramirez started college as a computer science major at Western Governors University in 2025. He wanted the career his generation had been promised: flexible, well-paid, future-proof. Six months later, he dropped the major and applied to nursing school.
🎙️ Listen to this article
The 20-year-old told the Guardian that he ran the math on his own timeline. By the time he graduated in 2028, the entry-level coding jobs he was training for would likely not exist in their current form. His family works in nursing. Nursing requires hands. Hands are hard to automate. "Even though AI might not be at the point where it will overtake all these entry-level jobs now," Ramirez said, "by the time I graduate, it likely will."
Ramirez is not an outlier. According to a Zety survey, 43 percent of Gen Z workers anxious about AI are moving away from corporate and administrative roles toward careers that rely on what career development expert Jasmine Escalera calls "human skills" — creativity, interpersonal connection, and hands-on expertise. Fifty-three percent of young respondents said they were seriously considering blue-collar or skilled trade work. The Wall Street Journal reported that the shift is measurable in enrollment data.
The question that used to be speculative — will AI take programming jobs? — has become operational. The answer is arriving faster than the people affected can retrain.
The profession that built Silicon Valley is being dismantled by the tools Silicon Valley built
On November 24, 2025, Anthropic released a new version of Claude Code. Engineers spent their holiday breaks testing it. They called it "Claude Christmas." The tool could autonomously build projects that would have taken weeks of human coding. Developers emerged from the holidays deeply unsettled. Some began using a phrase that had been circulating in tech circles since mid-2025: the "permanent underclass."
The SF Standard reported what followed. In San Francisco and San Mateo counties, where roughly 190,000 jobs are tied to tech, the anxiety became impossible to ignore. An essay by AI CEO Matt Shumer went viral in early February, arguing that tech workers have spent the past year watching AI surpass them at their jobs, and that other white-collar workers are about to experience the same thing. Fortune called it a comparison to February 2020 — the last moment before a catastrophe everyone could feel coming but hadn't yet arrived.
Then Microsoft's AI chief Mustafa Suleyman gave it a timeline: 18 months before anyone whose job involves looking at a computer screen is affected. Anthropic's Dario Amodei predicted half of all entry-level white-collar jobs gone within one to five years. Ford CEO Jim Farley said half of white-collar jobs, full stop. Federal Reserve Governor Michael Barr warned that millions could become "essentially unemployable" in the near future.
These predictions share a feature that separates them from earlier automation fears: they're coming from people who are building and deploying the technology making the predictions come true.
The 80/20 inversion explains the mechanism
To understand why coding is the first domino, you need one number. Daivik Goel, an engineer working on his own startup, described it to the SF Standard: engineers used to spend 20 percent of their time designing and 80 percent writing code. "But now it's rare that you write any code at all."
That ratio captures the structural shift. Software engineering, as practiced for the past three decades, bundled two distinct skills into a single job description. The first skill was judgment — understanding what to build, why, and how it fits into a larger system. The second was implementation — translating that understanding into functioning code. The two skills were inseparable because the translation step was hard. It required years of training, deep knowledge of languages and frameworks, and the kind of craft expertise that justified starting salaries above $120,000.
AI has separated them. GPT-5.3-Codex went generally available on GitHub Copilot on February 9. Microsoft says 30 percent of its code is now AI-generated. Meta CEO Mark Zuckerberg has predicted that by midyear, AI will write most of his company's code. The implementation skill — the one that anchored salaries, career trajectories, and an entire professional identity — is being commoditized at a speed that has no precedent in the history of white-collar work.
Tanmai Gopal, CEO of PromptQL, a billion-dollar AI unicorn, told Fortune that the latest AI models have the judgment and taste of an "average senior software engineer." Standard software engineering, he explained, relies heavily on converting established business context into technical code. AI excels at exactly that translation. "What used to be kind of sometimes considered the epitome of white collar was high-grade software engineering," Gopal said. "That's been all the rage for the last 30 years and I'm excited to see that go."
Coding is uniquely vulnerable because engineers built the training data
Why coding and not, say, nursing or plumbing or trial law? The SF Standard identified the mechanism: coding is entirely digital. Unlike jobs that require physical presence or navigating the unpredictability of human relationships, software can be written, tested, debugged, and improved entirely by machine. There is no physical world to interface with. The medium is the machine.
But there is a deeper irony. Engineers spent decades publishing code publicly on GitHub, Stack Overflow, and open-source repositories. They documented their patterns, explained their reasoning, and created an enormous corpus of labeled training data — for free. They built, in effect, the perfect training ground for the systems that would learn to do their jobs.
Today's coding agents don't just accelerate the work. They show signs of what the SF Standard called having "ideas" — proposing architectures, following their own roadmaps, executing sophisticated projects with minimal human input. James O'Brien, a computer science professor at UC Berkeley, described the existential dimension: "If suddenly we have a machine that's able to do all the things that society thought you were valuable for, that's very existentially upsetting."
The counterargument: engineers aren't being replaced, they're being promoted
The strongest version of the counterargument comes from Gopal himself. He told Fortune that the doomsday predictions contain "a grain of truth while also being massively overstated." Silicon Valley, he argued, is projecting its own experience onto the entire economy. "It's like, oh, this is the problem for 7 billion people on the planet, because I'm in Silicon Valley, so I obviously know what's best, right?"
Gopal's framing: humans are becoming "context gatherers instead of just workers." The skill that earns promotions and makes people effective has always been the ability to understand context — what needs to be built and why — not the mechanical ability to write the code. AI is stripping away the mechanical layer and revealing the judgment layer underneath.
Stanford economist Erik Brynjolfsson offers indirect support for this view. Writing in the Financial Times, he noted that all U.S. job gains for 2025 were revised down to just 181,000, while his calculation projected productivity growth of 2.7 percent for the year — nearly double the 1.4 percent average of the past decade. The implication: fewer workers, producing more. If that productivity accrues as higher wages and new roles for the remaining workers, the transition is painful but manageable.
The rebuttal: the data says displacement is already happening
But the data tells a more complicated story. ADP, the largest payroll company in the United States, found that professional and business services roles, alongside information services jobs in media, telecom, and IT, collectively lost 41,000 jobs in December 2025 alone. In the same month, employment grew in healthcare, education, and hospitality — the hands-on, people-facing sectors that AI cannot easily reach.
The World Economic Forum projects that AI could displace 92 million roles worldwide by 2030. Challenger, Gray & Christmas, the outplacement consultancy, reported that employers cited AI as a factor in nearly 55,000 job cuts during 2025.
The SaaSpocalypse that wiped $2 trillion from software-as-a-service valuations earlier this month makes the financial dimension concrete. Bank of America Research called AI a "double-edged sword" — not purely an upside play, but a force that could "cannibalize" the very software businesses that were supposed to benefit from it. When AI can write software, the companies that sell software are exposed. When those companies are exposed, the engineers they employ are exposed. The chain is short and the links are tight.
Gopal acknowledged the tension. Cynics, he said, have a point that doomsday predictions conveniently arrive around the time of the next multibillion-dollar funding round. There is a funding rationale behind the fear. But the fear is also grounded in observable reality: the latest models genuinely perform at the level of senior engineers for standard software tasks.
The transition will be brutal for the people living through it
History offers precedent but not comfort. Every major automation wave — mechanized farming, assembly-line manufacturing, computerized office work — eventually created more jobs than it destroyed. The "eventually" is doing a lot of work in that sentence. The transition periods lasted decades. The workers displaced by automation in the 1980s and 1990s did not personally benefit from the new economy that emerged in the 2000s and 2010s. Their children did, sometimes.
The speed of AI-driven displacement is different. Previous automation waves moved slowly enough for labor markets to adapt through retraining, geographic migration, and generational turnover. AI coding tools went from novelty to production-grade in roughly 18 months. The labor market has no mechanism that operates at that speed.
Ramirez's decision to switch from CS to nursing is rational at the individual level. But it carries a collective irony: if enough talented people abandon computing, the field's ability to guide and oversee AI systems — the judgment work that Gopal says defines the new role — will atrophy at exactly the moment it becomes most important.
What this demands
If you are an engineer, the action ladder is clear. First: assess honestly how much of your daily work involves judgment versus implementation. If the answer is mostly implementation, the timeline for disruption is measured in months, not years. Second: invest in the context-gathering skills that Gopal describes — domain expertise, system design, the ability to evaluate AI output critically rather than accept it. Third: track the data. ADP's monthly employment reports, Challenger's layoff tracking, and your own company's internal metrics on AI-generated code will tell you more about your personal risk than any CEO's prediction.
If you manage engineers, the obligation is different. The companies that handle this transition well will be the ones that retrain and redeploy rather than simply replace. The ones that handle it badly will discover, as the manufacturing sector did in the 1990s, that the talent loss compounds for years after the layoffs stop.
Ramirez is 20. He has time to build a career in a field AI can't touch. The 35-year-old senior engineer with a mortgage and two kids doesn't have the same runway. The question the industry owes that person an answer to isn't whether AI will change software engineering. It's what, specifically, they should do about it on Monday morning.
