The Ladder They Pulled Up

Somewhere in San Francisco, a software engineer at one of the city's largest technology companies sits at his desk and does his job. His manager sends him a task. He opens a terminal. He types a prompt into an AI coding tool called Claude Code. The tool writes the code. He reviews it — or, increasingly, does not — and submits the result. His manager is satisfied. The feature ships. The cycle repeats.

He is, by any conventional measure, employed. He earns a good salary. He works at a company whose name would impress his parents. And when a reporter from the San Francisco Standard asked him how he felt about all of this, he chose a word that had nothing to do with productivity or efficiency or the future of work. The word was grief. "The skill you spent years developing," he said, "is now just commoditized to the general public."¹

He is not entry-level — he has years of experience, and still the ground has shifted beneath him. For the thousands of computer science graduates now entering a job market that has quietly closed its doors to them, the situation is worse. They are discovering that the first job — the one that was supposed to teach them the skill in the first place — may no longer exist.

The numbers landed in August 2025, and they landed hard.

A team of economists at Stanford University, led by Erik Brynjolfsson, analyzed payroll records from millions of American workers through ADP. Since the release of ChatGPT in late 2022, employment for software developers aged 22 to 25 had declined by nearly 20 percent. Across all occupations most exposed to generative AI, entry-level employment had fallen 13 percent — even after controlling for macroeconomic shocks and firm-level hiring patterns. For experienced workers in the same occupations, employment held steady or grew.²

The study was titled "Canaries in the Coal Mine," and the metaphor was not subtle.

Hiring of new graduates by the 15 largest technology companies has fallen by more than 50 percent since 2019, according to SignalFire, the venture capital firm.³ Computer science graduates now carry a 6.1 percent unemployment rate — higher than liberal arts majors.⁴ And yet the Bureau of Labor Statistics still projects software developer employment to grow 17 percent from 2023 to 2033, far faster than the national average.⁵ The industry is not shrinking. It is restructuring. And the restructuring has been lethal to anyone standing on the bottom rung.

To understand why this matters, you have to understand what entry-level software engineering work actually was.

It was an apprenticeship. Companies hired junior developers and assigned them the work that no one else wanted: fixing small bugs, writing boilerplate code, building modest features under the supervision of engineers who had done the same a decade earlier. The juniors were slow. They made mistakes. They asked questions that revealed how little they understood about the difference between a textbook and a production system. And in the friction between what they thought they knew and what the work actually demanded, they learned. Not syntax or algorithms — they had learned those in school. They learned judgment. When to push back on a requirement that sounds simple but introduces architectural debt. How to debug a failure that spans three services and two cloud providers.

That tacit knowledge is what separates a person who can write code from a person who can build systems. Every senior engineer working today acquired it through the apprenticeship. And the apprenticeship is being dismantled.

On a February 2025 earnings call, Salesforce CEO Marc Benioff said his company would hire no new software engineers that year, citing a 30 percent productivity boost from AI.⁶ Two months later, Microsoft CEO Satya Nadella estimated that 20 to 30 percent of Microsoft's code was now written by AI.⁷ The tasks these tools handle most capably — boilerplate, documentation, routine bugs, simple features — are precisely the tasks that once served as the junior developer curriculum. The grunt work was never just grunt work. It was the training ground. And it has been automated away.

AWS CEO Matt Garman was blunter. On a podcast in August 2025, he called the idea of replacing junior developers with AI "one of the dumbest things I've ever heard" and posed the question the industry must answer: "How's that going to work when ten years in the future you have no one that has learned anything?"⁸

If companies stop hiring juniors now, they will face a shortage of mid-level engineers by 2030 and a crisis-level scarcity of senior architects by 2035. The senior engineers reviewing AI-generated code today were the juniors who wrote terrible code ten years ago. Cut off the supply, and the pipeline stops.

Efficiency without succession planning is liquidation.

The Stanford study identified the mechanism: a distinction between two modes of AI deployment. In occupations where AI primarily automated tasks, entry-level employment collapsed. In occupations where AI primarily augmented human capabilities, employment for young workers remained stable or grew.⁹ Companies that use AI to eliminate junior roles are borrowing against a future they cannot afford. Companies that use AI to accelerate junior development are building a bench.

But that is the industry's problem. The graduate's problem is more immediate, and the honest answer to it is harder than any platitude: the traditional path — study, apply, get hired, get trained — has narrowed to the point of near-invisibility. What remains demands something no previous generation of software engineers was asked to do. It demands that they train themselves.

What this amounts to, in practice, is a generation that must construct its own apprenticeship. The most resourceful among them are already doing it — building complete applications from concept to deployment, not as tutorial exercises but as working software for real users with real problems. They design the architecture, write the code, operate the system in production, and fix what breaks. AI tools accelerate the process enormously, but the value is not in the speed. It is in the judgment that emerges from owning every decision and living with its consequences.

Others are finding in open source what the corporate apprenticeship once provided: production-scale codebases, code review from experienced engineers, and a public record of contribution that no résumé can replicate. GitLab has reported that nearly a third of its first 40 engineering hires were open source contributors before they were employees.¹⁰ The path is available. It simply requires initiative that the previous generation's path did not.

The skills that matter most now are the ones AI handles worst: system architecture, complex debugging, security as adversarial thinking, and the ability to translate between engineering, design, product, and business — the cross-functional fluency that, across twenty years and two continents, has distinguished every engineer who survived a paradigm shift from those who did not. The engineers who endure are never the fastest coders. They are the ones who understand what to build, not merely how to build it.

And the opportunities themselves are not scarce — they are merely distributed differently than ambition expects. Not every job worth taking carries a prestigious address or a six-figure starting salary. Across the country and across the Pacific, communities are struggling with digital transformations they cannot afford at prevailing rates. In Guam, a U.S. territory in the western Pacific, government systems that should give citizens transparent fiscal data still run on workflows designed for a previous era. Healthcare, education, and tourism — the pillars of the island's economy — need modernization that mainland firms price beyond reach. The island loses talented young people every year, a quiet exodus that hollows out the institutions that need them most. Guam is one example. There are hundreds of others. The question is not whether the work exists. It is whether the young engineers who need experience are willing to go where the work is, rather than where the résumé points.

The traditional entry-level position required you to show up and be trained. The new path requires you to show up having already trained yourself — through work chosen because it mattered, not because it was assigned. That is not the bargain this generation was promised. But it may produce a more resilient kind of engineer, one whose competence was forged through initiative rather than inherited through assignment.

The engineers who thrive through technological upheaval always share a posture that has nothing to do with any particular language or framework or tool: a willingness to be useful, a refusal to wait for permission, and a conviction that the purpose of knowledge is service.

The industry has pulled up the first rung of the ladder. The task now is not to wait for someone to lower it back down. It is to build your own.