The West Forgot How to Build Things. Now It's Forgetting How to Code.

A line keeps showing up in developer forums lately: “We lost manufacturing the exact way we’re now losing coding.” It sounds like a quip until you sit with it. Since the 1980s, the West sold itself a tidy story — we design, they build — and shipped its factories to East Asia. We know how that movie ended. The sequel is now being filmed in your IDE.

The Manufacturing Déjà Vu

TSMC’s Arizona fab is the cleanest cautionary tale. The buildings went up, the equipment arrived, and then the project stalled because the United States no longer had enough engineers who knew how to actually run a leading-edge fab. TSMC had to fly technicians in from Taiwan. Three decades of “factories are dirty, let someone else do it” produced a country where the people who could do it had quietly disappeared.

Shipbuilding tells the same story. The US Navy now takes more than twice as long to build a hull as Korean or Japanese yards. The blueprints exist. The welders don’t — or more precisely, the apprenticeship pipelines, the supplier networks, the muscle memory of an industry, all of it is gone. Rebuilding a broken skill chain takes far longer than breaking it did.

The Same Pattern, Now in Code

Cut to software. Junior developer hiring has collapsed in the past two years, and a lot of companies are openly making the argument: why train a new grad when a senior engineer plus Copilot can ship the same work? Seniors lean on AI for leverage. Juniors get squeezed out of the entry rung. On a quarterly earnings call, this looks like efficiency.

The bill arrives in ten years. When today’s seniors retire, there is no bench behind them. Senior engineers don’t spawn at level 5. They’re forged through years of debugging at 2 a.m., getting torn apart in code review, and slowly building the intuition for why a system breaks the way it breaks. If AI handles the grind, humans skip the apprenticeship. The code still runs. The understanding of why it runs that way doesn’t.

Systems That Work Without Anyone Understanding Them

Parts of the industry are already worried about what some are calling dependence without comprehension. AI ships code; fewer engineers can pop the hood when it misbehaves. Most days that’s fine — most drivers don’t know what’s under the hood of their car either. The trouble starts the day something breaks.

The stakes are sharpest in security and infrastructure. Zero-days, kernel panics, the subtle race conditions that surface only at scale — an LLM can produce a confident-sounding explanation, but it can’t own the fix. Someone has to go all the way down. That bench is thinning.

East Asia Is Playing a Different Game

The contrast with Korea, China, and Japan is striking. They never let manufacturing go, and they’re not letting computer science training go either. China deploys AI coding tools aggressively and graduates hundreds of thousands of CS majors a year. Korea treats software talent as national strategy. The bet: use AI as a tool, but keep building the humans who can build and reason about the tools.

The West doesn’t have a clean answer to “why bother training engineers when the AI can do it?” Decision-making optimized for next-quarter cost and stock price has a long track record of trading away long-run capability. We’ve seen this move before.

The Takeaway

Hollowing out happens in a generation. Rebuilding takes two or three. The lesson from manufacturing is sitting right there — the question is whether software runs the same play. Using AI well and using AI as an excuse not to develop people are not the same thing. Where, exactly, does your organization plan to find its next batch of senior engineers in 2031?