I'm not going to open this one with comfort.
Between 2023 and 2025, employment for "computer programmers" in the United States fell 27.5 percent.
Not a niche subfield. A category employing hundreds of thousands of people, contracting by more than a quarter in 24 months.
Entry-level tech hiring fell 25% year-over-year in 2024. Stanford's Digital Economy Lab found employment for software developers aged 22–25 had declined nearly 20% from its 2022 peak.
Marcus, a composite character drawn from dozens of real profiles, started coding at 14, taught himself from Stack Overflow, landed a junior role in 2022. By 2024, he was a mid-level full-stack developer earning $92,000.
In January 2026, he was laid off.
The company had adopted Claude Code six months earlier. Output up 40%. Engineers required down proportionally.
Now here's the part they're not headlining:
In the same period that "computer programmer" employment fell 27.5%, employment for "software developers", the category that includes architecture, design, system integration, and technical decision-making fell by 0.3%.
Not twenty-seven. Zero point three.
And the Bureau of Labor Statistics projects software developer roles to grow 15–18% through 2034.
Demand for AI Engineers: growing at 20.78% CAGR. Projected 11.72 million roles by 2030.
So what's actually happening?
Compression, not elimination.
The routine, specification-to-code translation work that junior roles were organised around, it's being automated. The architecture, the orchestration, the judgment work that sits above it, it's expanding.
One senior engineer with Claude Code can now do what previously required a team of three. That's not replacement. That's leverage.
But here's the critical insight from the METR 2025 controlled study:
Experienced developers using AI tools were 19% slower, not faster.
Not because the tools lack capability. Because AI productivity gains require skill in wielding AI, not just access to it.
That's the wielder's edge. And that's what the next 16 days are about.
Tomorrow: exactly what AI handles reliably in 2026, and the six categories that will always require you.