Generative AI as Seniority-Biased Technological Change

Original Article ↗ Hacker News Discussion ↗

Shrinking junior pipeline & “where do seniors come from?”

  • Many commenters worry that cutting entry-level roles now will leave too few qualified seniors in 10–20 years, or force promotions of underqualified people, worsening product quality and “enshittification.”
  • Others think seniors themselves may later be cut as AI improves, so companies are implicitly betting on AGI timelines rather than on long-term human pipelines.
  • Some argue the problem is deferred: current seniors in their 30s–50s exist, but the gap will emerge once they retire.

AI capability vs hype and macroeconomy

  • Strong disagreement over whether juniors are being replaced by actual AI performance or by management’s expectations and hype.
  • Several point to high interest rates, weak demand, post‑COVID overhiring, tax changes, offshoring, and visa policy as alternative or compounding drivers of reduced junior hiring.
  • Some say AI is a convenient cover story for cuts companies wanted to make anyway.

Changing work and training models

  • AI + seniors can remove many of the “grind” tasks that used to train juniors, reducing their marginal value.
  • There’s debate over whether AI-assisted coding and “agentic coding” can truly teach deep understanding or just enable superficial “vibe coding.”
  • University instructors describe banning LLMs for foundational coursework while allowing them in open‑ended projects as a compromise.
  • Suggestions for new pipelines include internships, open source, non‑SWE roles that involve coding, and even long-term contracts.

Incentives, short-termism, and tragedy of the commons

  • Many note that firms have little private incentive to invest in juniors who may job‑hop, especially when judged on quarterly metrics.
  • This is framed as a classic tragedy of the commons: everyone relies on someone else to train future seniors, so the pipeline shrinks.
  • Some call for government intervention or subsidies; others predict more visas or offshoring instead.

Data and study skepticism

  • Several question the LinkedIn/Revelio dataset: representativeness, duplicate postings, and the very low measured AI‑adoption rate.
  • Others argue the design (AI adopters vs non‑adopters in same sectors) should at least partially control for macro trends, but confounders remain “unclear.”