How AI Is Disrupting the Engineer Growth Ladder from Junior to Senior

Alasdair Allen, former astrophysicist and current CTO of Negroni Venture Studios and interim CTO of medical‑tech startup Evaro, has a background in open‑source hardware, machine learning, and embedded systems. He previously led documentation for Raspberry Pi, authored over a hundred papers and eight books, and helped define standards for real‑time event interoperability.

In his QCon London talk “Artificial Intelligence Swallows the Middle Tier: The Engineer’s Path”, Allen warned that AI‑driven automation removes the “learning ladder” that junior developers rely on to gain experience, pattern recognition, and system intuition. While AI can generate code, it lacks the craftsmanship needed to understand legacy systems, production patterns, and complex business rules.

“If AI usage blocks junior engineers from developing skills, the productivity gains may come at the cost of the very expertise needed to validate AI‑generated code.”

Allen highlighted that many organizations prioritize short‑term AI productivity, neglecting the long‑term cultivation of senior engineers who can supervise these tools. This creates a talent pipeline collapse: junior hires are decreasing, and senior engineers lose opportunities to practice the deep judgment required for large‑scale, high‑load systems.

He cited two studies: the METR randomized controlled trial showed experienced developers’ speed dropped 19% when using AI, despite their belief of a 20% speed increase; Anthropic’s research found junior engineers’ mastery fell 17% with AI assistance, without any speed improvement.

AI excels at short, well‑defined tasks, but its advantage sharply declines as complexity rises. In open‑source projects, none of fifteen AI‑generated pull requests were merged, illustrating that AI can produce functional code but fails to meet the rigorous quality standards of seasoned reviewers.

“Junior engineers no longer spend years reading legacy code or debugging at 3 am; they simply point an AI agent at the code and ask for a summary.”

Allen explained that AI agents can read code and test documentation but cannot access production environments, observe long‑term request patterns, or understand which code paths carry real load—knowledge that resides only in experienced humans.

Hiring data shows a slowdown in entry‑level positions, especially in high‑risk industries, while opportunities for engineers over 25 remain stable. The result is a market where “people are not being fired; they are simply not being hired.”

Drawing an analogy to medical residency, Allen argued that the mundane tasks performed by junior engineers are essential for learning, not efficiency. Companies that ignore this will, in ten years, retain only senior developers while the rest retire.

For immediate action, Allen suggests establishing structured learning paths and rotating foundational knowledge, measuring understanding rather than speed, and treating documentation as onboarding material for a senior engineer unfamiliar with the codebase. He emphasizes that AI is a tool, not a teacher; skipping the comprehension phase with AI will “mortgage your future.”

“We are still in the diagnostic stage, not the solution stage. The focus should be on cultivating judgment, not just output.”