AI does not just change how engineering work gets done — it changes how many engineers you need to do it. That is the core argument in a March 2026 essay by Juan Cruz Martinez, a software engineer and developer relations manager writing in his newsletter The Long Commit. Martinez, who is turning 40 and facing roughly 30 more working years, draws a sharp line between past tech transitions — cloud migration, microservices, DevOps — and this one. Previous waves changed the shape of engineering work. AI changes its leverage. When one engineer with AI tools can do the work of three, organizations do not need differently-skilled engineers; they need fewer of them.
<a href="/news/2026-03-14-xai-turmoil-musk-fires-cofounders-tesla-spacex-fixers-coding-product">Claude Code is his exhibit A</a>. Martinez argues it has crossed a qualitative threshold, producing clean, well-structured code that rivals senior engineer output — and that recognition pushed him to ask what engineers are actually being paid for <a href="/news/2026-03-14-codespeak-wants-to-replace-code-with-markdown-specs">when the code itself is no longer the hard part</a>. His answer is judgment: the contextual wisdom to know which thing to build, why a technically correct solution might be wrong for a specific team or codebase, and how to evaluate AI output critically. He advocates doubling down on system architecture, domain expertise, and business context — skills built through years of watching real systems succeed and fail in production — and urges engineers to build professional equity outside any single employer through writing, teaching, and income diversification. He is also careful to separate AI's actual current capabilities from executive narratives driven by hype, warning that real careers are being eliminated not because AI replaced those roles, but because leadership prematurely bought the narrative that it would.
Hacker News pushed back. A top commenter flagged the irony of AI-adjacent professionals warning against a hype cycle they help drive, raising conflict-of-interest questions about who shapes this discourse. Others questioned whether even the architectural judgment Martinez identifies as the human moat will hold — pointing to emerging full-system platforms like Replit as potential eroding forces. One commenter reached for the COBOL analogy: pre-AI engineers as an aging priesthood of legacy-system knowledge, with no apprentices learning the craft beneath them.
That image points to the piece's sharpest problem, one Martinez surfaces without fully resolving. Boilerplate coding, documentation drafting, test generation — work that has moved from junior engineers to AI almost overnight — is precisely the low-stakes, high-volume terrain through which engineers have historically built the production intuition he now calls irreplaceable. The cohort that should be accumulating hard-won judgment through hands-on work right now is instead being squeezed out of the market. A company that eliminates entry-level positions in 2025 may not feel the consequences until its current senior engineers begin aging out in the 2035 to 2045 window — a generational timescale that renders the knowledge-transfer crisis nearly invisible in quarterly planning cycles.