An essay published by Tech Trenches argued that the software industry risks repeating a pattern seen in Western defense manufacturing: optimizing away people and capacity that appear unnecessary, then discovering during a crisis that the missing knowledge cannot be restored quickly.
The author, who says he runs engineering teams in Ukraine, begins with weapons production. The essay describes Raytheon's effort to restart Stinger missile manufacturing after a two-decade purchasing gap, including bringing back older engineers familiar with paper-era designs and obsolete components. It also points to Europe's delayed effort to supply one million artillery shells to Ukraine, citing shortages across explosives, propellant, factories and trained workers. These examples are presented through the author's synthesis; the underlying reports are not separately supplied in the evidence packet.
A more unusual case in the essay concerns Fogbank, a classified material used in nuclear weapons. When production resumed after a long shutdown, key expertise and records were gone. The author says researchers eventually found that an unintended impurity in the old process had been functionally important. The lesson drawn is that some operational knowledge is neither fully documented nor even consciously recognized until the production system disappears.
The essay maps that problem onto software careers. Junior developers traditionally acquire judgment through years of supervised work before becoming mid-level, senior and principal engineers. If companies use AI tools to eliminate entry-level roles, the author argues, they may enjoy near-term savings while weakening the future supply of people able to review systems, diagnose unusual failures and make architectural decisions. Money cannot instantly recreate that experience once demand returns.
Several figures are offered to support the warning. The essay cites a randomized METR study in which experienced open-source developers using AI coding tools took 19% longer on selected real-world tasks, despite initially expecting to work faster. It also references a LeadDev survey in which 54% of engineering leaders expected copilots to reduce junior hiring over the long term, and a survey of university computing departments that found 62% reporting enrollment declines. Those results cannot be independently evaluated from the single supplied source and may not describe every team or type of AI-assisted work.
The comparison has limits. Software can be copied and distributed differently from physical ammunition, and AI systems may change both training and productivity in ways that historical manufacturing analogies cannot predict. The essay's value is therefore less a forecast than a risk framework.
Its central question is whether organizations are measuring only today's output. If fewer beginners receive meaningful work, mentorship and responsibility, the apparent efficiency of current tools could conceal a slower loss: the human pipeline required to understand and maintain tomorrow's systems.
