A little over a decade ago, the controls design for a first-of-its-kind full digital-control system for a United States nuclear plant was led by an engineer facing a fascinating paradox. On paper, it was a beautiful machine, engineered to run itself the way a modern airliner does, with human operators watching over a system that rarely needed their active intervention. Yet, a conscious decision was made that, to any efficiency-minded observer, looked entirely backward: the design team deliberately left manual steps inside sequences that the system could easily execute entirely on its own.
They were solving a very specific, highly critical human-factors problem. An operator who only ever supervises automation slowly stops being an operator. Their hands go cold, and their mental model of what the plant is actually doing gets fuzzy, abstract, and detached. Then comes the inevitable day when the automation hands control back. It is almost always the worst day possible, because automation only quits when it is deeply confused or in trouble. But by the time that happens, you have a person sitting in the chair who hasn’t truly operated the equipment in years. The manual steps were there to keep the human current and engaged. It was inefficient by design, done purely on purpose.
That plant, as it happened, was never ultimately built. It was shelved amid the complex politics and shifting economics that routinely surround nuclear power in the country, for reasons that had almost nothing to do with the underlying engineering. But the core design instinct outlived the project itself, evolving into a concept that speaks directly to the debate now consuming every corporate boardroom: What happens to human expertise when artificial intelligence does the work that used to build it?
AI Is Disrupting the Engineering Career Ladder
The empirical data highlighting this shift has simply gotten too large and consistent to wave away. A Harvard University working paper covering some 65 million workers across more than 280,000 U.S. firms found that after companies adopted generative AI, junior employment fell roughly 9 percent within six quarters relative to nonadopting firms, while senior employment kept right on growing. A Stanford University analysis of ADP payroll records points in the exact same direction: the youngest workers in the most AI-exposed occupations lost ground after late 2022 while their more-experienced colleagues held their ground. The Stanford researchers found that these employment losses concentrate precisely where AI automates the work; where it merely augments human effort, junior employment holds steady or even rises.
The underlying causal story is still fiercely contested, and intellectual honesty requires stating that clearly. Researchers at the Federal Reserve Bank of New York attribute much of the rise in young-graduate unemployment not directly to AI, but to remote work, arguing that firms are reluctant to hire inexperienced individuals whom they cannot adequately train and mentor at a distance. But it is worth noting what these competing explanations share. Whether a generative model is absorbing the formative entry-level work or physical distance is severing the mentorship network around it, both descriptions point to the exact same broken mechanism: the traditional apprenticeship channel through which expertise passes organically from senior staff to junior staff. Either way, "entry-level" has quietly come to mean "three years of experience required."
When you strip away the technological noise, you are left with one deceptively simple reality: You cannot become a senior engineer without first being a junior one. True expertise is not something that can be downloaded into a brain via a software update. It is earned through failed builds, frustrating dead-end debugging sessions, and the puzzling "why on earth did that work" moments that a capable AI assistant will now happily spare the newcomer. Spare them enough of those foundational struggles, and you inevitably produce a cohort that can supervise a model on paper but has never developed the internal gut sense to know when that same model is confidently and catastrophically wrong.
Most of the public commentary on this phenomenon stops at the diagnosis, or reaches for macro-level policy solutions that treat the loss of junior jobs purely as an economic problem. Yet it is fundamentally an engineering problem as well, and safety-critical fields have already spent decades learning how to solve it.
Aviation’s Lessons About the Automation Paradox
My own career started at the sharp end of industrial automation. My very first job out of school was verifying and validating the software inside the digital jet-engine controller that decides, faster than any human pilot ever could, how a fighter plane’s engine responds to shifting atmospheric conditions. Even then, in the late 1980s, the central tension was clearly visible to everyone in the room: The machine vastly outperforms the human in routine cases, but the human is all that stands between the aircraft and utter disaster in the rare cases that the machine simply didn’t anticipate. This tension is formally known as the automation paradox, in which increasingly capable automation gives human operators less practice and fewer opportunities to hone their craft, while leaving them alone with only the most difficult, high-stakes situations.
Aviation has learned, repeatedly and at great human expense, what happens to human skills when they quietly atrophy inside that operational gap. The canonical example is Air France Flight 447, which plunged into the Atlantic Ocean in 2009. The proximate cause was mundane: iced-over airspeed sensors fed the autopilot bad data, and the system did precisely what it was designed to do in response—it disconnected itself and handed control of the airplane back to the flight crew. What followed was not a hardware failure, but a profound competence failure. A recoverable situation became an unrecoverable disaster because the pilots, conditioned by thousands of hours of passively watching the automation fly the plane, could not read a high-altitude aerodynamic stall and hand-fly their way out of it. The airplane was working as intended. The human training that the automation had slowly eroded was not.
The commercial aviation industry’s response to this reality is deeply instructive, echoing the exact move made in that nuclear control room years prior. The industry did not rip the autopilots out of the aircraft. Instead, it deliberately built manual practice back into standard operating procedures. In 2017, the Federal Aviation Administration issued Safety Alert for Operators 17007, titled "Manual Flight Operations Proficiency," declaring that manual flight is the foundational bedrock upon which other technical flying skills are built. The safety alert formally recognized ongoing skill decay as an aviation hazard in its own right.
Following this guidance, several major airlines amended their internal procedures to actively encourage pilots to hand-fly both the initial climb and initial descent phases of flight during benign weather conditions, knowingly trading a tiny sliver of fuel efficiency to keep the crew’s raw flying skills alive and sharp. That specific trade is the entire point. A perfectly optimized system that routinely produces incompetent operators is not truly optimized at all. It has simply shifted its inherent failure mode somewhere that a spreadsheet cannot easily see.
Manual Gates Could Preserve Engineering Skills
Put the hard-won aviation lesson and the nuclear design instinct side by side, and they point directly to one crucial design pattern that modern workplaces now desperately need in AI-augmented environments: the deliberate and purposeful manual gate.
A manual gate is a specific point in a complex workflow where a human being takes over the controls, not because it is the fastest or most efficient way to get the task finished, and not just as a safety interlock, but specifically to exercise and preserve a human skill that would otherwise atrophy from disuse. The defining feature of this approach is that it is chosen intentionally. An organization decides, as a matter of foundational design, which core competencies its human staff must keep alive because those are the exact capabilities the team will need on the bad day when everything goes wrong. Then, the organization engineers the necessary friction into the process to keep those skills warm.
Picture how this concept might function on a modern software engineering team that relies heavily on AI to generate the bulk of its code. The team could place a strict manual gate around the skill it can least afford to lose: deep debugging. When a complex defect surfaces in a critical software module, the assigned engineer—deliberately, and often a junior member of the team—must first independently reproduce the failure, trace it back to its root cause, and write an automated test that captures the bug, all with the AI assistant switched completely off. Only after the engineer commits to their own diagnosis does the AI model come back online to propose fixes, generate alternative solutions, and sweep the broader codebase for similar underlying bugs. The engineer then compares their independent diagnosis against the model’s output. When the two disagree, that is the design working exactly as intended, surfacing the technical disagreement before the bad day arrives instead of during it.
This design framework reframes the role of the junior engineer entirely. The prevailing instinct today is to let AI handle all entry-level work simply because it is faster and cheaper in the short term. But some of that foundational work is not mere overhead to be ruthlessly eliminated. It is the primary training apparatus of your future senior staff, and organizations should protect it the way they protect any other piece of critical infrastructure. It may not look efficient this quarter, but dismantling it quietly mortgages your core technical capability a decade down the line.
Why Companies Must Keep Training Junior Engineers
None of this operational philosophy is free, and pretending otherwise would be an insult to the executives and managers who have to sign corporate budgets. A deliberate manual gate is, by its very construction, significantly less efficient in the near term than full, uninhibited automation. Keeping junior workers engaged in formative work and forcing them to run manual sequences costs real resources now in order to protect vital capabilities later.
That is a notoriously hard sell in a modern market that judges most corporate leaders strictly on quarterly financial results. A hired executive who carries "unnecessary" human workers that artificial intelligence could theoretically replace will hear about it from the board of directors long before any long-term payoff materializes. The math only truly works for entities insulated from that relentless pressure: a founder with voting control, a private company with a stable vision, an institution with a genuinely long horizon, or a regulator willing to step in and require workers to demonstrate their core skills regularly, much as commercial pilots are required to do. Which means the organizations most likely to successfully preserve their own internal expertise are those structurally able to spend short-term profit margin on long-term capability; everyone else will ultimately need an outside push from industry standards or regulation.
The argument ultimately rests on a simple premise: deliberate inefficiency is not waste. In safety-critical engineering, professionals have always known it as insurance, and they buy it on purpose. As artificial intelligence takes over the routine tasks where human expertise was historically forged, the smart strategic move is not to resist the march of automation. Rather, it is to keep our hands deliberately on the controls by design, ensuring that when the automation inevitably fails—as complex systems always eventually do—there is still someone sitting in the chair who actually knows how to fly.
