Manual Gates: How to Preserve Engineering Skills in the AI Era

The apprenticeship problem AI creates

A decade ago, a nuclear engineer designed a fully automated digital control system for a U.S. power plant—then deliberately inserted manual steps the system could execute on its own. The reasoning was straightforward: operators who only supervise automation lose the muscle memory and mental models they need when the system inevitably fails. The plant was never built, but the design principle has become urgent as generative AI reshapes how companies build technical expertise.

Recent research shows the scale of disruption. A Harvard working paper tracking 65 million workers at more than 280,000 U.S. firms found that junior employment fell roughly 9 percent within six quarters after companies adopted generative AI, while senior roles continued growing. Stanford analysis of ADP payroll data confirms the pattern: the youngest workers in AI-exposed occupations lost ground after late 2022 while experienced colleagues held steady. The losses concentrate where AI automates work rather than merely augmenting it.

The causal mechanism matters less than the outcome. Whether AI absorbs entry-level tasks or remote work severs in-person mentorship, both break the same channel through which expertise transfers from senior to junior engineers. The result is a market where "entry-level" increasingly means three years of experience required.

Why it matters

You cannot become a senior engineer without first being a junior one. Expertise develops through failed builds, debugging dead ends, and moments of unexpected success that capable AI now spares newcomers. Spare them enough formative struggle and organizations produce engineers who can supervise models on paper but lack the instinct to recognize when those models are confidently, catastrophically wrong. This is not just an economic problem—it is an engineering safety problem that other high-stakes fields have already solved.

Aviation's answer to the automation paradox

The aviation industry confronted this tension decades ago. Air France flight 447, which crashed into the Atlantic in 2009, became the canonical example. When iced airspeed sensors caused the autopilot to disconnect, pilots conditioned by thousands of hours watching automation fly could not hand-fly their way out of a recoverable high-altitude stall. The airplane worked; the skills automation had eroded did not.

The industry response was instructive. Rather than removing autopilots, aviation built deliberate manual practice back in. The FAA's 2017 Safety Alert for Operators 17007 declared manual flight "the foundation upon which other technical flying skills are built" and formally recognized skill decay as a hazard. Airlines amended procedures to encourage hand-flying during climbs and descents in benign conditions, knowingly trading fuel efficiency to keep crew skills alive.

Engineering manual gates into AI workflows

This points to a design pattern for AI-augmented work: the deliberate manual gate. Unlike a safety interlock, a manual gate exists specifically to exercise and preserve skills that would otherwise atrophy. Organizations decide which competencies they cannot afford to lose, then engineer the friction required to keep them warm.

On a software team, this might mean requiring engineers—often junior ones—to reproduce failures, trace root causes, and write automated tests with AI assistance switched off before the model proposes fixes. When human diagnosis and AI output disagree, that is the design working, surfacing the gap before the crisis instead of during it.

This reframes entry-level work entirely. Some of what looks like overhead to eliminate is actually the training apparatus for future senior staff. Protecting it may not optimize this quarter's efficiency, but dismantling it quietly mortgages organizational capability a decade out.

The cost of deliberate inefficiency

Manual gates cost something now to protect something later—a hard sell in markets judging leaders on quarterly results. The math works for founders with control, private companies, institutions with long horizons, or regulators willing to require demonstrated proficiency. Organizations unable to spend short-term margin on long-term capability will need external pressure.

Deliberate inefficiency is not waste. In safety-critical engineering it has always been insurance, purchased on purpose. As AI takes over the work where expertise is forged, the smart move is keeping hands on the controls by design—so when automation fails, someone in the chair still knows how to fly.

These details were first reported by IEEE Spectrum in an essay by the nuclear engineer who designed that never-built control system.