Engineer proposes 'manual gates' to save junior work from AI automation

Richard Mitchell, a systems engineer with roughly 38 years in safety-critical controls (jet-engine control verification, power-plant optimization, digital nuclear control, aircraft simulation) and founder and CEO of AuraSpark Technologies, argues that AI is closing off the entry-level work through which engineers historically became experts, and proposes a deliberate countermeasure borrowed from aviation and nuclear engineering: the 'manual gate.'

Mitchell opens with his own experience leading controls design for a first-of-its-kind fully digital control system for a U.S. nuclear plant. The project was later shelved for political and economic reasons unrelated to the engineering, but before that happened, the team deliberately left manual steps inside sequences the system could already run on its own, because an operator who only ever supervises automation slowly stops being able to operate the plant, and the day automation hands control back is always the worst day if the human in the chair hasn't practiced in years.

He backs the broader trend with two studies. A Harvard University working paper covering roughly 65 million workers at 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 companies that had not adopted it, while senior employment kept growing. A Stanford analysis of ADP payroll records found the youngest workers in the most AI-exposed occupations lost ground after late 2022 while more experienced colleagues held theirs, with the losses concentrated specifically where AI automates the work; where it only augments the work, junior employment holds steady or rises. A competing explanation comes from researchers at the New York Fed, who attribute much of the rise in young-graduate unemployment not to AI but to remote work, since firms are reluctant to hire inexperienced people they cannot train and mentor at a distance. Mitchell treats the two explanations as pointing at the same underlying failure either way: the apprenticeship channel that passes expertise from senior engineers to junior ones is breaking down, which is why 'entry-level' postings have quietly come to require experience.

He draws the aviation parallel from his own early career verifying and validating software in a digital jet-engine controller. He cites Air France flight 447, which crashed into the Atlantic in 2009 after iced-over airspeed sensors fed the autopilot bad data; the autopilot disconnected and handed control back to the crew, and the pilots, having watched automation fly for thousands of hours, could not recognize a high-altitude aerodynamic stall and hand-fly out of it. Mitchell calls it a competence failure, not a hardware failure. Aviation's response, he notes, was not to remove the autopilot: in 2017 the FAA issued Safety Alert for Operators 17007, 'Manual Flight Operations Proficiency,' declaring that manual flight is the foundation other technical flying skills are built on, and some airlines changed procedures to have pilots hand-fly the initial climb and descent in benign conditions, trading some fuel efficiency to keep raw flying skills alive.

From these two cases Mitchell generalizes a design pattern he calls the manual gate: a point in a workflow where a human takes the controls not because it is the fastest way to finish the task, but specifically to exercise a skill that would otherwise decay. As a hypothetical illustration, he describes a software team placing a manual gate around debugging: when a defect surfaces in a critical module, the assigned engineer, deliberately often a junior one, must reproduce the failure, trace it to root cause, and write an automated test that captures the bug with the AI assistant switched off; only once the engineer has committed to a diagnosis does the model come back on to propose a fix and scan the codebase for similar bugs, and the engineer compares their own diagnosis against the model's.

Mitchell is direct that this costs something now to protect something later, and is a hard sell to a market that judges leaders on quarterly results: a hired executive who keeps 'unnecessary' humans AI could replace will hear about it from the board before the payoff arrives. He argues the approach only fits organizations insulated from that pressure: a founder with control, a private company, an institution with a genuinely long horizon, or a setting where a regulator requires workers to demonstrate skills regularly, as pilots must.

Key facts

A Harvard University working paper covering about 65 million workers at more than 280,000 U.S. firms found junior employment fell roughly 9 percent within six quarters after companies adopted generative AI, relative to nonadopters, while senior employment kept growing.
A Stanford analysis of ADP payroll records found junior-employment losses concentrate specifically where AI automates work; where it only augments work, junior employment holds steady or rises.
Researchers at the New York Fed offer a competing explanation, attributing much of the rise in young-graduate unemployment to remote work rather than AI, since firms are reluctant to hire people they cannot mentor at a distance.
Citing Air France flight 447 (2009) and the FAA's 2017 Safety Alert for Operators 17007 on manual-flight proficiency, author Richard Mitchell proposes 'manual gates': workflow points, such as a debugging step with AI switched off, deliberately kept inefficient to preserve human expertise.
Mitchell argues manual gates are realistic mainly for founder-controlled companies, private firms, long-horizon institutions, or settings where a regulator requires periodic skill demonstration, since they cost efficiency that quarterly-results-driven executives struggle to justify.

Why it matters

Mitchell's argument reframes AI-driven productivity gains as a hidden tradeoff: automating junior-level engineering work does not just cut headcount, it removes the failed builds and dead-end debugging sessions through which engineers historically developed judgment. His concern is a workforce that can supervise a model on paper but has never developed the instinct to catch it when it is confidently wrong, a gap that only shows up on what he calls 'the bad day,' when the automation hands control back and there is no one left who has recently practiced the underlying skill.

Who it affects

The piece is aimed at engineering organizations, especially software teams, that lean heavily on AI assistants for day-to-day work, and at the junior and early-career engineers whose formative tasks are being automated away. Mitchell frames the same dynamic as visible in safety-critical fields generally, citing his own background across nuclear-plant controls and jet-engine control-system verification, and treats aviation's decades of experience with automation-eroded pilot skill as a template for any field where AI now performs the entry-level work.

How to use it

Mitchell's concrete proposal is the 'manual gate': choose a specific skill an organization cannot afford to lose, then engineer deliberate friction to keep it alive. His illustrative example is a debugging gate: when a defect appears in a critical module, an engineer, deliberately often a junior one, reproduces the failure, traces it to root cause, and writes a test with the AI assistant switched off before the model is allowed back on to propose a fix; comparing the engineer's independent diagnosis against the model's afterward is meant to surface disagreements before they matter on a real incident.

How solid is it

The employment findings rest on two studies Mitchell cites but does not name in detail: a Harvard University working paper covering roughly 65 million workers and 280,000 firms, and a Stanford analysis of ADP payroll data, neither given authors or a citation in the piece. Mitchell explicitly flags that the causal story is contested, noting the New York Fed's competing remote-work explanation. The debugging 'manual gate' itself is presented only as a hypothetical illustration, introduced with 'picture how this might work,' not a workflow the source describes any company as having actually deployed, and no outcome data is given for the FAA's 2017 alert itself.

Risks and caveats

This is an opinion essay by an author who runs his own engineering-technology company, AuraSpark Technologies, not an empirical study. Mitchell himself concedes manual gates are less efficient in the near term than full automation and that the cost is real, not hypothetical; he argues the approach only works for organizations structurally insulated from quarterly-performance pressure, meaning it may be inapplicable to most public companies without an external push such as regulation.

“Manual flight is the foundation upon which other technical flying skills are built.”

— FAA Safety Alert for Operators 17007, "Manual Flight Operations Proficiency" (2017)