Protecting Engineers' Skills in the AI Era
TL;DR
A former nuclear and avionics controls engineer argues that companies using AI should build deliberate "manual gates" into their workflows. One example is debugging with the AI assistant switched off. The gates let junior engineers gain the skills they'll need when the automation fails.
The article
- The author's nuclear plant control design left manual steps in sequences the system could run alone. He did this because an operator who only supervises automation loses the skill needed on the day it hands control back.
- A Harvard working paper on 65 million workers found that junior employment fell about 9 percent within six quarters of firms adopting generative AI. Senior employment kept growing. A Stanford study of ADP payroll data shows the same pattern where AI automates work rather than augments it.
- The New York Fed blames remote work for much of the young-graduate slump. The author says both explanations describe the same broken apprenticeship channel. Nobody becomes a senior engineer without first being a junior one.
- Aviation shows the cost. Air France 447's crew couldn't hand-fly out of a stall after the autopilot disconnected. The FAA's 2017 SAFO 17007 named skill decay as a hazard. Some airlines then added hand-flying to their procedures.
- A "manual gate" is a chosen point where a human does the work to keep a skill alive. On a software team, a junior engineer switches the AI assistant off to debug. The engineer reproduces the bug, finds the root cause and writes a failing test. Then the engineer compares that diagnosis with the assistant's.
- The gates cost money now. The author expects only founders, private firms, long-horizon institutions, or regulators requiring regular skill checks to pay for them without outside pressure.
Why it matters
Engineering managers deciding how much entry-level work to hand to AI get a concrete workflow pattern and the aviation precedent behind it. The original adds the author's firsthand control-room and jet-engine-controller background. Several commenters say the prose itself reads as machine-written.
The discussion
A small, mostly hostile thread. Most commenters spent their time on whether the article was itself written by AI, which they saw as undercutting a piece about preserving human skill. The smaller substantive part split between people who think skill loss is a real organizational risk and one long reply calling it an irrational fear.
An AI-written essay about AI risk
The top thread and several standalone comments say the article reads as generated text. One linked a Pangram detector result; others pointed at stock phrasings like "honesty requires saying so" and the punchy closing sentences. One commenter found the author's company site equally generated and called him a hypocrite for not training a junior to write it. Even those who liked the argument said the AI prose weakened it.
This appears to be a completely AI-generated article, in which AI is telling us how to protect ourselves against AI.
The second sentence is an ai generation tell. Idk why ai always tries these sort of “wow you” sentences.
Skill loss is bigger than safety-critical work
Supporters said the article's focus on aviation is too narrow. The deeper risk is that organizations lose the unwritten, social knowledge of why their systems behave as they do. That problem predates AI, and it already killed firms that outsourced core functions. Another commenter noted that high turnover makes training juniors a prisoner's dilemma, since the company paying for the training may not keep the engineer.
without a continual replacement of labour that understands the system and the reasons for the system's behaviour, the ability to extend it or fix it is lost.
especially with the culture of job hopping a lot and high turnover you can't be sure that you're not just training staff that'll go elsewhere. It's like an industry wide prisoners dilemna.
Corporate text was already mostly waste
A side argument started from the slop complaint. One commenter said offices have produced low-value documents for decades, and AI just made them cheaper. The useful ones, like process docs, manuals and troubleshooting guides, never get budget. The pushback was that human work shouldn't be lumped in with slop, and that documents one team calls useless are often required by another.
Corporate world is obsessed with producing text to unhealthy degree. AI generation is the result, not the cause.
In my experience, what is "useless" to me is required by another team.
The handoff problem in practice
Commenters connected the nuclear operator story to self-driving cars that expect an inattentive human to take over in an emergency. A cynical reply said the human is there as a scapegoat. A developer posted a parody poem about ceding CI, code writing, review and merging to AI agents. It ends with a week of unreviewed code that he has an hour to debug.
A human who is not driving is not paying attention and lacks the spatial awareness to take immediate control in a critical situation...
until everything breaks and I have 1 hour to find why the million lines of code merged this week don't work anymore.
Consensus
The prose reads as AI-generated, which most commenters found off-putting in a piece about keeping human skill. Among those who engaged with the argument, most agreed that hollowing out junior work puts long-term capability at risk.
Dissent
One commenter rejected the premise. When a skill dies out, organizations retrain people or replace the system that needed it, and aviation is too extreme a case to generalize from.