Why We Still Learn After AI Can Write the Answers
Specific knowledge expires; judgment, verification habits, and responsibility do not. Two migrations — from aircraft maintenance to software, and into the AI era — and why learning still matters.
Moving from aircraft maintenance into software made much of my specific knowledge seem obsolete, but the experience survived. It left a bias toward reliability and a troubleshooting habit: gather symptoms, locate the cause, and verify the fix. Those habits still shape how I investigate software and production issues.
AI has brought a second shift. It writes code in unfamiliar languages and generates SQL, pages, and scripts. I may write fewer queries myself, but I still need to understand schemas, constraints, failure modes, and stability signals. I cannot outsource my understanding of what a model did or where its result might break.
I distrust AI success stories that omit prior experience, requirements, mistakes, testing, and responsibility. Real delivery is never one generation.
Accountability brings me back to verification. I must know what the work actually did, whether it met constraints, and why it counts as correct. Verification should be layered: reversible, inspectable work can rely more on automated checks; real users, production, money, and security require deeper human involvement.
More jobs now include managing agents. Management span is not the number of agents started, but the exceptions requiring judgment, the cost of verification, and coordination between tasks.
Learning has changed too. AI lowers the social cost of asking: I can start with a basic question, revise my description, and try again without worrying about taking up another person’s time. Curiosity becomes action more easily.
For an in-service graduate exam, I loaded reference materials into an agent, asked for a plan, and studied one concept at a time. The crucial step was retelling each idea through my own experience and letting the agent correct my interpretation. Feeling comfortable was not evidence of understanding; being able to restate and transfer the idea was stronger evidence.
An AI familiar with my background can translate unfamiliar concepts into examples from my work. But that does not replace systematic reading: following an author’s complete argument still reveals questions I would not have known to ask.
The same alignment principle applies to complex consultations. Before requesting a solution, I explain the context, goals, sources, standards, and constraints, then ask the AI to discuss its understanding first. Up-front alignment costs less than a long investigation aimed at the wrong problem.
My practical conclusion is simple: do not merely ask for an answer. Align on the problem, restate new knowledge in your own words, accept correction, and verify the result. Curiosity starts the process; dialogue builds understanding; reading expands its boundaries; verification and responsibility make the outcome trustworthy.