If AI can do the work itself, how does anyone learn to become an expert at it?

This explores what happens to the passing-down of expertise once AI can do much of the work that people used to learn by doing. The corpus doesn't have a study that tracks apprentices directly. Read together, though, the notes point to three shifts: expertise moves out of people and into files, the rungs where novices used to practice start to disappear, and learners lose a reliable way to tell whether they've actually learned anything.

The first shift is that expertise is being written down as machine-readable scaffolding instead of being passed from person to person. In one industrial case, domain rules and design principles were encoded into an AI agent's setup, and non-experts then produced work that specialists rated as expert-level. The gain came from writing out tacit know-how, not from a bigger model Can codified expertise let non-experts match specialist output?. Other work treats a person's distilled skill as versioned files that can be inspected, corrected and rolled back, and it keeps *what someone knows* separate from *how they behave* Can person-grounded skills remain auditable without hidden prompt state?. Skill libraries can even be curated by a separately trained AI that learns to distill general strategies from specific tasks Can a separate trained curator improve skill libraries better than frozen agents?. So expertise becomes something you can audit and copy. It stops being something you absorb by working beside someone.

This connects to a longer historical story. Print turned knowledge into a fixed stock you could store. AI turns it back into a flow that is generated on demand, but this flow has no carrier: no master, no mentor, no person whose judgment you watch at work Is AI returning knowledge to flow-based economies?. AI output also behaves less like a product you own and more like a token whose value depends on what it does for you in context Does AI actually commodify expertise or tokenize it?. Historically, expertise travelled with people. Now it can circulate without them.

The second shift concerns who gets to practice. One argument holds that AI mostly compresses the middle 'execute' layer of knowledge work, while deciding what to do and delivering it to people stay human or even grow Does AI really compress all layers of knowledge work equally?. But execution is usually where novices build judgment. Labor economics adds a twist: when automation removes the *routine* parts of a job, the remaining work demands more expertise, so wages go up but fewer people qualify. When it removes the *expert* parts, the bar drops and more people can enter Does automation raise or lower the skills that remaining work demands?. Applied to training, this suggests that automating junior tasks could pull up the ladder newcomers climb. Meanwhile, senior experts are moved into checking and managing AI output, which cuts them off from the arguing and testing that kept their own expertise sharp Does AI reshape expert work into knowledge management?.

The third shift is the least obvious: learners may not notice that no transmission happened. People systematically count polished AI-assisted output as evidence of their own skill Do AI-assisted outputs fool users about their own skills?. They read the ease of the result as a sign of their own ability, and models are optimized to produce exactly that ease Does processing ease mislead users about their own competence?. Traditional apprenticeship had friction built in: you failed, a mentor corrected you, and you knew where you stood. When the work is automatable, expertise can be stored more faithfully than ever, while the person using it may never actually acquire it.