If AI now does the entry-level work juniors used to learn from, can mentoring still teach them when to trust it?

This explores whether mentoring inside an organization can teach junior workers good judgment about when and how to rely on AI. The short version is that the corpus doesn't test mentoring programs directly. What it does show is why mentoring has become harder, and also why it matters more than it used to. The clearest evidence comes from interviews with South Korean software engineers Does generative AI prevent juniors from getting entry-level work?. Generative AI is taking over the routine entry-level tasks that juniors used to learn on. Seniors now hand that work to AI instead of to a junior. The awkward part is that seniors and juniors see the problem differently, so the people best placed to mentor may not notice that the old apprenticeship path has disappeared.

That gap matters because judgment about AI can't simply be handed over as a rule. One line of research describes what it calls the 'LLM Fallacy' How does AI-assisted work reshape how people see their own abilities?: people start treating the AI's output as evidence of their own skill. This is a mistake about themselves, separate from trusting a wrong answer or leaning on the tool too much. Making the AI more accurate doesn't fix it, and neither does forcing users to double-check. What helps is making clear which parts of the work came from the person and which came from the machine. That is a natural job for a mentor: sitting next to a junior and asking which part they actually understood.

There is also a social barrier. In four experiments with over 4,000 participants, people using AI expected to be judged as less competent and less diligent. They were less willing to tell managers and colleagues they had used it Do people fear judgment when they use AI at work?. If juniors hide their AI use, mentors can't see the habits they would need to correct. Meanwhile recruiters reward 'AI skills' on résumés without checking whether candidates actually have them Do AI skills help candidates get more job interviews?. So the incentives point toward looking fluent with AI rather than using it well. A mentoring culture would probably need to make AI use openly discussable before it can teach anything about it.

Some nearby research suggests what good guidance might look like. 'Learning to Guide' found that AI works better when it points out which parts of a problem deserve attention than when it simply hands over a decision. Guidance avoids the anchoring effect, where people lock onto the AI's answer, and the human stays responsible for the call Can AI guidance reduce anchoring bias better than AI decisions?. That is roughly how good mentoring works too. Another study found that AI suggestions can hurt reasoning even when they are correct, because they break a person's concentration Does AI assistance always help reasoning or does it carry hidden costs?. A mentor could teach juniors when to keep the assistant out of the way. Research on AI agent teams adds a warning: mixing different viewpoints only improved ideas when real senior expertise was present Does cognitive diversity alone improve multi-agent ideation quality?. Those results are about AI agents, but the idea carries over to people: experience has to be part of the mix for diversity to help.

Here's the bigger idea you might not expect. AI now produces polished intellectual work without the thinking that normally sits behind it Does AI separate intellectual form from the thinking behind it?. A junior's work can look expert while the reasoning is missing. That makes mentoring more valuable than before, because a mentor can probe whether real understanding sits behind the finished product. It also makes mentoring harder, because the finished product no longer reveals much about the understanding behind it. The corpus doesn't yet have studies testing which mentoring practices actually work. That's an open gap in this collection.