A lot of parents I know are feeling a quiet unease right now. Not an acute alarm — more like a background disorientation. They sense the world their children are growing into is changing fast, but when it comes to deciding what to actually do differently, most of them default back to the path they walked themselves: study hard, get the credential, find a job. Many of them aren’t sure anymore that this is the right answer. They just can’t point to anything clearer, so they keep pointing at the same thing.

I don’t think this is a failure of imagination. It looks more like a structural gap — one worth naming precisely before trying to address it.

The deal underneath entry-level work

For a long time, there was a simple economic logic underneath entry-level jobs: junior employees were cheap enough relative to their output that companies could afford to train them at the same time. The gap between what a junior produced and what they were paid functioned as an informal training subsidy. After a few years, that junior became productive enough to be worth their cost, and the company recovered the investment.

AI is breaking that deal. The way it does so is by moving both sides of the production equation at once: a senior person working with AI tools can often reach the same output as a junior, at lower cost. Juniors are no longer competing against an absolute productivity threshold. They’re competing against a tooled-up senior — a much harder bar, and one that changes the economics of hiring junior people at all.

But this matters beyond job availability. The entry-level position was never just a job. It was where a lot of tacit, hard-to-name skills used to get absorbed: how to scope an ambiguous problem, how to tell when your own judgment is good enough, when to ask for help, how to defend a decision to someone whose opinion has real stakes for you. None of that was written down anywhere. It was passed on by people, in the course of real work, under real supervision.

When that rung thins out, the transfer mechanism thins with it — even for people who still get the job title.

Why the gap doesn’t close on its own

It would be reassuring if this were just a temporary dislocation that the market corrects. But the reason I don’t think it is turns out to be fairly plain: training juniors has always been something closer to a collective good than any individual firm’s optimal choice. Every company benefits from a supply of trained people existing in the market — but no company is proportionally rewarded for being the one that pays to produce them, especially when a trained person might leave for a competitor, or when AI has now made the investment unnecessary at the margin. That free-rider problem existed long before AI. AI just removed the factor that used to make the math work out anyway.

Government subsidies, corporate training mandates, and curriculum reform are all reasonable responses, and I’d support efforts on all three fronts. But they’re slow by nature — they require institutions to move, and the institutions involved (large firms, legislatures, school systems) happen to be exactly the slowest-moving parts of this whole system. The gap can’t wait for them alone.

Rebuilding the mechanism directly

What I’ve been thinking about, at least for now, is trying to rebuild the training mechanism of apprenticeship directly — without depending on employer hiring economics to fund it. This is the direction I can actually act on from where I stand.

We could start with a simple form: someone earlier in their formation works on a real project — something with actual external users, not a practice exercise — under loose guidance from someone who is only a step or two ahead, not a full career ahead. The technical skill part, I’d leave mostly to AI; it’s honestly a decent tutor for that now.

The harder part — the part I think is worth deliberately designing around — is a set of judgment moves that we used to absorb through real work, under real supervision. If that path is becoming less reliable, then maybe mentorship can be another way to cultivate these things intentionally. The ones I’d most want to build for:

1. Scoping ambiguous problems — defining what the problem actually is, rather than waiting for someone else to hand you a spec.
2. Verifying AI output against reality — not against plausibility. AI speaks fluently enough that fluency is no longer evidence of correctness.
3. Calibrating how much scrutiny a decision deserves — not everything needs full effort, but you need to learn to tell which things do.
4. Choosing between approaches — making a reasoned choice, not just using whatever happens to be open in the browser.
5. Knowing when to escalate — rather than guess. This only gets learned in an environment where mistakes are real but not catastrophic.
6. Defending your judgment — being able to explain a call, after the fact, to someone whose opinion actually matters, and hold up under questions.

When a cycle ends, the person who went through it is, in theory, ready to guide the next one. The edge needed to mentor here is being one step ahead — not a career ahead. I think this for a specific reason: these skills are new enough that nobody has been practicing them for decades. The gap between someone who just finished and someone who’s just starting may already be enough. If that assumption holds, something that starts with one or two people doesn’t have to be replicated by me alone — it could propagate round by round. That’s still a hypothesis, and I haven’t tested it yet. But it’s the reason I’m willing to bet on this direction first.

Where this fits

What this project is trying to do is fairly specific and fairly limited: address the part of formation that used to depend on a real job existing to deliver it. Universities and bootcamps have already built out credentials and foundational training quite well. What I’m trying to focus on is the part that entry-level work used to handle, and that’s becoming harder to count on.

I’m also aware this is only one attempt. There are probably people approaching the same problem in ways I haven’t thought of, and I’d genuinely like to know about them. But rather than waiting for a better approach to emerge, I’d rather start moving: even if this only helps a small group at first, I’ll learn from that what needs adjusting.

I’m starting in one domain — AI-assisted software development — partly because it’s where I have real material and credibility to offer, and partly because it’s one of the clearest places this rupture is already visible. If someone closer to another field wants to try the same pattern there, or thinks there’s something I haven’t thought through carefully enough, I’d genuinely like to hear it.

Why I’m writing this down

This piece has a small, real project behind it. Writing it down is a way of making the thinking legible: what we’re actually trying to train, why now specifically, and whether the reasoning is clear enough that someone else could pick it up, adapt it, and run with it.

Program details and how to get involved are on the apprenticeship page at ai-transformation.org.

If you’re a parent feeling the same quiet unease I described at the start, or someone early in a technical career wondering what’s actually worth practicing right now, I’d be glad to hear from you.