by Fiona Reynolds

There’s a quiet crisis unfolding in the workforce, and it doesn’t look like the dramatic robot takeover we’ve been warned about. It looks, instead, like a missing internship.

A recent New York Times article on AI and programming jobs paints a picture that should stop every school leader, curriculum designer, and business executive in their tracks. Experienced software engineers are using Generative AI to do in minutes what used to take a junior developer days. And here’s the thing: it’s working. Productivity is up. Code ships faster. Experienced engineers love it.

But junior developers? They’re not getting hired. The entry-level work — the grunt work, the learning work, the work that used to be the curriculum of early careers — is being automated away.

And we haven’t fully reckoned with what that means.

The Wisdom Loop

Here’s what strikes me about the Times article, and what I can’t stop thinking about: experienced engineers can leverage AI precisely because they have the knowledge and experience to evaluate its output. They know when the code smells wrong. They know when the AI has taken a shortcut that will create problems three months down the road. They understand the problem space well enough to ask better questions, catch the errors, and know what “good enough” actually looks like.

In other words, you need deep expertise to be a good supervisor of AI.

But how did those engineers get their expertise? They got it by doing the entry-level work. By making mistakes on small projects. By having a senior developer look over their shoulder and say, “I see what you were trying to do here, but here’s why this approach creates problems.” They learned by being in the messy middle of figuring things out.

This is what I’d call the wisdom loop — and right now, AI is breaking it.

I had that same loop as a teacher and leader. I became a good teacher by trying to figure out how to support the child at the back of the room with their hoodie covering their head, trying to blend into the background. I read, I talked to other teachers, I tried new strategies, and I learned. The toolkit I built allowed me to design learning for lots of different students quickly and effectively.

If AI absorbs all the entry-level work, the next generation of professionals never gets the formative experience that creates the judgment needed to supervise AI well. We end up with a workforce that has tools no one knows how to steer.

This Isn’t Just a Tech Problem

You might be thinking: okay, this is a coding issue, a software world problem. But I’d push back on that hard. We’re talking about entry-level work across industries — the analyst who reviews spreadsheets, the junior accountant reconciling the books, the new marketing hire writing the first draft of the copy, the research assistant pulling together a literature review. All of that work is increasingly being done, or assisted, by AI.

In each case, the same paradox holds. The senior professionals who know what good looks like can use AI brilliantly. But they know what good looks like because they did that work first, with their hands, with their brains, with the patient (or impatient) guidance of someone more experienced.

We are potentially looking at a generational skills gap that doesn’t announce itself loudly. It shows up quietly, years from now, when companies discover they have plenty of AI tools and not enough people who understand the work deeply enough to direct them well.

So What Do Schools Do With This?

This is the question I keep returning to, and I don’t think there’s a simple answer, but I do think we’re asking the wrong questions right now.

We have spent a lot of energy in education debating whether students should be allowed to use AI on assignments. That debate, while worth having, is a small boat in a much bigger ocean. The deeper question is: what are we actually preparing students to do?

If the skills of the future are fundamentally about direction, evaluation, and judgment — about knowing what good looks like well enough to get AI to produce it — then our curriculum needs to reflect that.

In practice, this means:

In a writing class, the goal isn’t just to produce a good essay. It’s to develop the aesthetic judgment to recognize when an essay isn’t good — and articulate why. To know what makes an argument weak, a transition clunky, a thesis buried. That judgment doesn’t come from having AI write the essay. It comes from struggling with sentences yourself.

In a science class, the goal isn’t just to complete the lab report. It’s to understand methodology well enough to spot when a conclusion doesn’t follow from the evidence — including when AI draws the wrong conclusion from a dataset.

In a business class, the goal isn’t just to analyze the case study. It’s to develop the situational wisdom to recognize when an analysis is technically correct but misses the point entirely.

The thread running through all of these? Deep content knowledge, married to the habit of critical evaluation, applied to the outputs of tools that are increasingly doing the first draft of everything.

Schools that are doing this well are teaching students to be discerning consumers and directors of AI — not just users. There’s a significant difference.

What Businesses Need to Understand

Companies are currently benefiting enormously from AI taking over entry-level tasks. The return on investment is real and it’s immediate. But businesses would be wise to think two steps ahead: where are the experienced professionals of 2035 going to come from, if the developmental experiences of 2025 have been automated away?

Some forward-thinking companies are already grappling with this. There’s a growing conversation in certain industries about deliberately preserving some entry-level work — not because AI can’t do it, but because humans need to do it to develop. Think of it like the argument for doing mental math even when calculators exist, or reading long-form books when summaries are available. The output isn’t always the point. The cognitive struggle is.

For businesses thinking about their talent pipelines, there are some urgent questions worth asking:

What formative experiences do your best employees have in common? How much of that experience is replicable if the entry-level role looks fundamentally different? Are you investing in other ways to develop junior talent — apprenticeship-style mentoring, structured project rotations, deliberate stretch assignments — if the traditional tasks are gone? And critically: are you communicating any of this back to the schools and universities that are trying to prepare your future workforce?

That last point matters more than we acknowledge. The feedback loop between business and education is weak on a good day. Right now, we need it to be strong.

The New Core Competency: Knowing What You’re Looking At

The core competency of the AI era might be epistemic judgment — the ability to evaluate the quality of information, outputs, and reasoning. Not just to produce things, but to assess things.

This is not a new idea. It’s essentially what we’ve always meant by “critical thinking.” But AI gives it new urgency and new specificity because AI can produce fluent, confident, plausible-sounding outputs that are completely wrong. And the people who will catch that, who will add real value in a world of abundant AI-generated content, are the ones who know the subject well enough to push back.

That means schools need to double down on depth over breadth. We need to  develop real domain knowledge, not just surface familiarity. We need to teach students to ask better questions, not just any questions, but informed questions that come from understanding the terrain.

And it means businesses need to think carefully about how they develop that depth in their people, if the traditional on-ramp of entry-level work has narrowed.

We don’t yet know exactly what the workforce of 2035 will look like. But I’d bet heavily on this: the people who will thrive in it are the ones who understand things deeply enough to direct, evaluate, and push back on the AI doing the work beside them.

The question for us, right now, is whether we’re building an education system and a talent development ecosystem that produces those people.

Because if we’re not, the AI will get smarter, and the people supervising it will know less and less about what they’re actually supervising.

That’s not a future any of us should be comfortable with.