A parent asked us a version of this question last year, and it has come up in almost every conversation since.

Her son wanted to study computer science. She had just watched an AI write a working program in about forty seconds, from a sentence of plain English. She was not being cynical. She was being a parent, and the question was reasonable:

Is she encouraging her child toward a field that will not exist by the time he graduates?

It deserves a real answer rather than reassurance.

The acronym covers science, technology, engineering, and mathematics, but as a category it is close to useless, since it lumps together a marine biologist and a network administrator. What holds it together is a method rather than a subject: observe, explain, test, be wrong, revise.

That distinction matters enormously for the AI question, because AI has become very good at the domain detail and has changed almost nothing about the method.

Be honest about this part, because teenagers can tell when adults are not.

Some real skills lost most of their market value in a very short time. Writing routine code from a clear specification. Producing a first draft of standard documentation. Recalling syntax. Assembling a competent summary of known material. A student who spent four years becoming good at exactly those things is entering a harder market than they expected.

Pretending otherwise does not help anyone. The floor moved.

But notice what all of those tasks have in common. Each one starts after somebody has already decided what needs doing, and each one is judged by a standard somebody else already set.

That is the part AI handles well. It is also the smaller part of the work.

Four things went up in value:

AI will answer whatever you ask it. It will not tell you that you are asking about the wrong thing. Almost every serious failure in engineering starts as a well-executed answer to a badly chosen question.

A model will produce a confident, plausible, entirely incorrect answer, and it will do so in the same tone it uses when it is right. Catching that requires understanding the underlying material. The people who can no longer be replaced are the ones who know enough to notice.

Real problems come with budgets, regulations, physical limits, and people who will not use your solution the way you imagined. None of that is in the prompt.

Persuading a team. Handling the disagreement in week six when the design is not working and everyone is tired. No tool has touched this.

None of these are new skills. They are the same ones that made engineers valuable in 1985. What changed is that they used to be bundled with a lot of executable work that hid how much they mattered. That bundle came apart, and now they are exposed.

Here is the uncomfortable implication for schools and programs.

The traditional model teaches domain knowledge first and defers judgment until much later, usually until a job. You learn the syntax for years, then eventually somebody lets you decide something.

That order was always debatable. It is now actively backwards, because the first half is the half that got automated.

A student who has only ever solved problems that were handed to them, with a known correct answer, graded against a rubric, has practiced precisely the skill that no longer distinguishes them. They may have excellent grades and very little of what the work now requires.

The alternative is not to abandon fundamentals. Judgment without knowledge is just opinion, and you cannot tell that an answer is wrong if you do not understand the subject. The alternative is to teach both at once, on problems where the answer is genuinely unknown.

If you are evaluating options for your teenager, the AI question turns into a fairly practical filter.

If every student works on the same assignment with a known solution, they are practicing the automated half. If students identify their own problems, they are practicing the half that matters

A program where every project succeeds is a program where nothing was genuinely at risk. Failure is not a bug in the learning; it is most of it.

Banning it entirely is unrealistic and prepares students for a world that no longer exists. Allowing it with no discussion teaches nothing. The useful middle is students using these tools while being expected to understand and defend every part of what they produce.

Ask whether anyone outside the program ever sees the work. A rubric can be satisfied. A real user cannot be argued with.

Yes, but the reason has shifted.

The case for STEM was never really that it leads to a specific job. Jobs have been appearing and disappearing throughout the entire history of the field. The case is that it teaches a way of approaching unfamiliar problems that transfers to almost anything, including work that does not exist yet.

Your teenager will change careers several times. Some of what they eventually do has not been invented. What they need is not a durable set of facts, which is exactly the thing AI now supplies on demand. What they need is the ability to face something they do not understand, break it apart, test it, be wrong, and keep going.

That is what STEM is, underneath the acronym. It has not been automated, and there is no particular sign that it is about to be.

Our program is built around students finding a real need with a community partner, prototyping a response, piloting it, measuring whether it worked, and then doing it again with what they learned.

There is no answer key. Some projects do not work. Students are guided by working professionals rather than instructors, which means they are getting judgment from people who exercise it for a living.

We did not design it in response to AI. It happens to be what the current moment rewards.

Our cohorts open and close on a schedule for students in grades 8 through 12 across Vancouver and Richmond. If applications are closed when you read this, join the list and we will let you know before the next intake opens.