It is easy to forget sometimes in the excitement, but nobody has been using (2026) AI for 20 years. We're all still new. I am sure that in the next year, someth

It is easy to forget sometimes in the excitement, but nobody has been using (2026) AI for 20 years. We're all still new. I am sure that in the next year, something will be found that is fairly exciting, and something we could all be doing right now, but it's simply that nobody has thought of it yet. Or something that is today common practice will become generally considered an anti-pattern and common practice will have some replacement for it that, again, nothing stops us from doing it today but nobody has thought of it yet, because we're all newbs. (One candidate example for this is the discussion I've seen in the last few days about not trying to negate something, to say "Don't do X", but instead stay positive because eventually the negation gets lost in the context window and you're better off just not putting the idea in the LLM's mind at all, where "Don't do X" comes to be seen as an LLM antipattern.) One of the consequences of none of us having used AI for long enough is that we don't know how to onboard developers in an age of AI. This will be, by necessity, transient. Eventually we're going to max out what a person can do and we'll need more people. The supply of existing engineers will be limited. We will be forced to discover how to onboard new engineers. But at the moment we've got our hands full, and we don't know how to do it. The irony is, the best time to join a field is often exactly when the enrollment dips and the worst can be precisely when it is the most popular. Start a programming college program today and the odds that in 4 years we'll have onboarding figured out and have developed some sort of need for fresh developers is pretty decent. But I don't know what to do about the fact that the standard CS curriculum was already of debatable relevance to me in the late 90s and I don't know of what relevance it will be in four years except to guess that it very likely to be even less. I do know that we are again affected by the fact nobody has been doing this for 20 years, like I mentioned above. There is no body of "wisdom" for an AI-powered world to draw on to construct a new curriculum. Universities would be inclined to do the obvious thing and try to chase our current practices with AI but those aren't going to be stable enough to build a curriculum on any time soon, and a real fundamentals-based curriculum may involve less AI than people may think. I know one advantage I have over my younger peers at this point is just a knowledge of what terms to say to the AI to get it to do what I want, words like "event sourced" or "message bus" or "stored procedures", where simply knowing that the concept exists is the bottleneck. I could see a programming curriculum based on touring through a whole whackload of concepts with their pros and cons, or at least, where that is a much larger portion of it. Ask me in 5 years though and I'd almost certainly suggest a completely different curriculum than I would now, though.