I got into one of those convos with a friend recently, the kind that starts with “how’s your day going” and ends forty messages later with both of us questioning what “learning to code” even means anymore.

Here’s the question that kicked it off for me. At what point in the process of humans inventing something new, from prototyping it, showcasing it, mass-producing it, selling it, and making it commonplace; do you stop teaching people, ESPECIALLY kids, how to build the thing it’s replacing?

Nobody teaches covered-wagon construction anymore, and nobody (well maybe a handful of diehard historians) is mad about that. But there was a window, somewhere between “the car is a novelty” and “most every family owns one,” where that question would have been hard to answer. I think we’re in that window right now with code.

I say this as someone who’s been inside this conversation for a while. Back in 2023, a seasoned developer in my network told me, flatly, that it was over for “learning how to code.” At the time I didn’t really register what he meant. Young kids would show up in my livestream sessions asking the best ways to start learning to program, and his answer was consistently: don’t. The tools are going to change out from under you before you finish learning them…

Code is writing code now. Is that a good thing? Right now, mostly not really, it’s a whole lot of slop, very expensive and resource-intensive slop, genuinely rough output that still needs a human checking the work. But “right now” is doing a ton of work in that sentence. Eventually, and very soon, the fundamentals of computer science and code itself are going to change, not just the tools sitting on top of them. The time it takes for humans to agree on this shift is shrinking faster than ever.

That’s a different claim than “AI will replace programmers,” and I think it’s the more interesting one. It’s not that the job disappears. It’s that what counts as the actual skill shifts underneath it. Brackets and commands, functions and formulas, knowing the syntax well enough to spot a bug by eye, that’s not nothing of course. However, it’s increasingly not the thing that matters most. And that’s frightening to a ton of folks that spent their secondary education lives grinding through learning it.

I’ve also heard the counterargument plenty. That AI can only solve problems up to a certain point of specificity, so the fundamentals are still safe to teach. I think that’s accurate as a description of right now and inaccurate as a forecast. The gap closes fast when the release cadence looks like what’s happening currently in this industry.

Here’s what’s making this feel urgent to me instead of theoretical: new AI models aren’t dropping yearly anymore. Not even monthly. Not weekly. They’re dropping daily, from insanely talented and competitive teams all over the globe, both man and machine in tandem, and the trajectory points toward something closer to hourly. If you want to see it instead of taking my word for it, go look at a model marketplace like OpenRouter or Hugging Face on any given day. New releases stacking up, half of them from labs and teams most people haven’t even heard of yet.

Someone in a dev Discord I’m part of put it sharply not long ago: it’s basically over for traditional development as its own separate category. AI itself is on track to become “the traditional developer” within a few months/years. Not just a tool one uses, but the thing actually doing the work. Currently, that amount of work may be small, but it will indeed try to grow until it is 100% in a majority of cases.

All of the newest models are focused on agentic tasks. If they aren’t, they are behind. AI teams basically mastered the chatbot side of things already. Code is next. Beyond that, honestly, who knows. But one thing is for sure. Time will tell.

I think about AI’s trajectory less like a series of global product launches and more like a ladder of communication ubiquity: chatbot level, smartphone level, internet level, eventually printing-press level. Each rung down that list means the same thing: less impressive, more unavoidable, more infrastructure than tool. Someone half-joked to me about how a regular person would even know if their account got compromised in an increasingly AI world like that. My answer was simple: you’ll ask your agent. It might even already be on autopilot for you, and alert you.

I made a prediction recently that I keep coming back to. This current 2026 generation of blockbuster video games might be the last ones built entirely without AI in the production pipeline. That’s actually wild to sit with. Game studios at that scale represent some of the largest, most resourced dev efforts in the industry, and even that might not be enough insulation from where this is headed.

But betting early on any of this AI and emerging tech advancement is a gamble, full stop. And it’s a gamble that mostly gets taken by people who have the leverage to absorb being wrong. Whether it’s the risk of teaching your kid a skill that becomes obsolete, or the risk of restructuring a team around a technology that’s still mid-slop, the people with the most resources take that risk first, and they teach it to their children before the rest of society has even clocked that the window is open. That access gap is worth sitting with longer than the tech-optimism headlines usually allow.

There’s also a real mismatch happening inside a lot of organizations right now, between what leadership is asking teams to do with the money they’ve committed to AI, and what the people actually doing the technical work say AI is currently capable of. Those two things frequently do not align, and a lot of the current frustration, from workers, from creators, from anyone whose job touches this tech, is coming out of exactly that gap.

So what’s the actual skill, if it’s not syntax? I keep landing on the same answer: problem-solving. Solution-thinking. The ability to sit with a messy, half-defined situation and creatively brainstorm a way through it. Those are the people who stay in demand no matter what the underlying tools look like because that’s the layer that doesn’t get automated away. It gets amplified.

I don’t have a clean answer to the question I opened with. I don’t think anyone does yet, which is exactly why it’s worth asking out loud instead of assuming the answer is obvious in either direction. But I’ll leave you with the version of the question I keep sitting with.

If code is now the covered wagon, what does the car actually look like, and how would we know it if we saw it?

Drop a comment or reply to share your thoughts.