The Cement Truck Won't Replace the Engineer

A civil engineer calculates how much cement a foundation needs, checks the load-bearing math, signs off on the structural drawings — and then doesn't personally mix a single bag of it. A crew does that. Nobody looks at that division of labor and concludes the engineer's job is disappearing. The opposite is true: the engineer is the reason the building doesn't fall down, and the crew is the reason it gets built by Friday.

Software engineering is about to look a lot more like that.

The mixing and pouring is getting automated

For the last two decades, writing code — the actual typing of functions, wiring up a CRUD endpoint, fixing an off-by-one error — was the job. It was also the bottleneck. Now AI agents can generate, refactor, and test large chunks of that code directly. That's not a hypothetical; it's already how a lot of software gets written this year.

This piece isn't about researchers building the models — that's a different profession with a different set of concerns. It's about the people who build products, services, and pipelines with them: developers, architects, data engineers. The builders.

For that group, the mundane part of the job — the mixing and pouring — is being handed to a crew of agents. What's left, and what's suddenly in much higher demand, is the part that was always the actual engineering: understanding the problem, deciding what should be built, designing how the pieces fit, and making the call on trade-offs no model can make for you because it doesn't know your users, your constraints, or your business. Coders, in the narrow sense of "people who translate a fully-specified task into syntax," get reduced. Engineers — the ones who can think it, analyze it, and orchestrate a team of humans and agents to ship it — become more essential, not less.

That's a promotion in disguise, not a layoff notice. But it only works out that way for people who were already doing the thinking part of the job, not just the typing part.

We've watched this play out before

Rewind about twenty years, to the early days of the commercial web. Knowing HTML, CSS, and a bit of JavaScript was, on its own, a genuinely lucrative skill. You didn't need a computer science degree — a lot of very well-paid people at the time were self-taught or came out of a short vocational program — and you could build a client a website and get paid well for it, because almost nobody else could do it.

That job didn't vanish. But it's almost unrecognizable in isolation now. "I know HTML and CSS" stopped being a business on its own; it got absorbed into a bigger bundle — digital marketing, brand and identity work, ongoing maintenance of cloud infrastructure that didn't even exist back then. Fewer people make a living purely from hand-coding a page from scratch. More people make a living from the wider, messier job that swallowed it.

That's the pattern worth paying attention to: the narrow hard skill didn't disappear, it got absorbed into something broader, and the money moved toward the people who could operate at that broader scope. Software engineering, at a larger scale, is entering the same cycle.

The market is pricing in generalists

If there's one uncomfortable takeaway here, it's this: for the last twenty years, the market rewarded specialization — go deep on one language, one framework, one narrow skill, and you were set. That's flipping. The market is now pricing in generalists: people who can gather requirements from a confused stakeholder, analyze a messy problem, design a workable solution, orchestrate the implementation across both human teammates and AI agents, get it tested and deployed, keep an eye on it once it's live, and then stand up and explain — clearly, to non-technical people — why any of it mattered.

None of that is a technical skill in the traditional sense. It's judgment, communication, and the ability to hold a whole system in your head instead of just the function you were assigned. It was always valuable. It's just no longer optional.

What's next

This is the first stop on a longer walk through what AI is actually doing to the computer science job market — not the panicked kind, the useful kind. If engineers are being asked to orchestrate more than they type, the obvious next question is what school is supposed to teach them before they're handed a badge for the job site. Right behind it is a stranger one: why so many companies would rather hire one senior who already knows how to run an AI crew than train three juniors the way they used to. Somewhere in there is a conversation worth having with any teenager who likes taking things apart — about what's worth learning outside of class, long before a CS degree is even on the table. And underneath all of it sits a number nobody likes to say out loud yet: what companies are spending on tokens and subscriptions compared to what they used to spend on payroll, and whether that trade is actually working out the way it looks on the slide.

Consider the foundation poured. The next few posts are the framing.