Every few decades, software engineering renegotiates what “the job” means. Assembly programmers watched compilers take the part they were proudest of. Sysadmins watched the cloud swallow their racks. Each time, the people who thrived weren’t the ones who defended the old layer — they were the ones who noticed where the craft had moved, and climbed.
It happened again. Machines write most of the lines now. And the loudest reactions — “engineering is dead” on one side, “nothing really changed” on the other — are both wrong in instructive ways.
What actually got automated
Not engineering. Transcription. The activity that got cheap is the one in the middle: turning a decision that was already made into syntax that compiles. That was never the whole job, but it was the visible job — the part interviews tested, bootcamps taught, and juniors were hired to do.
What got more valuable
Everything surrounding that middle step:
- Deciding what to build. Ambiguity doesn’t compress. Someone still turns a fuzzy business problem into a buildable shape.
- Shaping the system. Architecture is a bet about the future cost of change. Models execute those bets brilliantly; they don’t place them.
- Directing the work. Decomposing a system into pieces an agent can execute is a spec-writing skill most engineers were never taught.
- Verifying the result. Generated code fails plausibly. Catching plausible wrongness is a discipline, not a vibe.
- Owning the outcome. Someone’s name is on the deploy. That accountability is the one thing that cannot be delegated to a model.
The uncomfortable middle
Here’s what makes this transition harder than the compiler one: you can’t verify what you don’t understand. The engineers getting 10× leverage from AI are, almost universally, the ones who could have built the thing slowly by hand. The fundamentals didn’t become optional — they became load-bearing for the leverage itself.
That’s the design principle behind everything we teach: understanding first, delegation second, verification always. Not because we’re nostalgic for the old layer — because that’s the only sequence that actually compounds.
The craft isn’t dying. It’s hiring at a higher altitude.