The AI Shift

I have spent a meaningful portion of my adult life arguing about tabs vs. spaces. I have strong opinions about function naming conventions. I once rewrote a sorting algorithm at 11pm not because it was wrong, but because it offended me aesthetically.

I tell you this so you understand the kind of person who finds the current moment in software deeply, personally unsettling.

The Bottleneck Moved (And We Weren't Ready)

For most of computing history, the constraint was simple: humans type slowly, think carefully, and need coffee. Writing code took time. Not because we were bad at it — because software is genuinely hard to specify, and half the requirements only revealed themselves once you'd written the wrong thing three times.

Then AI showed up and said: "what if we wrote the wrong thing three times in forty seconds?"

And somehow that's... better? Studies say developers finish work 55% faster. Another found a 26% jump in completed tasks. These are real numbers from real researchers. I have checked. The gains are not made up.

The catch is that when the coding speeds up, something upstream breaks. Turns out "write the code" was never really the hard part. It was the thing we did while we figured out what we actually wanted. Now that crutch is gone and everyone has to confront the terrifying truth: requirements are hard, and we were using implementation time to avoid writing them down properly.

Congratulations. We optimized our way into doing the hard part.

The Product Looks Exactly the Same

Here is the twist nobody warns you about: the software coming out of this revolution is completely normal.

React apps. REST APIs. Mobile apps that remind you to drink water. Slack integrations nobody asked for. Same as always. Your users have no idea. Your pull requests look identical. The deployed service does not announce that it was built by an AI that hallucinated three variable names and had to be corrected like a very confident intern.

The disruption is entirely behind the curtain. Same shirt, completely different factory. The thing that changed is not what gets built — it's who builds it, how fast, and whether they spent their afternoon actually thinking or just vibing in a chat window with a language model.

Personally I find this funnier the more I think about it. The robots took over and nobody noticed because the robots make the same CRUD apps we always made.

You Are Now a Manager. I'm Sorry.

Here is where I have to deliver some bad news.

The new job — the job AI is actually creating — looks a lot like... management. You set intent. You delegate. You review work you didn't do yourself. You spend a lot of time writing things down clearly so someone else can execute them without asking seventeen follow-up questions.

If you became an engineer specifically to avoid that career path, I understand your frustration. I've been there. I am there.

But here's what I've come to accept: the craft didn't disappear, it moved. Writing a prompt that actually produces the right code is a skill. Breaking a problem into pieces small enough for an agent to handle without going off the rails is a skill. Reviewing AI-generated code — which looks very confident and professional right up until it definitely doesn't work — is its own entirely new skill that no one has fully figured out yet.

The musicians who survived the DJ era didn't quit music. They learned what the room needed and figured out how to deliver it with the tools available.

We're in that transition now. The engineers thriving are the ones who realized that "where does quality actually get made?" is a more interesting question than "did I write this line of code personally."

The answer, increasingly, is earlier. In the problem statement. In the decomposition. In the review.

It's still ours to own. It's just a different kind of ownership than we trained for.

Also I still have opinions about variable names. Some things don't change.