Jun 26, 2026

YAML: Yet Another Meditation on Layoffs

Lately, almost everyone is talking about AI taking over human jobs. In fact, it has already started, and many assume the engineers being laid off today are losing out to AI. I want to weigh in with my thought process on this.

First of all, we are living in an astonishing era. I bet life has never been this exciting yet this uncertain. The uncertainty scares people away, and people tend to be defensive because of their survival instinct. In my opinion, being careful is natural for any living being. There might be creatures that lack a survival mechanism, but humans are not one of them, so it is only natural for people to worry. Before I get into my main argument, I want to state my view on this skepticism clearly. I personally don't think AI will take over software engineers, and here is why.

One. It's statistics, not magic

The way the current AI breakthrough works comes down to one thing underneath all of it: statistics. So why has it been treated like magic? Two reasons. First, for a long time we had no compute power to train models at the scale the theory demanded. Second, and just as important, the transformer architecture introduced in 2017 was a genuine algorithmic breakthrough, not just old math finally getting enough horsepower. Put those together (cheap compute, thanks to Mr. Jensen Huang and the folks at NVIDIA, plus attention-based models that scale) and you get today's systems.

So in my words, today's AI is not really "artificial" intelligence in the way people imagine. It is sophisticated pattern matching over statistical correlations in human knowledge, scaled to a degree that feels uncanny. By saying this, I don't want to underestimate the people who made these things real. If we compare real knowledge generation to mining gold from a pile of dirt, what current technology does is not add new piles but sift the same pile faster than any human ever could. Most people who think AI is so smart are not recognizing the size of that dirt pile. It is mountain-like, so vast that sifting through it feels like creation.

But sifting is not creating. The models are extraordinary at recombining what humans have already explored, and yes, they sometimes surface combinations no single person had put together before. What they rarely do is step fully outside the distribution of human knowledge and find a paradigm that wasn't buried in that mountain somewhere. And this distinction is the crux of why I'm not worried.

Software engineering, at its core, is not the act of typing syntax. It never was. Syntax is the dirt. The actual job is translating fuzzy, contradictory, half-understood human desires into precise systems, and most of the time the desire itself doesn't exist anywhere in the training data yet. When a stakeholder says "make it feel faster," there is no Stack Overflow answer for their product, their users, their constraints. That is a fresh pile of dirt nobody has mined before. AI can hand you the average of every "make it faster" conversation that ever happened, but averaging is not understanding.

Two. The bottleneck moves up the stack

There is a subtler point people miss. The more code AI generates, the more code needs to be understood, reviewed, debugged, integrated, and accounted for. Demand for engineering judgment doesn't shrink when typing gets cheaper. It grows, because the bottleneck moves up the stack.

We have seen this before. Compilers didn't kill programmers; they killed the need to write assembly by hand and unleashed millions more programmers. Higher-level languages, open-source libraries, cloud infrastructure: every one of these was supposed to "replace" engineers, and every one instead multiplied what a single engineer could build, which multiplied demand. AI is the next compiler. It compiles intent into a first draft. Someone still has to own that draft.

Three. Accountability is a human contract

Accountability cannot be outsourced to a statistical model. When the system goes down at 3 a.m., when the data leaks, when a deploy corrupts the database, no one is going to interrogate the model. They are going to ask the engineer. Responsibility is a fundamentally human contract, and you cannot delegate it to something that has no skin in the game and no concept of consequence.

The core idea

So here is my core idea, stated plainly: AI is not replacing software engineers. It is removing the parts of the job that were never the point. The boilerplate, the lookup, the syntax recall, the boring 80 percent. What is left is the 20 percent that was always the actual job: judgment, taste, system design, understanding humans, and owning outcomes. The engineers who treat AI as a threat are the ones who mistook the dirt for the gold. The engineers who treat it as the most powerful shovel ever built are about to become more valuable, not less.

The mountain of dirt isn't getting smaller. We just got a better sieve. And someone still has to know what gold looks like.