Google fires engineers. Meta fires engineers. Amazon fires engineers. Not dozens — thousands. The same week, you watch the stock price of software companies quietly slide on the market. Then the headlines arrive: "AI is replacing developers." "Software engineers are next."
I wanted to know if any of that was actually true. So I tested it.
The experiment
I subscribed to the best AI available, set up the environment, and picked a project that was real — not a tutorial exercise. I was not going to build a to-do list or a currency converter. Those prove nothing. I wanted something with actual complexity: a trading portfolio management system with execution capabilities and analytics. A domain I know well, having built multiple systems in it. The choice was deliberate — I needed to be able to judge the output, not just run it and hope.
Route one: free range
I gave the AI a complete description of what I wanted. The features. The rules. The expected behavior. Exactly what a business owner communicates to a developer: here is what I want, here is what I expect to see, figure out how. Then I let it run. For hours. For days.
It finished — and honestly, that surprised me. Faster than I expected. The design came out well: animations, color choices, layout. The things most developers rush through or skip entirely. The AI treated them like they mattered.
Then I tried to run it.
Going back and forth to get it running took time. More time than it should have. The AI had made decisions I never asked for, skipped things I assumed were obvious, and confidently delivered features that did not behave as described.
Route two: with a leash
I went back to the start. This time I provided the architecture. The guidelines. We wrote tests before we wrote features. I reviewed code at every step and pushed back when something was off. The AI responded to the structure.
Two months later, everything worked. Not mostly everything — everything. Smooth, expected, correct. What I had specified was what I got.
The AI did not get smarter between route one and route two. I got more specific about what I wanted and how I expected it to be built.
Then I read the code
This is the part that stayed with me.
When a project ends, I read the code. Old habit. I opened it up and started going through it.
It read like a junior had written it.
Repetition everywhere. Functions that did the same thing, written three different ways, scattered across files. No abstraction. Logic that could have been one clean utility was copy-pasted instead — five times, ten times, with slight variations each time. The code worked, but it carried the weight of someone who knows how to write lines without knowing how to organize them.
So I started enforcing design patterns. Pushed the AI to refactor. It followed the direction. The code cleaned up. By the end, it read like something I would have written myself.
Total time, both routes combined: fifteen weeks. The same time it would have taken me to do it alone.
What I actually learned
AI is not going to kill the software developer. Not because software is too complex — it isn't always. Not because AI will plateau — it won't. But because software development is a delegate field. It requires someone on the other end who can define what "done" actually means, catch the difference between something that looks finished and something that works, and push back when the output is technically running but structurally wrong.
The people carrying carton boxes who got replaced by machines did not lose their jobs because AI was particularly clever. They lost them because the job had a ceiling — learn it once, repeat it forever — and they chose to stay there. That is a quiet choice, made over years.
Software developers who treat AI as a tool — who learn to direct it, review it, and push back on it — become faster. The ones who don't evolve will find themselves outpaced. Not by the AI. By someone who learned to use it properly.
AI is a very fast junior developer. Impressive output, questionable judgment, entirely dependent on whoever is giving the direction. That person's job has not gone away. It has gotten more important.
