“Are we all really going to be replaced?”

At lunch, a coworker slid his phone across the table. A web demo — built entirely by AI.

“Apparently, all it took was a single prompt.”

Lately, my YouTube feed is flooded with videos that feel overwhelming.
”Stop learning to code.” “99% of developers will be replaced.”
Clickbaity thumbnails take over the algorithm, basically declaring that coding is dead every single day.

The grind to become a developer. All the late nights, the frustration, and the breakthroughs. It felt like all that hard work was for nothing—all because someone who’s never written a line of code clicked a button.

I had to see for myself. Was this just clickbait fear-mongering for views, or a real threat breathing down my neck?
If I gave the AI a task I knew absolutely nothing about, could it really deliver with just a single ‘click’? I decided to put it to the test.

Deliberate Ignorance

3D programming — something I’d always been curious about, but had never once touched. Complete, total ignorance. The goal: modeling a 3D jellyfish drifting through the deep sea. I saved one photo of a jellyfish from Google. jellyfish.png

I opened a terminal. “Create a 3D jellyfish model using jellyfish.png as a reference.”

The AI spat out a massive wall of code—over a thousand lines in one go—along with 11 cryptic GLSL shader files. It was overwhelming.

But the real problem lay elsewhere: I had no way to judge whether this thousand lines of code were garbage, the right solution, or if there was a much better way to do it.

Trash In, Trash Out

When I ran it, something appeared on the screen—but it was glitching out in the creepiest way.
I was hoping for a jellyfish, but what I got was a distorted mesh that looked more like a ‘monster in a jellyfish suit.’ Without any 3D domain knowledge, I couldn’t give the AI meaningful feedback.

“Make it more natural."
"No, not that.”
That was the full extent of the feedback I could give. Completely helpless.

There was no silver bullet. Bugs started popping up everywhere. The tentacles were twisting in all the wrong ways. All I could say was, ‘Something looks off about the rotation.’

Whenever I switched browser tabs and came back, the physics would just explode. The velocities spiked out of control, and I had no idea why. I couldn’t tell if the AI’s code was a real solution or just a patchwork of duct tape.

It turned into a game of Whac-A-Mole: every fix spawned a new bug. Plugging one hole just made another one burst open. Without a fundamental understanding, my ‘band-aid’ fixes were just polluting the entire structure.

It was only the next day, after studying Verlet Integration from scratch, that I could finally strip away the patches and cut out the root cause.

“The power to give meaningful feedback comes from your fundamentals and domain knowledge.”

Middle-Out

I had to study. The question was how.

The first option: Bottom-up.
Grinding through linear algebra and WebGL specs. It’s the ‘correct’ way, but I didn’t have the luxury of time to bury myself in textbooks while my render was broken.

The second option: Top-down.
Reverse-engineering the AI’s output. But this is a trap—it’s a fast track to becoming a passive consumer, forever babysitting code you don’t truly understand.

I chose Middle-out.
I leveraged React—my most reliable weapon—as the anchor and only filled in the gaps when I hit a wall. I didn’t study for the sake of studying; I ‘transfused’ knowledge only when necessary.

When a shader bug popped up, I’d hunt down a specific GLSL tutorial on Udemy, absorb just enough to fix the immediate problem, and get back to the code. No mindless copying, no endless theory. Just sharpening my tools as I went.

As I filled these gaps, my questions evolved. ‘Why isn’t it showing?’ became ‘Is the transmission pass missing a transparency property?’

5 Weeks. 47 Commits.

Looking back, my learning was neither strictly top-down nor bottom-up. With React as my anchor, I reached up toward AI to gain speed and productivity, and down into the fundamentals—WebGL, shaders, and physics—every time I hit a wall. What I knew became the reins; what I didn’t know became the direction. Middle-out: learning driven by necessity.

GitHub: diver-jay/r3f-drei-water-caustics-effect

The fear was just ambiguity. Once I started, that ambiguity disappeared. AI can write code I don’t understand—but without the logic, ‘not that’ is the only feedback I can give.

The era of AI has changed how I learn and how I lead.

I no longer dread the technological shifts ahead. I see them clearly now.