← Ascendy 한국어

meta

Build in studying — learning while you build, how non-engineers make vertical AI

· Ascendy Engineering

TL;DR

About this piece. Written from an interview (/interview) — I answered, an AI editor pressed, and the position developed. Assertion and speculation are kept apart (the learning/aptitude logic is one operator’s N=1 observation). Company, family, and unannounced plans are generalized or omitted.

It started from a request at home

All of this started from a request at home: “I have way too many photos — isn’t there a cloud that organizes and finds them with AI?”

I looked. Google Photos, iCloud, every cloud I knew — none did it the way I wanted. And yet I had no intention of building it. I’m a cloud engineer. Not an AI engineer. Serving models and running inference was, I thought, not my field.

Then I just vented the problem to GPT and Gemini and Claude. It was closer to grumbling. And all three said the same thing — “you can build this.”

What AI changed is distance, not skill

Stop here and you can see exactly what changed.

It’s not that I lacked the idea. I just didn’t know how to build it. Before, it would have ended at “oh, nice idea — but how would I build this? How do I serve the model, run inference, stand up a server… ugh, too hard.” You think it, and it ends there.

Now that sentence becomes “oh, I’ll just have Claude and Codex build it.” What AI tore down isn’t ‘the time to learn a skill’ — it’s ‘the distance between an idea and shipping it.’ The idea in a domain expert’s head no longer dies at the wall of “how would I build this.” That’s the core.

Don’t misread it: entry got easier, depth didn’t vanish. Anyone can stand up a plausible MVP now, but the parts where a product carries real weight — security, infra, scale — are still done best by seasoned engineers. What got easier is the start, not the finish.

Build in studying — learning while you build

So I said, “I know nothing, walk me through it from the beginner level.” And talking with the LLMs, I picked up Triton, vLLM, ONNX, TensorRT, LangChain, LangGraph, and pulling models from Hugging Face.

But this learning was completely unlike any learning I’d done in my life. Before, I’d buy a book or take a course — learn first, then build. This time there was no such order. Ask what I don’t know → build → hit a wall → ask again → propose an idea and debate it. It was sitting next to an engineer smarter than me and building together, the artifact and my skill growing at the same time.

There’s a phrase, build in public. This is different. Not building while publishing, but building while studying. So I call it build in studying. (I found that non-experts building directly with AI is already called domain expert building. Similar — but what I want to stress is learning in the process, more than the building itself.)

Why it matters: people say “a domain takes years to accrue, and AI handles the skill.” But build in studying shakes even that premise — because acquiring the domain itself speeds up.

What’s scarce has changed

Vertical AI — AI specialized to a specific industry or domain — being the hottest current trend fits the same logic. A product that goes deep into one domain — owning that domain’s data and workflows — is stronger than a general tool, and investors bet on exactly those verticals.

Here a reversal happens. Before, the person with the skill (the engineer) was scarce, and the domain was layered on top. Now that the skill’s barrier has dropped, the scarce thing has moved to “the domain sense to know what’s worth building” and “the ability to learn something new, fast.”

Which means a career that looks unrelated to engineering suddenly becomes an asset. My own path was non-traditional too — crossing from the humanities and social sciences deep into engineering. What once looked like an “inconsistent résumé” becomes, in the build-in-studying era, all raw material for vertical AI.

But not everyone makes it

Stop here and it’s naive optimism — “with AI, anyone can build anything.” That isn’t true.

No base is fine. But not everyone makes it. Each person has a field that fits. Some are brilliant at law but can’t think in engineering at all; some ace engineering math but can’t follow literature, history, philosophy. AI amplifies your learning ability and your matching aptitude — it doesn’t manufacture an aptitude you don’t have.

Build in studying worked for me because I happen to have no aversion to learning unfamiliar fields — I actually enjoy it. A strong learner who uses AI well has that ability amplified. But in a field that doesn’t fit you, AI won’t save you either. Someone with athletic talent might use AI for diet, conditioning, training and become an athlete — but I couldn’t. Athletic talent is partly innate, and AI can’t close that gap.

The same thing happened at work

This isn’t only a side-project story. With no course and no book, studying alongside AI, I’m now building an internal domain AI system, alone, inside the organization I work for. And colleagues come to me with their AI questions. Somewhere along the way I became the “expert.”

The funny part is I did nothing special. I just worked with them for a long time, and learned a new domain — AI — together with AI. AI is, after all, just another domain; learning it by building rather than from a book shortened the time to accrue it. That’s build in studying working in real life.

Takeaways

The raw material of vertical AI isn’t a flashy tech résumé. It’s the domain you already have × a disposition to enjoy learning. If you have an idea you shelved because you couldn’t code, this is the era where it no longer dies.

But honestly — no base is fine, yet not everyone makes it. AI is a tool that amplifies you, not one that creates a you that isn’t there. So the question isn’t “but I’m not an engineer,” it’s “what do I know, and what do I enjoy learning?”

Authorship & citation: Written by Ascendy Engineering; quotable with attribution. Found something wrong? Let us know via a GitHub issue.

Tags: ai, vertical-ai, vibe-coding, learning, career, opinion