A Month of Agent Development — I Became a Creator, Not Just a Developer

If you keep getting distracted while building, that’s not distraction

Opening

A month ago, 70% of my headspace while developing was occupied by thoughts like these. How do I implement this feature. Why is there an error here. How do I connect this API.

Now it’s different. I’m in the middle of building something and I keep drifting into other thoughts. Why does this app need to exist. What does the user actually want. Does this feature connect to the core value.

At first I thought something was off. Was I losing focus. But at some point I realized — it wasn’t distraction. It was leveling up.

It Started Simply

I began developing two things almost simultaneously: a stock portfolio app and myWiki. Both for the same reason — I wanted to use them myself. I wanted to properly track stock returns, and I wanted to organize knowledge coming out of AI conversations so I could actually retrieve it later.

At first, implementation was fun. Writing code alongside AI felt novel, and I enjoyed watching features accumulate one by one each day. The satisfaction of building things directly was still there.

But at some point, thoughts like this kept surfacing while I was working. AI is going to change the work environment fundamentally. Now that implementation is solved, knowing the domain and the business seems more important.

The moment it clicked was while building myWiki, when I had to define the curation criteria. “How does AI decide what to keep and what to discard?” — this wasn’t a problem I could solve with an algorithm. I had to make the judgment myself: “is this knowledge valuable to me?” I wasn’t writing code — I was establishing principles. That’s when it first became clear. As implementation got easier, the real decisions hiding behind it began to surface.

”AI Builds It Differently Than I Had in Mind”

Once I started delegating implementation to the agent, a strange discomfort emerged.

AI would often build things differently from what I had imagined. At first I thought I was giving bad instructions. I tried being more specific with conventions, providing more context. But it kept happening.

Then I understood something. The fact that it felt “different from what I had in mind” meant there was a clear image in my head. I thought I had no ideas, but I actually had standards. Someone who feels “this isn’t right” has a “this is how it should be” already.

I tried full automation too. The results were underwhelming. The quality wasn’t where I needed it. So I arrived at my own conclusion: I stay involved in the core business decisions, and automate everything else. Nobody taught me this — it came from actually trying it.

The Bottleneck Moved Up

Reading a book about stocks, my first thought wasn’t “I should invest.” It was “I should build an app.”

At the time I figured it was just my personality. Later I understood what it meant. Someone who automatically translates any domain into “what tool could I build here?” — a maker’s instinct.

And I started to see how this connects to the AI era. Developers become managers, managers become founders. Each layer has a different core function.

• Developer: “how to build it” (implementation)
• Manager: “who builds what” (coordination, priorities)
• Founder: “why and what to build” (direction, value)

AI made the bottom layer nearly free. As implementation got easier, the real bottleneck emerged — not knowing what to build. The place where implementation used to block me was now asking: “why are you building this?”

One day someone asked me, “what’s the hardest thing about developing right now?” I was in the middle of working on two apps simultaneously and implementation wasn’t blocking me at all. But when I answered honestly, I surprised myself. “I have no ideas.” Already building, but not knowing what to build next — that was what felt most stuck. AI cleared the bottom layer, and the real bottleneck that had always been there came to the surface. I’ve written separately about how this shift changes the developer role overall.

What It Means to Become a Creator

It’s not about moving away from implementation. It’s about going up while keeping what you know about implementation.

Building the stock portfolio app, when Korea Investment API integration and complex profit/loss calculations got tangled up, full delegation to AI broke things. Financial calculations can’t be off by even a single won. So I defined the policies first. This formula for average cost, this handling for partial sales, this rule for sync conflicts. AI just translated those into code.

That moment clarified my role. What I was doing wasn’t implementation — it was policy design. And to design those policies well, I needed domain knowledge. The time I’d spent running accounting systems connected directly to building the stock app.

There was a period when I planned once a week specifically to use my Claude Design tokens before they reset — a forced planning session just to burn the weekly quota. Later, when that limit merged into a shared pool with Claude.ai and Claude Code, the routine disappeared. Looking back, it wasn’t really about saving tokens — it was a signal that planning was needed. An external mechanism was forcing me up to a higher layer.

A World Where What You Can Do Has Changed

It wasn’t about fun. What I could do had changed. It feels like a world where if you want to do something, you can.

Building with AI is fun. Building things directly is still satisfying. But on top of that, a larger feeling emerged — if I have an idea, nothing will stop me.

I tried going to a study cafe to study. It didn’t work. But once I started AI-assisted development on the stock app, location stopped mattering. I’d take my laptop anywhere — a cafe, wherever — and just open it up. Studying failed, but building worked. At first I figured I was just lazy. Now I read it differently. Someone who learns by making doesn’t level up by sitting down to study.

Closing

If you’re building something with AI, it might be worth checking in: what occupied most of your thinking this week — the code, or the product?

If it’s 70% code, you’re still in the implementation phase. If thoughts about the product, the user, and why this needs to exist are taking up more space — that’s not a bad sign. It means you’re leveling up.

And if that transition feels awkward, like you can’t focus — you’re probably just like me, building something while the real thoughts keep drifting in.

That’s not distraction.