I Read a Book About Stock ETFs and Decided to Build a Stock App Instead

I knew nothing about stocks but wanted to build it — after a year-long pause, finishing in 2 weeks

I knew absolutely nothing about stocks.

About a year ago, I read a book called Start Investing with Stock ETFs. Most people finish a book like that and think, “Maybe I should buy an ETF” — then download a brokerage app. But I went somewhere a bit unusual. Not “I want to invest in stocks” — I wanted to build a stock app. Without knowing a thing about stocks.

Looking back, that was the starting point of my AI development journey.

I didn’t do it alone. A group of 7–8 coworkers started it as a study project together. Since it was my idea, I naturally ended up leading the planning and pulling the development along.

My main role back then was “explaining things.” I organized backend guides chapter by chapter (intro, project structure, shared modules, models, repositories…), explained what ORM meant with SQLAlchemy and SQLModel, and established a DDD structure with app/domain/modules-style domain separation as the team convention. More than writing code together, I was laying the foundation so we could write code together.

The way I worked back then was very different from now. I used Gemini CLI — Google’s free CLI tool. I’d train it by writing conventions into GEMINI.md, manually pass context with @file, then give instructions one tick at a time → review → reorient. If I didn’t like it, I’d have it redo things until I did. Back-and-forth. Very hands-on.

And then it just fizzled out.

It didn’t fall apart dramatically. Everyone had their day jobs, and keeping up side work alongside full-time work wasn’t easy. That’s how it naturally stopped.

I picked it back up in May this year.

I started in earnest again, but my old design tripped me up at first. I’d sketched it as a dashboard showing multiple accounts at a glance — too deep, too heavy. So I scrapped it. Instead of a multi-account overview panel, I centered it on “one account you’re currently managing”, with the home screen as that account’s portfolio hub. Sidebars, web tables, the admin-style dashboard — all stripped away, redesigned as a mobile app flow.

(The principle I settled on is still there in the spec doc — “Prioritize the flow of selecting one account to manage now, over showing multiple accounts at once.”)

But the real shock came from something else.

After months away, I picked the tools back up — and it was a completely different world. The quality of AI coding tools had leveled up entirely in the meantime. What used to require tedious back-and-forth to train now handled much bigger chunks on its own.

I’m not someone who researches before jumping in. I’m the type who just goes for it — I dove in without knowing what was best or even how to use it properly. And I finished in almost 2 weeks. What had been dragging along for a year — I picked it back up and wrapped it in 2 weeks.

I still don’t know much about stocks.

But I started with nothing but the desire to build, and the tools happened to grow ahead of me, which let me see it through to the end. The motivation for starting doesn’t have to be “I already know a lot.” If you want to build it, build it.

What I actually built — running agents in parallel with worktrees, experimenting with multi-agent task splits, running into the KIS API — that’s in the next post. And the story of getting blocked at launch by a single line in the terms of service, after finishing everything, is there too.