What Building From Scratch Taught Me, and What AI Changed

In 2017 I built my first genuinely successful website, dnddiceroller.com. I know almost exactly how long it took me, because I was proud enough of the effort to count: about 80 hours.

Eighty hours of hand-writing HTML, wrestling CSS into shape, debugging JavaScript that rolled virtual dice, and testing the thing on every browser I could get my hands on. It was hard. It was also some of the most satisfying work I'd ever done.

Last week I rebuilt something of comparable scope in an afternoon. Same me, same skills, plus AI. That gap has been rattling around in my head ever since, so I want to write it down properly.

The long way, in 2017

By 2017 I'd been coding as a hobby for over 20 years. BASIC as a kid, then C++ and Java, and from 2013 onward the web. I wasn't a beginner. But dnddiceroller.com still cost me 80 hours, and here's roughly where they went:

• Figuring out the what. What should a dice roller even do? How many dice, what notation, what happens on the screen? Most of that was trial and error.
• The boilerplate tax. Page structure, layout, responsive breakpoints, the hundred little bits of scaffolding that every site needs and no site is interesting because of.
• The long tail of bugs. The roll that worked on desktop but not mobile. The layout that broke at exactly one width. The off-by-one that only showed up on a d100.
• Testing everything by hand, over and over, because there was no one else to catch what I missed.

None of that was wasted. I learned an enormous amount by doing every part of it myself. But if I'm honest, maybe ten of those eighty hours were the interesting part: the actual design decisions and the satisfying problem-solving. The other seventy were tax.

The short way, in 2026

I started working seriously with AI about two years ago. The thing that surprised me wasn't that it could write code; it was which hours it gave me back.

It didn't replace the ten interesting hours. Those are still mine: deciding what to build, how it should feel, which trade-offs are worth making. What it collapsed was the seventy hours of tax.

• Scaffolding a page is a sentence, not an afternoon.
• The boilerplate writes itself while I stay focused on the parts that matter.
• Whole categories of bug get caught in the diff before they ever reach a browser.
• I have a tireless collaborator to think out loud with at 1 a.m.

So the 80-hour build becomes a few hours. But here's the part that matters: only because I already did the 80 hours once. I know what good looks like. I know when the AI is confidently wrong. I know which corner it just cut, because I've been cut by that corner before.

What actually changed (and what didn't)

The tools changed. The judgment didn't.

AI didn't make the hard part easy. It made the easy part instant, which finally leaves room to spend real time on the hard part.

Building dnddiceroller.com the slow way is exactly what makes me effective with AI now. Every hour I spent hand-debugging CSS in 2017 is an hour of pattern-matching I get to reuse every time I read an AI-generated diff. The experience didn't become obsolete. It became leverage.

That's the honest answer to "doesn't AI make your skills worthless?" No. It makes taste and judgment the scarce things, and those only come from having done the work.

Why this turned into a profession

For most of those 30 years, coding was a hobby. Something I did because I loved it, on evenings and weekends, with no particular plan.

AI is a big part of why that finally changed. When the tax drops from seventy hours to a handful, the math on "should I build this real thing?" changes completely. Ideas that used to be too expensive to attempt are suddenly a weekend. So in 2025 I registered Iron Code Studios and started turning the hobby into a profession.

This blog is me documenting that transition in the open, including the parts I'm still figuring out.

The takeaway

If you're early in your own journey, don't read "AI does it in a few hours" as "so don't bother learning it the hard way." Read it as: the hard way is how you earn the right to go fast later. Do the 80 hours. Then let the tools give most of them back.

That's the trade I'd make again in a heartbeat, and thanks to AI, one I now get to make far more often.

More on the day-to-day of this in How I Actually Work With AI Coding Agents. And if you're curious how this very site was built, here's the write-up.