I've been building every day since a design interview showed me a glimpse of what designers are experimenting with using the new tools.
In my free time, I’ve been inspired to push PRs nearly every day, shipping products that are fun to design end-to-end and that solve my own problems. I’ve been building a recipe app for nearly a year, and my tools and full-stack skills have slowly improved.
So why am I building a digital Yahtzee score tracker? Because as niche as it is, I can’t stand the options that are out there, and building has never been more of a no-brainer.
Ever since camping, my partner and I have been addicted to simple dice rolling. Trying to save money, we figured we could just buy a couple of dice and skip buying the box set. But that meant we’d have to draw our own scorecard, and the digital solutions looked like they were made in the 90s. Quickly, I thought this could be an easy side project, and that the end result would make playing so much more enjoyable.
Not only would it mean no more lugging around pen and paper, but I also wanted to change how players think about inputting scores. At least for me, I don’t think in terms of adding while playing Yahtzee. I think about the game in terms of dice. So I knew it needed a way to score a round by selecting dice, not doing math.
I was also inspired by CMYK Games, which proved to me that board games don’t always have to go for a hyper-sensational style.
I’d also been wanting to try Claude Code for some time. I’d heard about Opus and its generous usage rates.
Starting from my phone
All you need to start any project is the Claude Code iOS app and GitHub Mobile. While it’s not the most seamless experience, it was simple enough that I could start a repo and open the first PR instead of doomscrolling. I’d open the Vercel preview and check out changes on my phone, where the app would mostly be used. For a few days, I iterated like this in small moments of downtime, adding or tweaking a feature here and there.
Switching to agents (and moving fast)
Eventually, I needed to really look at the code and move faster to reach the polish I wanted. So I opened Conductor, an agent orchestration UI for Claude Code and Codex, and started iterating on multiple issues at once. And like many people, it changed how I think about how we can, and maybe should, use agents in parallel. For a small side project, I didn’t care to review code too hard or think about every edge case. I have no users besides myself and my partner, so I made branches, checked if they worked, and merged rapidly.
What “good enough” looks like for a tiny app
This ability to move superfast is great at the early stages and means I can get a super functional prototype, and for my purpose that's all I need. I don't need it to scale to thousands or be heavily efficient. All I needed was for it to work for me and my friends.