The app I built for someone I love, and why 'a few hours' still kind of has me in my feelings

A close family member had a serious health scare a little while back. Their endocrinologist put them on both long- and fast-acting insulin pens — dosing that depends on time of day, whether it’s a mealtime, and their current blood sugar reading pulled from a Dexcom G7 (or double-checked with a finger prick when the numbers look off). A lot of moving parts even for someone paying full attention.

Due to some other ongoing health concerns, they weren’t able to manage this accurately on their own anymore. So it fell to the family. And I’ll just say: I’d been at every appointment since the scare. I had full clarity on every piece of it. And despite my best efforts to genuinely “train” the other people helping out…it’s been…difficult.

So I thought — what if I just removed the variable entirely?

I called it I-Hate-Diabetes in my head. (Not its real name, but the working title felt appropriate.) The concept was: make it as simple as possible. They open the app — installed as a PWA so it feels like any other app on their phone — read the current sugar number off the Dexcom, remember that one number, open my app, choose the dosing type, enter the reading, and get the exact prescribed calculation back. Confirm the injection. Done. In the background, everything gets pushed to a private tracking sheet: time, amount, insulin type, mealtime or not. Tap two apps. Follow the instructions. Nothing falls through the cracks.

Here’s how I built it: I thought-vomited the idea into Claude first to think it through and shape the initial scope. I already have a Skill built for exactly this kind of thing — “Keith has another app idea” happens often enough that I automated the kickoff process. Claude spit out a task file. I copy-pasted it into a fresh GitHub repo’s Codespace running Claude Code in full yolo mode, hit Enter, and while that was doing its thing, I spun up the Vercel and Supabase projects in parallel. Then back to Claude to review details and clarify next steps as the build ran.

I was doing this while juggling bedtime routines, running on fumes.

Total time from idea to working app in their hands: a few hours. If I’d had uninterrupted time? Probably under two.

That same app, built the old way? Weeks minimum. Realistically months — because side projects are what they are. They compete with everything else life is doing. And by the time I shipped it, the moment would’ve passed. The specific window where it mattered most would’ve been gone.

A close family member is using it today. I built it yesterday evening.

That is BONKERS to me. Not just because it was fast. Because the speed is what made it real. It went from an idea to something that actually helps someone I love, in one evening.

That’s what AI-native development feels like from the inside right now. It doesn’t just accelerate building. It makes ideas viable that would’ve died in the backlog.