I had never opened Xcode. A week later, I was using the app I built.

import WidgetKit
import SwiftUI

/// This was ten. Six of them restated
/// "what's on today" in slightly different
/// shapes … what is left matches what
/// the app is actually for.
@main
struct SecondMindWidgetBundle: WidgetBundle {
    var body: some Widget {
        TodayWidget()
        StreakWidget()
        TodayLockWidget()
        if #available(iOS 18.0, *) {
            CompleteNextControl()
            AddReminderControl()
        }
    }
}

I designed it, then I shipped it

Second Mind is a native iOS app written in Swift: a habit, routine, and reminder companion I designed and built for exactly one user: me. It runs on my phone, on-device, with Apple Health and location wired in.

The starting point was genuinely zero. I asked what an iPhone app is even written in, the answer was Swift, and I installed Xcode that day. Inside a week the first version was on my phone and I was using it. What you see below is the third version. The first two are further down the page, unretouched.

It runs on my phone, not in a mockup

That’s a real iOS home screen. The widget is reading live data straight out of the app’s own store: next habit, current streak, how much of today is left.

SwiftUI

Every screen written as code. No template, no page builder.

SwiftData

Nine on-device models: habits, routines, reminders, memories.

HealthKit

Steps, sleep and workouts read as triggers.

CoreLocation · MapKit

Geofenced places, so a routine can fire when I get home.

WidgetKit

Three home-screen widgets, fed by the app's own store.

Cloudflare Worker

Ask Mind's backend: a sentence in, typed JSON out.

178 Swift files across 61 commits, every one of them mine to review.

I used AI as an implementation partner while writing the Swift, but the product decisions, the design system, and the judgment calls below are mine, and I reviewed every change that went in.

The problem

My problem was never motivation. It was capture. The small, recurring things (a supplement, a birthday, a stretch) died between a notes app, a reminders app, and a habit tracker before I ever wrote them down. And the few reminders I did set were blind to context: they fired on a clock, never when I actually arrived somewhere or finished a workout.

I didn't need another productivity app. I needed a second mind: one that captures a thought in a sentence, remembers it, and brings it back at the exact moment it matters.

How my judgment changed

The interesting part of this project isn't the final screens. It's the distance between them. I rebuilt the whole app three times. Same person, same tools; what changed was judgment about hierarchy, visual weight, and when the app should act on its own.

The MVP leaned on defaults and empty states. The dark rebuild gave me a real system but fought the content for attention. The current version is quieter on purpose: a clear morning hierarchy, health context in view, and the AI folded into routines I already had instead of a separate "chat" tab.

The decision that shaped the app

Every capture flow lives or dies on friction. The core call was how Ask Mind should behave when I type or say something like "remind me to take vitamins at 8 AM."

One vocabulary, every screen

Because I was designing and shipping alone, drift was the real risk. So the app runs on one token vocabulary, named the same way in Figma and in code: primitives hold the raw palette, semantic tokens describe intent, and screens are only allowed to use the semantic layer. New screens can't quietly diverge from old ones.

The honest outcome

This was built for one user, so I won't dress it up as market validation. The evidence I can actually stand behind is simpler: it's still the first thing I open every morning, and it survived being rebuilt three times because each version was easier to live with than the last.