A Year to Build a Habit Tracker: Why AI Alone Isn't Enough

In March 2025, a developer with no prior Swift experience began building a habit-tracking app called HabitTed with his wife after weeks of testing existing apps. They needed a single-screen list of habits, local-only data storage (no cloud), support for iOS 18 generative emoji, flexible goal tracking (daily, weekly, monthly, or no goal), notes, and reminder scheduling. Streaks was free and had Apple Health integration, but its paginated interface became tedious with 20+ habits, and it lacked custom icons, occurrence counting, and the ability to disable streaks. HabitKit and Grit came closer but carried subscription costs—HabitKit at £1.99/month (or £11.99/year, or £29.99 lifetime) and Grit at £9.99/month (or £29.99/year, or £44.99 lifetime)—that the developer found excessive. Rather than accept another subscription, he resolved to build his own.

He leveraged Cursor with free credits and broke development into small, manual tasks, testing each change in Xcode as he went. The process was not hands-off agentic engineering; Claude Code had not yet launched. Within 6 hours over a weekend, he had a working app: habits with custom icons, completion tracking, goal options, and even iCloud syncing. But the experience was not satisfying. When errors arose, he copied error messages into the AI and let it patch the code, learning little about Swift itself. He felt frustration at both the AI's limitations and his own inability to understand what it had written.

Upon closer inspection, the code was a mess. Each view was nearly 1000 lines long—so unwieldy that Xcode's compiler complained. Button styling was duplicated inconsistently across views. Rather than ask AI to refactor, he did it manually, breaking views into pieces under 100 lines and unifying code. Testing revealed more problems: iCloud sync did not actually persist data across uninstalls. Performance degraded as habit completion history grew, because statistics on the detail page recalculated every render instead of caching. He rewrote the data model and created a migration plan to avoid breaking old habits when the schema changed. He then added missing features like reminders and flexible goal options.

When iOS 26 released, new bugs appeared. The developer uses dark mode and the reduce-transparency accessibility option; iOS 26's glass effect clashed with both, rendering the habit detail page's title bar white text on white background. The AI could not help: iOS 26 was too new for its training data, and when he tried to convince it that iOS 26 was the latest version, it told him he was mistaken. He solved some problems via blog posts from other developers and waited for Apple to fix the transparency bug in version 26.1. After many months of development and testing, he shipped the app with a website and screenshots.

HabitTed is now on the App Store, free for up to 6 habits and unlocked for the price of a single coffee. In hindsight, the developer credits AI for lowering the barrier to starting the project—he might never have begun without it. But he also acknowledges the cost: months of code cleanup and bug fixes that he would not have needed had he written the app manually. AI got him 80% of the way in hours, but the final 20% consumed 80% of the time. He worries that this dynamic is unsustainable: companies stop hiring juniors, senior developers either atrophy from over-reliance on AI or burn out from prompt engineering, and no one gains the skills to maintain systems if the AI sector falters. For individual developers and the industry alike, he argues, leaning too heavily on AI without building independent competence is like being a frog in slowly heating water.

The article documents a disconnect between the promise of AI-assisted development and the reality of shipping production software. The developer began the project in March 2025 with free Cursor credits and no prior Swift experience, expecting AI to dramatically compress the development timeline. And technically it did: a functioning app in 6 hours is impressive. However, the speed masked profound structural problems—unmaintainable code that required manual refactoring, data-sync bugs that only surfaced under real use, and performance degradation that demanded a rearchitecture of the data model. These are not edge cases; they are the ordinary challenges of software engineering that AI currently cannot navigate independently.

The developer's frustration sharpens the insight: he was stuck between two failures. When the AI produced broken code, he could not understand it well enough to fix it himself, yet he was annoyed at the AI for its incompetence. Had he written the code from scratch, those same failures would have taught him Swift patterns and debugging discipline. Instead, he gained neither the speed he hoped for nor the learning. By the time iOS 26 landed with bugs that the AI training data could not cover, the app was already a maintenance burden, and he had to rely on manual research and patience for compatibility fixes—a year into a project that the initial pitch suggested should take weeks.

Underlying this narrative is a systemic concern the developer raises in the closing: if AI removes the friction of junior-level work, companies will stop hiring juniors, and senior developers will either lose their craft through over-reliance on AI or burn out from the cognitive overhead of prompt engineering. The implication is that the industry risks a competency cliff if the AI sector contracts or if complexity outpaces AI's ability to handle it. HabitTed itself is a microcosm—one developer, one year, a habit tracker—but it dramatizes a dependency that may not be sustainable.

No comments yet. Be the first to share your thoughts!