TL;DR: I built a complete iOS app with AI-assisted development, without being a professional software developer. The key insight: AI is a multiplier for experience, not a replacement for it. My 15 years of IT background were the reason it worked. Estimated time savings compared to traditional development: 60-70%.
When I had the idea for Iceland Explorer, a travel app for Iceland, I faced a fundamental question: Build it traditionally with a team and a six-figure budget? Or try a new approach?
I chose AI-assisted development. Here’s what I learned.
The Starting Point
My qualification: 15+ years of IT experience, primarily in leadership and project management. Programming wasn’t my daily work. Yet I wanted to build the app myself, from concept to App Store.
The question wasn’t “Can AI build an app?” but rather: “Can AI enable an experienced IT leader to build an app themselves?”
The answer: Yes. But differently than expected.
Traditional vs. AI-Assisted Development
| Aspect | Traditional (Freelancer/Agency) | AI-Assisted (Self-built) |
|---|---|---|
| Cost | CHF 80,000 - 150,000 | CHF 500 - 2,000 (tools) |
| Timeline | 6-12 months | 4-6 months |
| Control | Dependent on supplier | Full ownership |
| Learning effect | Low | Very high |
| Risk | Medium-high | Manageable |
The AI Tools I Used
For Iceland Explorer, I relied on the following tools:
- Claude (Anthropic): Primary coding partner, architecture decisions, debugging
- ChatGPT (OpenAI): Research, content writing, translations
- GitHub Copilot: Inline code completion directly in Xcode
- Midjourney: Visual concepts and UI inspiration
- Cursor: AI-optimized code editor with deep context understanding
What AI Does Well
Prototyping and Iteration
The biggest game-changer was prototyping speed. Instead of spending days on a feature, I could create a working prototype in hours, test it, and iterate.
- UI components: From description to working code in minutes
- Data models: Complex structures (routes, POIs, weather data) designed quickly
- Bug fixing: AI as a pair-programming partner that never gets tired
Concrete example: The offline map feature, one of the app’s most complex features, was initially built without persistence. Migrating the existing code to a Core Data model with offline capability would have taken 2-3 days the traditional way. With Claude as a sparring partner: 6 hours.
Knowledge Transfer
As an iOS development newcomer, I constantly needed guidance. AI served as a patient mentor:
- Swift/SwiftUI syntax explained in the context of my code
- Best practices for app architecture suggested (MVVM, Dependency Injection)
- Apple guidelines interpreted and implemented (Human Interface Guidelines, App Store Review Guidelines)
Content Creation
Iceland Explorer lives on content: descriptions of attractions, route suggestions, practical tips. AI helped with:
- Researching and summarizing information for over 200 points of interest
- Translating into multiple languages (the app is available in German, English, and Icelandic)
- Structuring content for optimal user experience
What AI Cannot Do
Product Vision
No AI in the world knows what Iceland travelers actually need. That came from my own experience, user interviews, and intensive research. The vision remains human work.
Before starting, I conducted 23 interviews with Iceland travelers. I discovered that the biggest pain point wasn’t missing information (there’s plenty of that online), but the inability to access content offline and in poor mobile coverage areas. This insight shaped the product fundamentally. No AI would have told me that.
Design Decisions
AI can write code, but whether an interaction “feels right,” whether the information architecture is intuitive, whether the visual design evokes emotions: that requires human judgment.
Quality Assurance
“Works” and “works reliably” are two different things. Every line of code had to be understood, tested, and validated. AI-generated code is not automatically good code.
A cautionary example: Claude generated a route optimization algorithm that was correct in 95% of cases. In 5% of cases, it suggested routes through inaccessible terrain. Without my own knowledge of Iceland and extensive testing, I would not have caught this.
The Development Process at a Glance
Phase 1: Conception (4 weeks)
User interviews, feature definition, wireframes
AI contribution: 20% (research, summarization)
Phase 2: Architecture (2 weeks)
Technology stack, data model, project structure
AI contribution: 60% (architecture decisions, boilerplate)
Phase 3: Core Features (8 weeks)
Map, POIs, routes, offline functionality
AI contribution: 70% (code generation, debugging)
Phase 4: Content (4 weeks)
Texts, translations, images, database
AI contribution: 50% (content creation, translation)
Phase 5: Polishing (4 weeks)
UX refinements, performance, edge cases
AI contribution: 30% (targeted optimizations)
Phase 6: App Store (2 weeks)
Screenshots, descriptions, review process
AI contribution: 40% (copy, keyword optimization)My Key Learnings
-
AI is a multiplier, not a replacement. It makes good developers faster, but it doesn’t make anyone a developer who doesn’t understand the fundamentals.
-
Domain knowledge remains critical. My 15 years of IT experience were the reason I could use AI effectively. I knew which questions to ask.
-
The 80/20 effect. AI got me to 80% quickly. The last 20% (polishing, edge cases, performance) required disproportionately more human effort.
-
Iteration beats perfection. With AI, I could iterate faster and incorporate user feedback more frequently. The end product was better because I could run more cycles.
-
Prompting is a skill. The more precise and context-rich my requests to the AI became, the better the results. The first month was inefficient because I was still learning how to use AI properly.
What This Means for SMEs
You don’t need to build an app to benefit from these insights. The principles transfer:
- Internal tooling: Simple automations and dashboards can be created quickly with AI assistance
- Prototyping: New business ideas can be validated faster and cheaper
- Knowledge work: AI as a sparring partner for strategy, analysis, and conception
The barrier to entry for digitization projects has never been lower. A Swiss logistics company used this approach to build an internal dashboard visualizing delivery times and driver utilization in 6 weeks. Cost: around CHF 3,000 (primarily my consulting time). The traditional route would have cost CHF 25,000-40,000.
Want to explore what AI-assisted development could look like for your organization? Let’s talk.