Riley Gerszewski
What can AI really do for product design? That was my first question and by the end I was blow away by what it can do for end to end product development. This case study demonstrates how as a designer I not only transformed the design workflow how I was also able to complete a sophisticated native iOS app. Some may call it vibe coding but that is too narrow. What about Vibe ML model training? Vibe API’ing? Vibe branch merge’ing? Also, is it technically vibe coding if you need to train AI on knowledge it does not have? Nope, it’s AI assisted product development.
Traditional product development moves slowly. As a designer, I've spent months designing and prototyping that approximate functionality. Even with RITE methodology, you're slowed by development bottlenecks. But as we know, the faster you can get it in the front of users - at any fidelity - the faster you can get the feedback required to confidently move forward.
AI assisted product development relies upon AI to know everything. Claude, ChatGPT, and the like can be thought to know everything - but to a certain date. And in this case the SOTA LLMs know nothing about iOS 26. I needed expert-level guidance on new Apple technology, but got "I don't have information about iOS 26" responses.
This case study will give you an overview on how I solved the problems. However, follow the links at the bottom of the page to articles I’ve written on Medium, for a deeper dive into the solution details.
I chose to build a women's diabetes management app that - recognizes how hormonal fluctuations impact blood sugar control. This gave me a complex problem requiring sophisticated health data integration, machine learning, and privacy-first architecture.
The app includes comprehensive health tracking, cycle-aware glucose predictions, and educational content that adapt to different menstrual phases. All built with on-device AI processing and HealthKit integration.
But the real project was testing AI-assisted development at every stage.
Traditional design tools create static representations. I needed something that could generate functional iOS code directly from design concepts.
My workflow became: UXPilot web app generates initial concepts → UXPilot Figma plugin refines designs → direct Swift code generation → feature integration into working app.
This eliminated the typical design-to-development handoff entirely. Instead of creating mockups that developers interpret, I was generating functional SwiftUI components that integrated directly with existing app architecture.
I developed what I call "RITE Coding" - applying Rapid Iterative Testing methodology directly to functional code development. When I had an idea for a new feature, I could build it completely in hours using AI assistance.
This wasn't about creating clickable prototypes. I was building fully functional features that leveraged existing HealthKit integrations, Core Data, and on-device machine learning. Features that processed real health data and provided actual insights.
The speed change was dramatic. Ideas that would traditionally take weeks to prototype could become functional features before breakfast was finished.
I created dozens of specialized AI assistants for different aspects of development:
Technical architect for data modeling and API integration
SwiftUI expert for interface implementation
Health domain expert for medical accuracy and regulatory considerations
Privacy engineer for security architecture
UX researcher for interaction patterns and user flows
Each assistant was trained with specific context about the project, allowing me to get expert-level guidance across disciplines I'd never worked in before.
Since no AI assistant knew about Foundation Models, I created comprehensive training datasets specifically designed to teach AI systems about iOS 26. This involved using Gemini's research capabilities to analyze all available documentation, then having Claude restructure it into AI-optimized training data.
The result: I had expert-level AI assistance on bleeding-edge Apple technology that didn't exist in any training dataset.
Built sophisticated machine learning features that run entirely on-device:
Pattern recognition for glucose-cycle correlations
Personalized prediction algorithms
Privacy-preserving data analysis
Integration with Apple's Foundation Models for recipe generation
Designed and implemented:
Core Data schemas for health tracking
HealthKit integration for automatic data collection
Secure keychain storage for sensitive information
Real-time data synchronization and offline capability
Rather than traditional metrics like "30% reduction in production time," this exploration into AI-assisted product development yielded three fundamental shifts in perspective:
This isn't about AI replacing designers - it's about AI enabling designers to operate across the entire product stack. The workflow I developed - from UXPilot concept generation through Figma refinement to Swift implementation - suggests a future where design thinking can be applied directly to functional products in real-time. The bottleneck shifts from implementation capability to design judgment and user insight.
Building every aspect of an app - from initial concept through data architecture, ML implementation, API integration, and native iOS features - provided visceral understanding of how design decisions cascade through the entire development process. Design isn't just about interfaces; it's about understanding technical constraints, data flows, and implementation realities. This hands-on experience transforms how I approach design problems and collaborate with engineering teams.
By writing Swift code, implementing Core Data schemas, and debugging API integrations, the traditional handoff model becomes obsolete. Instead of "throwing designs over the wall," I can now participate directly in technical discussions, prototype with actual functionality, and contribute meaningfully to architectural decisions. This creates genuine partnership rather than transactional collaboration.
The project demonstrates that AI doesn't just make designers faster; it makes them more complete product builders. When the cost of iteration approaches zero and the time from concept to functional feature collapses to hours, the entire paradigm of product development shifts.
This wasn't about building the perfect diabetes app. It was about discovering what becomes possible when AI removes traditional constraints from the design process. The result is a new model for how designers can contribute to product development - not just through interfaces and user research, but through direct technical contribution across the entire stack.
