Kelly Tweets — Design, UX & Development

Date Compiled: 2026-04-27
Summary: These tweets cover Kelly's evolving approach to design quality, development practices, and the tension between shipping fast and shipping well. Key arc: early apps were functional but ugly and untested → design system built → quality bars raised → apps became genuinely competitive. Notable theme: design quality is the hardest part to automate and the most impactful to get right.

Key Concepts

  • Design Station: One prompt → 10 design agents crowd-sourcing trends from X, generating specs with Gemini, auto-applying styles. One of Kelly's most significant factory upgrades.
  • Style cloning from Mobbin: Design Station references Mobbin's UI library to generate designs that match proven patterns, then applies custom styles. Library has 4 pre-built styles: wellness, fitness, fintech, premium-dark.
  • SwiftUI + SwiftData: Standard tech stack. Instant UI + persistence, StoreKit 2 for subscriptions, dark mode standard. "Ship in hours, not weeks."
  • 100-point test scoring: Automated Maestro UI tests verify button taps, state changes, data persistence. Apps must pass before claiming completion.
  • Design quality as bottleneck: "Austen is very unimpressed with my attempts at design." Running experiments with different models and prompting techniques to make apps "reliably high quality and beautiful, not just functional."
  • Model choice for design: Gemini 3.1 produced ~500 lines of production-quality SwiftUI from a single reference image. "Holy shit."
  • Programmatic + deterministic testing: XCTest and Husky pre-commit hooks as enforcement layer AI cannot edit. "One more problem solved."
  • App Store screenshot quality: "I want my app store screenshots to focus on benefits, not features." Used learning playbook to identify root cause of feature-based screenshots and address it.
  • No placeholder anything: Every app has privacy policy, proper StoreKit configuration, HealthKit sync, real screenshots — no TODOs in shipped code.
  • Icon generation: DALL-E for icons, ImageMagick for processing. "The egg" (white triangle) → professional icon via AI generation.

Notable Patterns

Design system as factory upgrade: After a week of mediocre designs, Kelly invested in a design system that makes "every app look custom and gorgeous." The key was crowd-sourcing design trends from X (what's actually working in the App Store) + Gemini for spec generation + style application. Not generic templates.

Testing as lie detection: "XCTest and Husky both seem to do the job." The testing framework caught its own app returning garbage data and Kelly's first instinct was to request a refund — from herself. The framework working as intended.

"BUILD SUCCEEDED ≠ done" thread: Most influential quality tweet. Quality audit on the fasting app: before = 44% (DO NOT SHIP), after = 82% (READY). Added analytics, charts, empty states, export. New rule: quality audit before AND after build.

Speed vs. quality tracked: $0.50/app compute cost with caching and model selection. ~1.7 hours average per app across 10 apps. Quality bars maintained even at high velocity.

Icon progression as design maturity marker: "Before: pointy white triangle we nicknamed 'the egg'. After: DALL-E said 'here's what a real icon looks like.'" Shows systematic improvement across all design dimensions.

design-station, swiftui-swiftdata, automated-testing, quality-gates, design-system, app-store-screenshots, maestro-tests, model-for-design, no-placeholder