AI instead of a team: Flutter apps in 2 weeks

Business Analyst Builds Flutter Apps with AI: A 2-Week No-Code Experiment

A business analyst with zero programming experience launched two mobile apps on RuStore in just two weeks—using only AI assistants. Tech stack: Flutter + Supabase. Total effort: 80 hours. Downloads: 14. This experiment revealed how requirements-writing skills translate directly into effective code-generation prompts—and exposed the real limits of AI in complex development.

Comparing AI Models for Code Generation

Free-tier models produce non-compilable code with outdated UIs. GPT-5.2 improves compilation and design—but bugs persist. Claude Opus leads in code cleanliness, context retention, and modern UI implementation.

| Criterion | Free AI Models | GPT-5.2 | Claude Opus |

|------------------------|--------------------|--------------------|--------------------|

| Code Quality | Fails to compile | Functional, buggy | Clean, stable |

| UI/Design | 2010-era look | Acceptable | Modern, polished |

| Context Understanding | Loses track quickly| Medium | Best-in-class |

| Complex Integrations | Fails entirely | Works with hints | Partially successful |

| Cost | ₽0 | ~₽700/month | ~₽2,000/month |

Subscription costs were fully covered via corporate access.

Why Business Analysts Excel with AI

User stories, acceptance criteria, and business logic map naturally to precise AI prompts. Example prompt for a time-tracking dashboard:

Build the home screen for a Flutter time-tracking app. Requirements: 1) Top section: circular chart showing weekly 168-hour distribution across categories (sleep, work, sport, learning, leisure, other). 2) Below the chart: tap-to-edit category list with hour input per item. 3) Bottom: “Save Week” button. 4) Dark theme with accent color #4FC3F7. 5) Use fl_chart package for visualization. 6) State management via Provider.

A well-structured prompt cuts iterations dramatically—just like briefing a junior developer.

Projects: From Idea to App Store

Time Tracker “168 Hours”

Distributes your 168 weekly hours across life categories (sleep, work, sport, etc.) and compares planned vs. actual usage. Inspired by Laura Vanderkam’s book 168 Hours.

F1 Tycoon Manager

An F1 team simulation: hire drivers, develop your car, manage budget, and run race simulations. Designed to stress-test AI on game logic and state complexity.

Tech Stack:

• Frontend: Flutter + Dart
• Backend: Supabase (free tier for MVP)
• UI: AI-generated
• State Management: Provider

Stack chosen based on AI recommendations: cross-platform support, beginner-friendliness, and strong presence in training data.

Development Timeline

• Days 1–2: Installed Flutter & Android Studio. Built first animated screen—immediate “wow factor.”
• Days 3–5: Developed core time-tracker screens. Iteration loop: prompt → code → test → bug fix.
• Days 5–7: Integrated Supabase. Hit a crisis: cascading auth errors (email validation), network request failures.
• Days 8–12: Built F1 Tycoon. Struggled with race logic, balance tuning, and persistent state.
• Days 13–14: Generated AAB, signed APK, published to RuStore (free—no $25 Google Play fee).

Total time: 80 evening hours.

Key Problems & Workarounds

Context Loss

AI forgets architecture in larger projects (>50 files). Solution: a “memory file”—a plain-text doc listing folder structure, data models, and key decisions. To generate a clean file list:

find lib -type f -name "*.dart" | sort

Including this memory file in prompts reduced confusion by ~70%.

Cascading Bugs

Fixing one file breaks dependencies. Solution: always analyze all related files together—not in isolation.

Rate Limits

Claude Opus blocked for 4 hours mid-debugging session.

Free Model UI Output

Grey, dated interfaces—until upgrading to paid tiers.

Honest Metrics

• Total cost: ₽0
• “168 Hours” downloads: 1
• “F1 Tycoon” downloads: 13
• Monetization: interstitial banners + rewarded video bonuses
• Primary goal: prove viable app launch without coding experience

Where AI Shines—and Where It Stumbles

Strengths:

• UI components
• Stack selection
• Rapid prototyping (screen in <5 minutes)
• Error explanation
• Boilerplate generation

Weaknesses:

• Backend integrations (e.g., Supabase auth flow)
• Context handling in large-scale projects
• Cascading bug resolution
• Performance optimization
• Architectural decision-making

Key Takeaways

• BA skills are ideal for prompting: specs = ready-made prompts.
• AI lowers the entry barrier—but doesn’t replace senior developers (80 hrs vs. 15–20 hrs for same output).
• A memory file is mandatory for projects >20 files.
• Supabase integration demands multiple iterations due to subtle networking edge cases.
• RuStore offers a truly free, low-friction path to Android MVP distribution.

— Editorial Team