>>555617912

>>555619837
I'm not a developer. I'm a 38-year-old accountant from Norway who had never written code before 2024. Six months and 500+ hours later, I shipped Curve Clash - a real-time multiplayer mobile game that hit #4 in Family Games in my country.

I'm sharing this because I think there are useful lessons here, even for experienced devs.

The tools I used:

Claude & ChatGPT: For code generation, debugging, and architecture decisions

Cursor: As my IDE with AI integration

Flutter + Firebase: The actual tech stack

What worked:

"Vibe coding" is real - I could describe what I wanted ("make the curves leave a trail that other players can collide with") and iterate until it worked. The AI understood game logic surprisingly well.

Understanding > Writing - I can't write a Flutter function from scratch. But I learned to read code, understand why it breaks, and describe fixes. That's enough to ship.

Architecture matters more than syntax - The hardest bugs weren't code problems, they were architecture problems. AI helped me refactor toward better patterns once I understood what was wrong.

Real-time multiplayer is hard - Firebase works, but latency compensation, state sync, and edge cases took 40% of my time. If I did it again, I might use a dedicated multiplayer backend.

What I'd do differently:

Test on more devices earlier - Performance issues on older Android phones caught me late

Build analytics first - I added proper event tracking too late

Scope smaller - I built 4 game modes when 1 polished mode would have been better for launch

The game: Curve Clash is a multiplayer snake game inspired by "Achtung die Kurve" (90s LAN party classic). Free with a $3 Pro upgrade.

Not here to promote - genuinely curious if other devs have tried AI-assisted development and what your experience was like.

The project: An AI companion game for Android - virtual relationship with emotional dynamics, memory systems, and monetization. Built solo with Claude Code as my "junior dev."

The results:

Playable demo: 1 month

Public release: 2 months

100k downloads: 4 months

Here's what nobody tells you: AI can't make a production-ready game.

I tried letting AI lead. The codebase became unmaintainable in 2 weeks. Here's why:

The 3 pain points nobody warns you about:

1. AI sees the small picture, not the big picture.

This was my biggest problem. AI solves the immediate task in front of it. It doesn't understand how that task fits into your overall architecture.

I'd ask Claude to implement a save feature. It would create a perfect, working save system - as a completely new architecture that didn't blend with anything else in the project.

Week 3, I had:

SaveManager.cs (singleton pattern)

PersistenceService.cs (dependency injection style)

DataStorageHelper.cs (static utility class)

PlayerPrefsWrapper.cs (wrapper pattern)

Four systems doing the same thing. Four different architectural patterns. None of them talked to my existing GameServices registry. Logic spread across the entire codebase instead of one centralized API.

The AI didn't know I had a Service/System pattern. It didn't know I centralize all game services through GameServices.Instance. It didn't know my philosophy is "one button does everything" - single entry points
that handle full flows internally.

So every feature became its own island. Its own patterns. Its own way of doing things.

A human junior dev would look at existing code and think "how does this project do things?" AI just builds what you ask for - in isolation.

2. AI over-engineers everything.

Ask for a simple toggle and you get an abstract factory pattern with dependency injection and three interfaces.

File: 1613945311852.jpg (1.37 MB, 2168x2002)

1.37 MB JPG

>>555620221
>>555620830