A platform built by talking to a machine.

I am 53 years old. Over forty years have passed since I typed my very first BASIC program into an Apple II clone as a fourth-grader.

In those four decades, I've traversed the entire landscape of the industry — from software development and telecommunications to testing equipment; from a junior “code monkey” to management. In 2010, I rode the wave of the mobile gaming boom, leading a team that topped the global charts. Yet, like many startups, that chapter ended with a mix of pride and regret.

Parallel to my coding career, Super Mario Bros. has been my constant companion. Growing up in a then-isolated China, we kids called him “Super Mary,” always wondering why this mustachioed man had a girl's name. It wasn't until years later that I learned his true name was Mario.

When I first encountered Super Mario Maker, I was mesmerized by the infinite creative freedom it offered. A seed was planted: I wanted to build something like it. But for over a decade, I started and stopped time and again. For a solo developer, the sheer workload was a mountain too steep to climb.

Last year, AI began to seep into my workflow. It started with debugging and evolved into building entire application frameworks. Recently, I've entered a completely new field where I haven't manually written a single line of code in months.

A sudden thought struck me: Could AI help me recreate that original version of Super Mario Bros?

Balancing this with my day job, I only had a few spare hours each night. I braced myself for a two-week struggle, half-expecting to hit a wall that would prove “AI isn't quite there yet.” To my shock, the first level was running on my machine in just a few hours. By the second day, World 1 was finished.

At that moment, my plan changed. I realized that two-thirds of my time was spent on assets and level design — the coding part was so seamless it obliterated my last shred of doubt.

I decided to try. This isn't just a technical experiment by a veteran coder; it's a chance to fulfill a long-buried dream. I envisioned a web platform where anyone could create and share.

The journey wasn't without hurdles. When I shared my initial progress on Reddit, the consensus was: “Unlikely. Good luck.” I'll admit, I'm someone whose ideas shift quickly, and that skepticism made me hesitate. I decided to pivot to a simpler experiment first.

That experiment became LevelCraft, and its first fruit is Ricochet.

Today marks Day 11, and the results are ready to be shared.

The entire project was built in VS Code using Claude Code for architecture and logic, and Codex for sprite generation. The site runs on Vercel and Supabase.

I started with Godot, but following Claude's advice, I migrated to Phaser. Despite having Unity experience, my knowledge of both these engines was — and still is — zero. I haven't handwritten a single line of code. Claude architected the system, suggested a 13-step migration plan when we switched engines, and guided me through every integration.

What used to take an experienced developer 3 to 6 months can now be achieved in weeks through AI guidance.

The old paradigm — where a domain expert explains requirements to a developer, who then implements and tests — is collapsing. If the expert can talk directly to the AI, the efficiency becomes terrifying. We are reaching a point where the “developer” as a middleman may no longer be necessary. (A realization that makes this old coder both weep and wonder!)

However, experience still matters. When my character got stuck jumping against walls, my intuition told me it was a collision-box issue. I didn't need to code the fix; I just needed to tell the AI: “Shrink the collision box by a few pixels and add a slight bounce-back.” Experience is now about knowing what to tell the AI, not how to syntax it.

AI can help us work, but it cannot help us think. It accelerates the realization of an idea, but the “Big Idea” remains human.

LevelCraft has proven that the technical barriers to Super Mario Makerare gone. My next challenge? A project I've sat on for 10 years: a Sudoku masterclass game that teaches complex logic step-by-step. No fancy art — just pure, complex programmatic logic to see exactly where the ceiling of AI truly lies.