I’ve been curious about Google AI Studio for a while. Not the “generate a haiku” kind of curious — more the “could I actually ship something with this?” kind. The problem with testing any AI coding tool on a todo app or a landing page is that you learn nothing useful. Those are solved problems. I wanted to give it something with real logic: a multiplayer game where players race each other to find numbers in ascending order, with multiple game modes, real-time competition, and enough moving parts that a half-baked AI output would be obvious immediately. As a dev with a newborn at home, I don’t have weekends to burn on experiments that go nowhere. If Google AI Studio could handle this in a day, that would mean something. If it couldn’t, at least I’d know.

So I described the game. Not pseudocode, not a technical spec — just a description of what I wanted to exist. A grid of scattered numbers. Players racing to tap them in order. A classic mode for pure speed, a time trial mode that adds pressure, and two modes for the mathematically inclined: one where you hunt for prime numbers and another that follows the Fibonacci sequence. Multiplayer, browser-based, no downloads. I hit enter and waited.

What came back wasn’t a sketch or a partial scaffold. It was a working Next.js application with game logic, a dark-themed UI, and all four modes functioning. The classic mode tracked ascending order correctly. The time trial mode had a countdown. The prime number mode actually validated primes — it wasn’t just filtering a hardcoded list, it understood the mathematical constraint. The Fibonacci mode knew that after 1, 1, 2, 3, 5, the next number had to be 8, not 6. These aren’t trivial implementation details. Each mode needs different validation logic, different win conditions, different UI feedback. Google AI Studio handled all of it from a single natural language description. That’s vibe coding in the truest sense — you describe the vibe of what you want and the tool figures out how to make it real.

I’ll be honest: I didn’t expect it to work this well. I’ve used AI coding assistants that produce impressive-looking output that falls apart the moment you interact with it. This wasn’t that. The game was playable. The modes worked. The multiplayer mechanics were in place. It was genuinely the kind of output that, two years ago, would have taken me a few focused evenings to produce myself.

But “working” and “ready to share with people” are different things, and the gap between them is where the craft lives. AI Studio got the game to about 90% — the architecture was sound, the logic was correct, the UI was functional. The remaining 10% was the kind of polish that makes something feel finished rather than generated: edge cases in the scoring, visual consistency across modes, small interaction bugs that only surface when you actually play for a few minutes. This is where I brought in Claude Code. Not to rebuild anything, not to rearchitect — just to do what it does well: targeted fixes, refinement, the developer equivalent of sanding rough edges. AI Studio did the heavy creative lifting. Claude Code handled the craftsman’s finishing touches. Two AI tools with different strengths, used at the right moments, and the result was a complete game built in a single day.

What’s interesting about this workflow isn’t the individual tools — it’s that knowing when to switch between them is becoming a skill in itself. AI Studio excels at going from zero to something substantial. Claude Code excels at taking something substantial and making it solid. The overlap is smaller than you’d think, and the combination is more powerful than either one alone.

So here’s what that day produced: Find The Numbers, a multiplayer number puzzle you can play right now in your browser. You and your friends get the same grid of numbers and race to find them in ascending order. Classic mode is pure speed — tap 1, 2, 3, as fast as you can. Time Trial adds a countdown to sharpen the pressure. Prime mode challenges you to pick out only the prime numbers in sequence, and Fibonacci mode follows that famous spiral: 1, 1, 2, 3, 5, 8, 13. It’s free, it works on any device, and there’s something satisfying about a game that’s simple enough to understand in five seconds but competitive enough to keep you coming back.

The broader reflection is that vibe coding isn’t just a buzzword anymore — it’s an actual development workflow with real output. The gap between “AI-generated prototype” and “thing people actually use” has gotten remarkably small, and it keeps shrinking. A year ago I would have scoped a multiplayer game as a multi-weekend project. Today it’s a day. That compression isn’t just about speed; for someone juggling a dev career, side projects, and a seven-week-old who hasn’t figured out that 3 AM isn’t playtime, it fundamentally changes what’s possible to build. I sat down with an idea and Google AI Studio, spent a day in that flow state where the tool keeps up with your thinking, polished the edges with Claude Code, and ended up with a game that’s live and playable. The experiment worked better than I expected, and I have the game to prove it.