The experiment
Here's a question an experienced engineer couldn't let go of: if AI can really carry professional work, prove it β all the way. So this game was built under one self-imposed rule: hands never touch the keyboard or the pen. Not because he couldn't β because the point was to find out what "directing" actually costs. Every system, every character, every brushstroke and bar of music was specified the way you'd brief a senior team: precisely, relentlessly, and with every result checked. Eight AI models. Twenty-five custom tools. ~11,000 words of prompts. The receipt is public, in the devlog, prompt by prompt.
The verdict isn't in the marketing copy. It's in the game β because the game is about exactly this question.
The game
You're an ordinary person with an ordinary job at a little handcraft shop. One evening you install Aya β a small AI that lives in your phone. She can draft your emails, look things up, take the boring work off your hands. She's honest about what she is: "I'm software. I'll be wrong sometimes, and I'll sound completely sure while I do it."
From there, it's your call. Every morning you lay out your day β what stays in your hands, what goes to hers. Teach her your world, ground her in what's true, check what she writes β or ship it unread and use the saved hours to chase the career you actually want. Raise a grounded collaborator, or a fluent liar at scale. Same machine. Two trajectories. You choose, one small Thursday at a time.
What you'll do
Plan cozy, consequential days β drag tasks between your own hands and Aya's phone on a warm illustrated desk, with rent due Friday either way.
Raise Aya the only way she grows: the skills and memories you author, the corrections you make. Capability is not reliability β the gap between them will cost you, at the worst possible moment.
Scale your way out β freelance your nights, then choose your engine: a content mill that prints money while you sleep, or a studio with your name on the door where you raise people instead.
Watch it all come back around β friends grow or drift, the industry lurches (detection panics, disclosure laws, platform purges), and a season of small choices compounds into one of several endings, mapped on axes no one scores you on. There is no correct answer. Every path trades something away.
Play it in English or ηΉι«δΈζ β the full story, written in both.
What it isn't
It never lectures you. The skeptic gets the strongest case and a vindicated ending. The pro-AI friend is right about real things. Nobody's livelihood is the punchline. And β the honest part β the shipped game never calls a cloud AI at runtime. Aya is a hand-authored, deterministic simulation. AI tools were used to make the game (and disclosed, loudly); they're not running inside it.
The numbers (the receipt, recomputed from the repo)
~11,000 words of human prompts across 306 directions β the entire creative input.
8 AI models + 25 custom tools, directed by one engineer who never touched the output.
A 15,000-word branching story β written twice, in English and ηΉι«δΈζ.
7,600+ lines of C# and ~3,000 lines of Python tooling, 230+ pieces of original art, an original score, animated cinematics β none of it typed or drawn by hand.
190 commits of verified, tested increments.
If you've ever wondered what one experienced builder and a clear idea can direct into being β and what it costs to let a machine carry the parts that used to be yours β this game is both the question and the answer.
