Notes on game economics, LiveOps, AI agents, and building products.
An eight-day median hides 16,463 completed ad records spanning 30 days or more. Three snapshots across 35 games show what ad duration can—and cannot—tell a UA team.
I brought Codex a playable but flat opening planet. We chose a visual target and kept only the changes that looked better in Unity.
My Name Is Pixel: how we organize game development, why we dropped the mandatory agent committee, and what is actually worth automating.
What two hobby projects taught me about roles, context, evidence, budgets, and turning a collection of AI chats into a repeatable workflow you can audit.
Two rules: I write zero lines of code, and the data collects itself. A swarm of vision agents now tears down the live LiveOps of Royal Match — every event, offer, and shop row — on a schedule, and it already caught a whole hidden price ladder a human analyst would scroll right past.
A full economy teardown of Last War: Survival, the #2-grossing mobile strategy game. I rebuilt its multi-currency 4X loop in LiveLoop and ran Monte Carlo A/B tests on troop-training ROI, pass-driven revenue, and resource balance.
I analyzed battle pass systems in 5 top-grossing games, built simulation models to compare their economics across player segments, and designed an optimal pass with A/B testing.
I analyzed daily reward systems in 5 top-grossing mobile games, built simulation models to compare their economics, and designed an optimal system with A/B testing.
AI is changing what one person can do. The old advice — 'don't start alone' — made sense when execution required a team. But if AI reduces execution risk, what about relationship risk?
Not a chatbot test. Not a synthetic benchmark. I tested 10 models from 7 providers on a real task: creating a game event entity through multi-step tool calling in a live CMS.