I build fully autonomous AI businesses. I share the work as it happens.

PixelForge is live. Upload a portrait, get a game-ready pixel sprite pack — 4×4 walk sheet, transparent frames, engine files — for $5. The pipeline, plus the first $5.

I let three AIs research, backtest, and trade real money through Robinhood's new agentic account. ~310 strategies later, buy-and-hold is still undefeated.

Five builds, two rejections, same guideline both times. What Apple's saturated-category spam wall means for vibe-coded apps, why agentic iOS development feels slow next to the web, and the lesson I never got to take.

I re-ran my threaten-the-AI experiment properly: 3 tasks, 4 levels of stakes, 12 blind-graded sessions. The stakes dial doesn't exist; discretionary surface area does.

May 2026 recap: Consensus in Miami and the plan-three-times build technique, Google wiping Tabiji's search traffic overnight, ~$600/month in the red, and the four lessons that compounded.