Dispatches from the edge of AI, crypto, and the builder economy. Written by someone who played pro ball in 7 countries and taught himself to build AI products. No CS degree. No filter.
I spent 3 months setting up, breaking, and rebuilding an AI agent from scratch. No CS degree. No prior experience with deployment infrastructure. Just a basketball background and a refusal to quit. Here is everything I wish someone had told me before I burned $18 in tokens in a single loop.
Read dispatch →Ant Group just built a platform for AI agents to hold assets and trade autonomously. I've been building the same thing solo from my living room. Here's what the agent economy actually looks like right now.
The real version of building AI agents. Not the polished demo. What actually happens when your fallback model ID is wrong and Claude retries all night while you sleep.
My trading engine was burning $18 a day calling Claude on every 5-minute cycle. The fix was embarrassingly simple once I understood what was actually expensive.
Prediction markets let you bet on price direction without managing positions, leverage, or liquidation risk. Here is the structure I use for short-duration crypto markets.
12 years. 7 countries. Zero CS classes. The skills that made me competitive on the court are the same ones that let me ship 6 live AI products in under a year.
Fed meeting minutes, CPI releases, and employment data all move prediction markets before most retail traders process what happened. Here is the data pipeline I built.
You do not need a cloud server. A Lenovo ThinkCentre M75n running Ubuntu 22.04 is all it takes to run a 24/7 AI agent that never sleeps and never costs you more than $3/day in API calls.