Trading Wool

Trading Wool is a rapid experiment in building an autonomous AI trading agent using Alpaca. The goal was simple: give an AI agent the ability to analyze market conditions, make trading decisions, and manage leveraged positions, then see what happens when you actually let it operate. The first version answered that question pretty quickly. The agent went full degen, using available leverage far too aggressively and taking positions without enough consideration for risk. Rather than treating that as a dead end, I used the behavior as a feedback loop. I introduced guard rails for position sizing and risk exposure, changed how positions were managed, and added hedging mechanisms to make the strategy more controlled. I also integrated Grok for inference, allowing the agent to explain the reasoning behind its trading decisions instead of simply producing an unexplained action. Trading Wool is not intended to be a production-ready trading system or a claim that AI has solved algorithmic trading. It is a deliberately small experiment in agentic decision-making: give an autonomous system real constraints, observe where it fails, understand the failure, and engineer the next iteration. The result is a compact exploration of AI agents, automated trading, risk management, and the surprisingly important realization that sometimes your AI needs someone to tell it to stop YOLOing the leverage.