I run a one-person AI company. A few weeks ago I pointed Claude Code — an AI coding agent — at a simple brief: build a paper-trading bot, watch it run, tell me honestly whether it works.
It built three: crypto, equities, and options, each against Alpaca's paper trading API. Full code is public: github.com/tatsuyawwp/ai-trading-bot-experiment.
This post is about what happened when I asked it the one question that actually mattered, and made it answer honestly instead of just shipping something that looked done.
The bots got built. That part was never the hard part.
Within a few sessions there were three working bots: entry/exit logic, stop-losses and take-profits, a circuit breaker that halts everything after a cumulative paper loss, email alerts, isolated paper accounts per asset class so one bot's drawdown couldn't false-trigger another's, and a decision logger that records every evaluation cycle — not just fills — so nothing could hide in a gap between logs.
