AI agent builds paper-trading bots: honest backtesting reveals zero retail edge vs HFT
Original: I had an AI agent build 3 trading bots. It was losing to HFT before it even started.
Short summary
A solo founder used Claude Code to build three paper-trading bots (crypto, equities, options) against Alpaca's API, then asked the agent to honestly assess whether a retail bot can compete with HFT. The answer: no — colocated HFT is 10,000–100,000x faster and round-trip costs of 0.65–1.08% per trade eat any edge. Five backtested strategies for the crypto bot all failed a profit-factor > 1.2 gate, with the best (4-hour EMA crossover) reaching 1.23 but failing on out-of-sample data. The real lesson is forcing AI agents to prove viability rather than shipping something that looks done.
- •Claude Code built three working paper-trading bots with bug-fixing, circuit breakers, and decision logging
- •Honest assessment: retail bots have zero edge vs HFT on speed — 10,000–100,000x latency gap and 0.65–1.08% round-trip costs
- •Five backtested crypto strategies all failed a profit-factor > 1.2 gate, proving the value of a rigorous reject-or-adopt framework
Generated with AI, which can make mistakes.
Is this a good recommendation for you?
