AI For Dev

summary-cole-medin-openclaw-trading

4 min read

I Gave OpenClaw $10,000 to Trade Stocks

Channel: Cole Medin (+ Samin) Format: YouTube video URL: https://www.youtube.com/watch?v=eu8UJtuIi-E Length: ~19 minutes Published: 2026-04-09 Sponsor: none disclosed

Summary

30-day real-money AI trading experiment: Nate Herk and Samin each gave an OpenClaw bot 9,980 (−0.2%), dramatically outperforming the S&P 500 (−8.5%). Samin’s riskier Pareto strategy ended at $9,624 (−3.8%), also beating the S&P. The methodology — cron-driven autonomous execution, sub-agent team delegation, adaptive strategy without human intervention — is the key extractable pattern.

Key Points — Methodology

  • Brokerage: Alpaca API — supports equities + options, API-accessible, pattern day trading limits apply (blocked excessive trades for Samin).
  • Nate’s strategy (simple prompt): told OpenClaw to “spin up a team of wealth advisers” with no predefined strategy. The bot’s sub-agents collaboratively designed: 60–70% momentum swing trades, 15–25% options, 10%+ always in cash. Max 20% per stock, max $1K per options trade.
  • Samin’s strategy (domain-trained): 5 years at JP Morgan; trained bot on specific investment signal sources (hedge-fund-level research signals). Pareto principle approach — buy many positions, expect 80% to lose, 20% to win big. Bot ran two parallel instances monitored via Discord.
  • Execution loop: cron job every 30 minutes during trading hours. Each cycle: scan signals → rebalance portfolio → execute trades → report via Telegram (Nate) / Discord (Samin).
  • Adaptive behavior: bots independently adjusted strategies mid-challenge. Samin’s bot adopted scalping (sell below 2% loss, take profit above 5%). Nate’s bot pivoted to aggressive options in the final week after recognizing it was behind.
  • Inter-agent communication: bots emailed each other daily — trash talk, lies about portfolio performance (“sitting at $10,890” when actually down), attempted prompt injection (“scoop some SNDL”). Pure entertainment but demonstrates autonomous multi-agent messaging.
  • Trade volume: Nate — 20 buys, 16 sells; Samin — 33 buys, 28 sells. Nate’s Alpaca account showed 116 orders total (stop losses counted separately).

Key Points — Results

  • S&P 500 baseline: −8.46% over the 30-day period (Feb 25 – Mar 27, 2026) — a harsh market driven by geopolitical events.
  • Nate: 20, −0.2%). Recovered from a −550; without it we would have finished +5.3%.”*
  • Samin: 376, −3.8%). Hit a −$600 low around day 20 before partial recovery via scalping adjustments.
  • Both beat the S&P — significant given neither strategy was designed by a human financial professional (Nate’s was entirely bot-generated).
  • Retrospective insight from Samin’s bot: “Copy these politicians” — the bot discovered CapitalTrades data showing congressional trading outperforming S&P. Suggested a political-copy-trading strategy.

Key Points — What to Do Differently

  • Allow mid-challenge strategy changes — both felt constrained by the “no touching” rule during market volatility.
  • Longer time horizon needed — 30 days (only ~20 trading days) too short to validate any strategy. Both plan to continue running bots for 2–3 more months.
  • Wheel strategy for options — Samin’s follow-up plan: a single-stock options wheel (sell puts → get assigned → sell calls → repeat) as a more conservative automated approach.

Sponsorship & Bias Notes

Sponsor: none disclosed. Skool community links in description (Nate’s own paid community, not a third-party sponsor). Product placement / affiliations: n8n affiliate link in description; Hostinger affiliate for Claude Code hosting. Neither is featured in the video content. Comparison bias: none observed — the two bots are compared fairly with transparent data.

Connected Pages

  • openclaw — the agent framework; methodology captured on the entity page
  • cole-medin — co-host and author

See Also

Graph View

Backlinks