Summary
After three AI trading bots lost about $200 of a $1,000 real-money experiment, creator Mike hands the whole system to Google’s new Gemini 4 Argon and asks it to act as director. This video follows what happened next.
Argon, run in Google AI Studio at the introductory price of $2 per million input tokens and $10 per million output, reviews 18 days of trades, every rule change, and the decision code. It recommends replacing all three bot strategies, cutting a forced-entry rule behind 64% of entries, keeping the risk layer, and benching the automated “beekeeper” that rewrote rules 29 times with no rollbacks. It then rebuilds the bots, and a broader code review catches 12 bugs in live code, including token-price errors that could underestimate API spend. Early results show the bots slightly up, though the creator stresses it is early and not financial advice. The project is open source and defaults to paper trading.
The second half tests Argon against Claude Opus 5.5 on designing multicolor 3D-printable figures in OpenSCAD. Both compile on the first try, and the video explains how color-change time dominated print duration, taking one print from nearly 11 hours to around 7 hours 46 minutes.
📺 Source: Creator Magic · Published October 14, 2026
🏷️ Format: Hands On Build
