Summary
YouTuber Nate Herk documents a real-money experiment giving OpenAI’s Codex, powered by GPT-6 Astra, $10,000 to autonomously trade stocks over a seven-day challenge, with the explicit goal of beating the S&P 500. He details the multi-agent system he built, including six scheduled “wake-up” routines per trading day that read market news, evaluate positions, and execute trades, all coordinated through handoff messages so the agents behave like one continuous worker.
The video tracks daily performance, showing the portfolio’s ups and downs—including a rocky start where the account dipped to $9,876—and Herk’s interventions, such as pushing the agent to take more aggressive positions when too much cash sat idle. He explains the guardrails of the challenge: he can make up to two tweaks per day but cannot halt trading even if losses mount.
The case study offers a candid look at both the promise and friction of deploying autonomous AI agents for high-stakes financial decision-making, including how well (or poorly) GPT-6 Astra reasons about risk, urgency, and portfolio allocation compared to simply investing in an index fund.
📺 Source: Nate Herk | AI Automation · Published September 28, 2026
🏷️ Format: Workflow Case Study
