Summary
Nate Herk walks through the complete setup of an autonomous AI stock-trading agent built on GPT-6 Astra (OpenAI’s latest model) and the Alpaca brokerage API, documenting how he wired up a $10,000 live trading challenge. The tutorial covers the full architecture: six scheduled daily agent wake-ups mapped to key market windows (pre-open news scan at 7:45 a.m. Central through end-of-day position confirmation at 2:45 p.m.), each running as a stateless process that re-reads a shared project context to maintain continuity across sessions.
Key technical details include configuring Alpaca API keys and secrets inside a local environment file, structuring the project directory to isolate trading context from other Claude Code workspaces, and optionally subscribing to Alpaca’s $99/month real-time market data plan for faster quote access. Herk also outlines how the agent uses sub-agents for initial strategy research — fanning out roughly ten parallel researchers to build a consolidated strategy document — before locking in a day-trading approach for the seven-day challenge window.
The video frames two distinct use cases: outsourcing strategy formulation entirely to the agent for users without an existing trading methodology, and using the agent as an execution layer for traders who already have a consistent strategy but want automated monitoring and alerts. A prior Claude-based challenge that returned 8% above the S&P 500 in one month is cited as motivation for upgrading to the newer GPT-6 model.
📺 Source: Nate Herk | AI Automation · Published September 07, 2026
🏷️ Format: Tutorial Demo
