Summary
Nate Herk walks through building a 24/7 AI trading agent powered by Claude Opus 4.7 inside Claude Code, using a new Anthropic feature called routines to enable persistent scheduled execution. The project builds on a prior 30-day experiment where an OpenClaw-based agent running Opus 4.6 beat the S&P 500 by approximately 8% while trading with $10,000 in an Alpaca paper account. With Opus 4.7’s improved agentic financial analysis scores and Claude Code routines providing cron-style scheduling, Herk migrates the entire setup and expands it.
The tech stack is fully documented: Claude Code handles scheduling and orchestration across pre-market, market open, midday, and close intervals; Alpaca’s trading API executes paper trades; Perplexity API handles market research; and ClickUp receives end-of-day summary notifications. The agent maintains a trade journal and research log in the file system to preserve context and strategy knowledge across sessions. The video covers Alpaca API key setup, ClickUp webhook configuration, Claude Code project initialization, and the full migration of strategy documents and trading history from the previous OpenClaw agent using Claude’s plan mode.
The practical core of the video is a complete walkthrough of how Claude Code routines replace a traditional orchestration framework for always-on agents: the model can run on a schedule, write to its own knowledge files, call external APIs, and send notifications without any wrapper infrastructure. Viewers are given enough detail to replicate the setup with their own trading strategy and preferred notification platform.
📺 Source: Nate Herk | AI Automation · Published April 17, 2026
🏷️ Format: Hands On Build
