Descriptions:
Claude Code’s newly released Routines feature enables developers to run scheduled AI agents entirely on Anthropic’s cloud infrastructure — no local machine required. In this practical walkthrough, automation builder Nate Herk demonstrates how to configure and deploy these remote agents using both the Claude Code desktop app and the claude.ai/code web interface, covering all three trigger types: scheduled (minimum hourly cadence), API call, and GitHub events.
Herk addresses the key architectural constraint that trips up newcomers: because each routine run clones a linked GitHub repository and destroys it after completion, local .env files containing API keys are never available. The solution is Notion’s Cloud Environment settings, where credentials are stored as environment variables and injected at runtime. He also explains why browser-cookie-dependent automations break in remote sessions — every run is stateless — and how permission levels (Trusted vs. Full) affect what Claude can do unsupervised.
Practical limits covered include the 15 routine runs per day cap on the $200/month Max plan, the inability to run automations more frequently than once per hour, and how code-modifying routines differ from pure task automations (the former creates a new branch rather than discarding output). This is a must-watch for anyone migrating existing Claude Code workflows to always-on, serverless execution.
📺 Source: Nate Herk | AI Automation · Published April 14, 2026
🏷️ Format: Tutorial Demo
