Descriptions:
Nate Herk walks through three practical methods for deploying Claude Code agents so they run autonomously on a schedule — a question that comes up constantly for developers who have built agents locally but haven’t moved them into production. The video frames the choice along two axes: where the agent runs (local machine vs. cloud) and how deterministic its behavior is (scripted automation vs. full agentic loop).
The first method is Claude Code’s built-in cron system, using the cron-create, cron-list, and cron-delete tools to schedule session-scoped loops directly from natural language prompts. The second covers Claude Desktop’s scheduled tasks alongside Anthropic’s cloud-hosted Claude Routines, which run on Anthropic’s own infrastructure regardless of whether the user’s machine is on — though remote routines are capped at 5 runs/day on Pro, 15 on Max, and 25 on Team/Enterprise plans. The third method, previewed in the middle section, involves deploying agents to external cloud infrastructure such as Modal or a VPS for fully cloud-native, uncapped execution.
Throughout, Herk uses his own production automations — including a ‘School Wins’ engagement agent and a Hermes agent — as live examples, showing real prompt injection, project context, and skill execution. The comparison is structured enough to serve as a practical decision guide for developers evaluating deployment options for Claude-based workflows.
📺 Source: Nate Herk | AI Automation · Published May 15, 2026
🏷️ Format: Comparison
