Nate Herk

YouTube creator focused on AI automation and practical Claude Code workflows. Runs UpAI and hosts the AI Automation Society community on Skool.

Channels

  • YouTube: Nate Herk | AI Automation — practical AI automation tutorials
  • Community: AI Automation Society on Skool (free + paid tiers) — courses, unlimited support, resource docs
  • Site: skool.com/ai-automation-society-plus

Content in This Wiki

Key Ideas

  • Demonstrated ingesting 36 YouTube transcripts in ~14 minutes and a long article in ~10 minutes using Claude Code
  • “Claude Code is the car, the AI model is the engine” — framing for why model swapping works
  • Pointed out that setting only ANTHROPIC_MODEL in OpenRouter config causes silent fallback to paid Anthropic Haiku for tool calls
  • Human-as-board pattern: operate Paperclip in 30-minute board sessions (set high-level goals, approve hires) rather than managing individual agent sessions
  • “Businesses don’t want fancy AI — they want boring workflows that save time, save money, or remove mistakes”
  • The clogged pipe diagnostic: “If 500 new clients showed up tomorrow, what would break first?” — reveals the real bottleneck
  • “Give me 6 hours to chop down a tree and I will spend the first four sharpening the axe” — quoted re: planning. Spending more tokens upfront on planning is worth it if execution is faster and higher-quality
  • Ultra Plan in Claude Code uses 3 parallel exploration agents + 1 critique agent running cloud-hosted Opus 4.6 — multi-agent exploration is the structural reason it produces better plans than local single-agent linear thinking
  • Advisor Strategy: “It’s not a matter of which model is best — the question is for this specific task, which model should I be using?” — use Opus only for the hard steps, Haiku/Sonnet for everything else. Test hundreds of prompts before production; saving tokens is only worth it if quality doesn’t drop. /model opus-plan in Claude Code is the practical implementation.

See Also