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
- Andrej Karpathy Just 10x’d Everyone’s Claude Code — Demonstrates Karpathy’s LLM Wiki Pattern in practice; shows two working wikis and the executive assistant integration
- Ollama + Claude Code = 99% CHEAPER — Two methods for running Claude Code with open-source models (local via Ollama, cloud via OpenRouter)
- Claude Code + Paperclip Just Destroyed OpenClaw — Demonstrates Paperclip for multi-agent orchestration; building a persistent AI “company” using Claude Code agents with heartbeats, ticketing, and a CEO/engineer/QA hierarchy
- I’ve Built 500 AI Workflows, This is What Businesses Want in 2026 — The 5 highest-value AI automation workflows for selling to businesses, plus the clogged pipe framework for identifying which one a client needs
- Planning In Claude Code Just Got a Huge Upgrade — Walkthrough of Claude Code’s new Ultra Plan feature: cloud-hosted Opus 4.6 with 3 parallel exploration agents + 1 critique agent. Side-by-side dashboard build shows Ultra Plan at ~10–15 min vs local plan at ~45 min total.
- Unlock the Next Evolution of Agents with Human-like Memory (n8n + Zep) — Long-term knowledge-graph memory for n8n agents via zep; the cost-trap fix that drops per-turn tokens ~2.4×.
- I Tested Claude’s New Managed Agents — Hands-on review of Anthropic Managed Agents: cloud sandbox runtime at $0.08/hr; great onramp for Claude.ai users, missing crons/heartbeats for power users; three private-preview features (Outcomes, Multi-Agent Orchestration, Persistent Memory) define whether it becomes a real Claude Code competitor.
- How I Automated Faceless Shorts with AI in n8n — Free n8n template for fully automated faceless YouTube Shorts via Ideas Agent + Creator Agent + JSON-to-Video + 11Labs + Suno; the canonical n8n + generative-media short-form pipeline.
- Claude Just Told Us to Stop Using Their Best Model — Walkthrough of Anthropic’s Advisor Strategy (beta Messages API feature pairing Opus as advisor with cheaper executor); introduces
/model opus-planin Claude Code; side-by-side cost/quality demo. Anchors Pattern 6 on the design patterns page.
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_MODELin 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-planin Claude Code is the practical implementation.