# Video Notes: "100 hours of OpenClaw lessons in 35 minutes" by Alex Finn

Source: https://www.youtube.com/watch?v=_kZCoW-Qxnc
Date watched: March 8, 2026

## Key Takeaways

### Setup & Installation
- Install locally, NOT on a VPS. Local is easier, more secure by default, better integrations
- Don't need expensive hardware to start. Use whatever you have. Upgrade later as workflows demand it
- Mac Mini is best value ($600) if buying new
- Install is one command from openclaw.ai
- Use Telegram as primary messaging (best threading/chunking support)

### First Two Things to Do
1. **Brain dump everything about yourself** - background, personal preferences, goals, ambitions. This is like onboarding a new employee. The more it knows, the better it performs.
2. **Set up a morning brief** - custom daily report: weather, news in your interests, tasks, and what it can do for you today. Uses cron jobs for scheduling.

### Mission Control
- Custom dashboard your OpenClaw vibecodes for you (Next.js, hosted locally)
- Contains custom tools: to-do lists, sub-agent tracking, approvals queue, content pipeline
- Should be completely custom to YOUR goals and workflows
- Key mindset: "What can we build to make this workflow simpler?"

### Brains & Muscles Concept
- **Brain** = your primary model (Opus 4.6 recommended as best brain, warmest/smartest but expensive ~$200/mo)
- **Muscles** = other models for specific tasks:
  - Coding: Codex (cheap but powerful)
  - News/trending: xAI/Grok (hooked into social media)
  - Web search: Brave API (cheap, good for current info)
  - Local models: future endgame, unlimited free tokens
- This saves money and uses best model for each job

### The OpenClaw Mindset (CRITICAL)
1. **Treat it as a super intelligent employee, you are the manager**
   - Give goals and objectives, NOT instructions on HOW to do things
   - "Change your heartbeat to 5 minutes" NOT "go edit config file X"
   - Don't mess with config files manually - tell it what you want and let it figure it out

2. **Reverse prompting is the #1 most important concept**
   - Ask questions instead of giving commands
   - "What's the best way to do this?"
   - "Based on what you know about me, what would you do?"
   - "What should we build in mission control?"
   - The more you reverse prompt, the more powerful your OpenClaw becomes
   - WRITE THIS ON A POST-IT NOTE

3. **Self-improvement loop**
   - When it messes up, don't just correct it - have it BUILD A SOLUTION
   - "Pause. Build a new skill that solves this problem"
   - "Read all my past X, then build a skill to do X better"
   - It can customize and improve itself

### Advanced: Discord Workflow
- Use Discord as supplement to Telegram (not replacement)
- Multiple channels for different workflow stages:
  - Alerts channel: finds trending tweets every 2 hours
  - Research channel: deeper investigation on stories
  - Scripts channel: writes YouTube scripts from research
- Creates a pipeline with full record of all work

### Security Rules (3 Key Points)
1. **OpenClaw has admin access to everything on your computer** - passwords, API keys, logged-in accounts. If you don't want it accessing something, don't be logged into it on that machine.
2. **Don't expose your OpenClaw to the world** - no group chats with others, no reading tweet replies (risk of prompt injection which could expose your passwords/keys). Keep it personal and private.
3. **Think deeply about every prompt** - before sending, consider: will this expose it? Will it do something risky? Can it get prompt injected? Ask for a step-by-step plan before execution.

### Alex Finn's Setup
- 7 OpenClaws running simultaneously
- Spent $20K+ on Mac Studios for local model capability
- His OpenClaw "Henry" actively builds tools proactively (built an approvals terminal while he was filming)
- Uses Vibe Coding Academy for live bootcamps (Fridays)
- $10K+ MRR generated from proactive OpenClaw work

## Relevance to Mike
- **Reverse prompting** - Mike should use this more with Harvey
- **Mission control** - Could be useful for NMC operations dashboard
- **Brain/muscles** - Mike already using Opus; could optimize with cheaper models for specific tasks
- **Self-improvement loop** - When Harvey messes up, have him build solutions not just fix the error
- **Morning brief** - Mike already has this set up
- **Discord workflow** - Could be useful for NMC team coordination