Attribution note: all revenue figures ($4M/yr profit, $160K/mo agencies, $72K/mo, $92K/mo, $400K/mo) are the creator's self-reported claims, not verified.
The Core Doctrine
Sell before you build
Pitch productized automation ideas to prospects, build only what gets validated. "You don't build and then hope people come."
businessReserve LLMs for judgment
Wrap probabilistic models in deterministic frameworks. Error rates compound: 5 steps at 90% = 59% overall. Code handles execution; LLM handles judgment.
architectureDefinition of done
Every prompt/job closes with a measurable "definition of done" — the difference between satisfied and underwhelmed results.
promptingBoring systems win
Explainable, interpretable automations sell for $1.5K+ each. "You can print money so long as you solve a customer problem."
products60-30-10 model routing
Cheap models (Haiku/Flash) for volume, mid-tier (Sonnet) for standard work, top models (Opus/Gemini Pro) for synthesis.
economicsSecurity as differentiator
Vibe-coded economy: leaked keys = "millions". ~99% skip security audits — running the audit is a moat. Two passes, fresh context each time.
securityThe 5 Sellable Automations ($1.5K+ each)
- Search-intent scraping (n8n): LinkedIn job URL → Apify scrape → filters → dedupe sheet → GPT-4o-mini relevance → AnyMailFinder email → Perplexity research → personalized icebreaker → Instantly sequence.
- Podcast repurposing engine (n8n): episode URL → transcript → 10 JSON points → per-platform generators + images → scheduled posting.
- Invoice collection (make.com): overdue status → escalating follow-ups at 7/14/21/28/35/42 days.
- Cyclic content generator (make.com): keyword form → competitor research → outline refinement → section-by-section writing → Google Doc. Sold for $10K+.
- Email categorization (make.com): AI labels + auto-files Gmail — the foot-in-the-door SMB product.
Mobile App Workflow (Expo + React Native)
- Stack: Expo/React Native (one codebase, hot reload) — rejected Flutter/Firebase and Capacitor.
- Design framework: core function → core loop (action→reward, <30s) → accessory features → 5–7 screens max → retention hook (unfinished state, challenges, push check-ins).
- Testing: three-level loop — Chrome → screen mirror → real phone (haptics/push only verifiable on device).
- Database + auth: Supabase with RLS + automatic policies ("default-open DB policies are the classic vibe-code leak").
- AI features: Supabase Edge Functions → Claude API → coaching nudges; monetized as "your API key + user subscription with margin".
- Publishing: EAS build → Play Store (business email ~$6-7/mo, physical-device verification) / App Store (screenshots, privacy policy hosted, reviewer credentials, free pricing).
- Security audit: fresh context + giant checklist → 50-60 findings tiered → fix → second pass with new model instance.
Agentic Frameworks
- PTMRO — Planning, Tools, Memory, Reflection, Orchestration (the agent loop).
- DO — Directives (SOPs as markdown, human-readable) + Orchestration (LLM) + Execution (atomic deterministic scripts). "Cloudify" only the execution layer; leave the orchestrator out of the 3am loop.
- C.L.E.A.R. prompting — Clarity (measurable outcomes), Logic (steps + decision points), Examples (edge cases), Adaptation (iterate vs an eval), Results (validate, measure ROI).
- Self-annealing — system prompt instructs the agent to diagnose → fix → update its own scripts/directives; guardrails: cost thresholds, never modify keys, changelog.
- Skills over MCPs — skills load only ~63 tokens of front matter; convert MCP prototypes into skills for token efficiency.
- Fan-out/fan-in — cheap researchers → top-model synthesizer; stochastic consensus (10 agents, aggregate); implementer → fresh-context reviewer → resolver.
Context & Economics
/compactaround 250-300K tokens (saves 40-60%); fresh chats ~20K tokens.- Skills/front-matter at top of context; everything else behind grep/glob (the "iceberg" technique).
- Self-hosting n8n ≈ $5-10/mo vs per-execution cloud — the agency economics lever.
- Prompt length kills quality: ~20% accuracy drop from 250→3000 tokens; shorter + denser wins; one-shot beats few-shot; constrain output formats; Monte Carlo prompt testing (10-20 outputs, score, keep winner).
Business Model Notes
- Positioning formula: "I build X for Y." Copy structure: is this spam → who are you → why it matters → soft CTA.
- Niche selection: communities (Skool/Reddit) → problem → 2 services per niche → Apollo leads → Apify scrape → icebreakers → Instantly (warmed inboxes, 150/day/campaign, 2-3 variants, 2-day follow-up).
- Automation is a dying skill; the durable skills are communicating business requirements to AI and systems thinking.
- Revenue stack: consulting ($100K+), agency work (Leftclick), education (Maker School) — "an agency is an agency" regardless of deliverable.
- Routine/paywall products: site generation (Gemini 3.1 Pro one-shots, Claude Code + Kling hero videos), proposal generators (2 min vs 2.5-3h manual), landing pages as commodity.
Source Files
- videos.csv — full 323-video catalog
- text/ — cleaned transcripts of top 15 videos
- Summaries: courses, automations, strategy (assembled from transcript analysis)