AI is not here to replace you, but to amplify you. The key is not learning which tool to use, but knowing when to let AI do the work and when to do it yourself.
My Interpretation (Wisdom)
Many people’s understanding of AI stops at “chatting with ChatGPT,” but real AI application has three stages: Prompts (learn to ask) → Knowledge Base (build a personal knowledge base) → AI Agent (let AI take action autonomously). Most people are stuck in the first stage and think AI is not useful, when in fact they have not clearly stated the problem.
The core principle of AI writing is that your life experience and personal perspective are irreplaceable Input; leave Process to AI for editing and organizing, and remember to review the Output yourself. Content produced with AI but without your perspective is garbage, because anyone can produce the same thing.
Tools will keep changing, but the underlying logic does not: AI extends thinking; it does not replace thinking.
A question I often hear is whether to build an AI Agent Team. My answer is that most people do not need to do this deliberately. Humans need departments because one person can only master one or two specialties over a lifetime; it is a necessary compromise. AI has no such limit. Building departmental walls around it only moves the old problem of communication costs growing quadratically onto your own computer. One well-equipped generalist assistant with memory and skills covers most scenarios. Multiple Agents are an exception where one pays to save time; they are worthwhile only when subtasks are independent and do not need shared context.
Strategy and Methods (Knowledge)
Three-Stage AI Learning Path
- Prompts stage: Learn to communicate with AI using the TRIRO framework (Task, Role, Iterate, Reference, Output).
- Knowledge Base stage: Build a personal knowledge base so AI understands your context better.
- AI Agent stage: Let AI complete multi-step tasks autonomously using MCP, Claude Code, and n8n integrations.
Institutionalizing AI Collaboration
- Externalizing Judgment: Advanced models should not only solve the current difficult problem, but leave rules about when to stop, ask, change direction, and how to verify the result.
- Shared Upstream + Personalized Entry: Standardize general SOPs, scripts, dictionaries, and failure cases in the team’s shared upstream; preserve paths, private resources, tool routing, and taste at each person’s entry point.
Harness Engineering
- Situation: When AI moves from one-off conversations to long-running execution, multiple tools, handoffs among people, and high-stakes work.
- Definition: A prompt says what to do this time; context says what can be read this time. A harness combines instructions, data, tools, memory, permissions, and verification into a sustainable operating system.
- Method: Use Tool Neutrality and Context Compact to manage context and tool boundaries, Externalizing Judgment to preserve stopping and verification conditions, Shared Upstream + Personalized Entry to manage shared capabilities and local differences, and Automation for reliable triggering.
- Current level: For now, keep it as Knowledge H3 on this Theme. It currently appears mainly in AI courses and Agent-system implementation. Once it is reused reliably in a third independent setting, consider promoting it to a Knowledge Card.
IPO Principles (AI Writing)
- Input: Your views, experience, and stories—AI cannot replace these.
- Process: Ask AI to help edit, organize, and expand.
- Output: Always review the final result yourself to ensure it is “your voice.”
Related Content (Information)
Articles and Newsletters
- 2025-11-06: Complete AI learning path: prompts, knowledge bases, and AI Agents from beginner to advanced in three stages
- 2025-08-14: Hands-on guide to Manus AI: the most intuitive AI Agent, more than a chatbot
- 2025-08-08: New GPT-5 applications: five scenarios explained, including free users
- 2025-06-24: Latest Felo AI applications: presentation generation, interactive webpages, fact-checking, and essential workplace tools
- 2025-04-11: Latest ChatGPT 4o image generation: LINE stickers, hand-drawn menus, interior design, and product posters
- 2025-03-08: Seven useful free Chinese GPTs: build personal assistants with ChatGPT at no cost
- 2025-02-01: Memo AI tutorial: transcribe YouTube and podcasts in one click, export bilingual subtitles, video summaries, and more
- 2024-12-07: Build a free Chinese typo-correction AI bot that turns talks, meetings, and subtitle files into complete articles
- 2024-10-25: AI tools Raymond used at the end of 2024: a recommendation list with 16 use cases
- 2024-10-12: My AI translation workflow: bilingual reading of originals, PDFs, ebooks, and videos — six applications
- 2024-10-09: Recommended AI search tools: Felo Search as a more comprehensive alternative to Perplexity
- 2024-07-03: Keep It Shot: fix garbled screenshot filenames; AI renames files and can choose a folder
- 2024-06-17: GPT-4o: explanation and roundup of its nine latest applications and features
- 2024-03-18: ChatGPT tutorial: learn prompt principles and the TRIRO framework to use AI tools
- 2024-02-03: Recommended GPTs: a Chinese AI PDF book-summary assistant that saves learning time and supports multiple formats and languages
- 2023-11-29: ChatGPT tutorial: registration and upgrade principles explained simply, 2024 edition
- 2026-03-06: What is Claude Code? A complete beginner’s guide for non-engineers
- 2026-02-02: Vibe-coding learning notes: why reinvent the wheel? I made a practical mini-tool website
- 2026-01-15: Too much data to digest? Enjoy AI mind maps: hands-on review of Mapify and Xmind
- 2026-01-07: The best AI voice-input tool: hands-on review of Typeless
- 2024-09-20: AI note-taking tools: NotebookLM or Notion AI? Comparing personalized-assistant use cases
- 2023-03-07: Notion AI video and article tutorial: five writing applications
- 2025-08-14: Draft podcast-interview highlights in ten minutes: my AI writing-assistant workflow
- 2025-12-25: Who is Zhou Jia’en? From comedian to AI writer: use real experience to counter mediocre AI content
- 2026-03-22: Livestream planning: co-learning amid AI anxiety — Notion/N8N/AI Agent toolkit, Jensen Huang’s Token evaluation, and a cast-iron-pot analogy for using the wrong tool
- 2026-03-23: After getting AI Agents, can we throw away Notion and n8n? _27110823321839195
- 2026-03-20: How to choose a Zeabur dedicated server: a plain-language guide and letting AI help migrate
- 2026-03-20: Issue 176: Four sources for learning AI, Notion AI bonus livestream, automation requires a business first — Raymond Weekly
- 2026-03-18: Plain-language lesson: how AI controls software—choosing among API, CLI, MCP, and browser control
- 2026-03-16: Help an AI assistant understand webpages: comparing Firecrawl, Playwright, and WebFetch
- 2026-03-13: A record of building a complete AI-double assistant with Claude Code in 21 days
- 2026-03-08: Issue 175: Notion beginner in-person class, a non-engineer’s Claude Code journey, Manny joins — Raymond Weekly
- 2026-03-06: What is Claude Code? From Claude Cowork to Code: a complete beginner’s guide for non-engineers
- 2026-02-26: Issue 174: AI double at work, in-person Notion classes, repricing my own salary — Raymond Weekly
- 2026-02-20: Zeabur use case: deploy AI Hub / OpenClaw on a VPS; 50% discount coupon #2
- 2025-09-19: Notion 3.0 upgrade: AI Agent arrives, database permissions, offline mode, map view, and release notes
- 2025-08-05: How I started freelancing and raised my rates: highlights from Raymond’s interview with Lufan
- 2025-04-03: n8n Chinese tutorial: beginner interface, webhooks, and automation workflows
- 2025-02-26: n8n tutorial: use Zeabur to back up workflows and credentials and export workflows to others
- 2024-11-24: Install n8n in one click: get an unlimited-quota automation tool free for a limited time with enterprise features
- 2024-11-24: Start here: a 100-hour automation-work-method series using Zapier and n8n
- 2023-11-29: ChatGPT tutorial: registration and upgrade principles explained simply, 2026 edition
- 2023-09-05: Are you burned out? A product expert’s advice on combining expertise and business — interview with Han Han, Digital Life Atlas #22
- 2023-08-22: The most effective learning method is deconstruction: what the learning pyramid leaves out and examples of the Feynman technique
- 2023-08-02: How to become good at time management: don’t rely on willpower; two methods anyone can learn
- 2023-05-16: Behind ten years of writing by the Computer Playthings site owner: how to relax after work and turn life into projects — interview with Esor, Digital Life Atlas #19
- 2023-04-11: Canva product designer and education entrepreneur: finding long-term collaborators and keeping a remote team motivated — Digital Life Atlas #17
- 2023-03-14: The choices not made along the entrepreneurial path: the real life and underlying logic of a serial entrepreneur — interview with Lawrence Lin Yiru, Digital Life Atlas #15
- 2026-03-27: The most dangerous time to automate is before you have started earning money
- 2026-03-27: Don’t think about automation when you have not started earning money
- 2026-03-29: Connect Claude Code through iMessage so family members can use an AI assistant
- 2026-02-20: Zeabur use case: deploy AI Hub / OpenClaw on a VPS; apply for a domain, 50% discount coupon #20
- 2026-03-18: Plain-language lesson: how AI takes action—choosing among API, CLI, MCP, and browser control
- 2026-04-01: It’s done—what next? A complete guide to deployment paths through Zeabur and Cloudflare
- 2026-04-02: Claude Code CLAUDE.md and Skills tutorial: build reusable AI automation workflows
- 2026-05-20: Who is Anji? From Taiwan to the United States: impostor-syndrome struggles, self-publishing, and a Taiwan-made identity shift
- 2026-05-11: Issue 180: The meaning of an AI double, Esor livestream, Notion all-day workshop — Raymond Weekly
- 2026-05-10: You can outsource thinking, but not understanding: highlights from Karpathy’s latest interview and real competitiveness in the AI era
- 2026-05-23: Why most people do not need to deliberately build an AI Agent Team
- 2026-06-02: Issue 181: Codex desktop tutorial, free currency converter, Little Tiger’s voice course — Raymond Weekly
- 2026-06-04: Claude Code vs. Codex: full tutorial and desktop comparison for non-engineers
- 2026-06-05: Why does AI get dumber the more you chat? An introduction to context windows, Compact, and quota management
- 2026-06-09: Move an AI Agent: Claude Code to Codex, AGENTS.md, CLAUDE.md, and…
- 2026-06-19: Issue 182: Followed a star with my wife for 20 years, nearly migrated my AI brain, Waki’s AI thinking-collaboration course — Raymond Weekly
- 2026-06-25: Why does Notion need a Taiwan market lead? Understanding the strategy behind a job description
- 2026-06-28: Issue 183: Turn a course into an AI assistant, analyze Notion’s Taiwan market-lead role, choose an Agent — Raymond Weekly
- 2026-06-29: A midyear reflection at 31: 19 things life taught me
- 2026-06-29: A midyear reflection at 31: 19 things life taught me
- 2026-06-30: What is an AI Agent? Understand harnesses through the seven-layer architecture of Hermes
- 2025-11-06: Complete AI learning path: chat, collaboration, and agentic AI in three stages from Prompt to AI Agent
- 2026-07-01: Building your own AI assistant team: Hermes Agent, Claude Code, or Codex?
- 2026-07-02: Computer Playthings’ Esor: decades of Evernote notes were not wasted—AI Agent and the Second Brain
- 2026-07-11: Is Zeabur worth it after the price increase? Deployment platforms: Vercel, Supabase, Cloudflare, Netlify
- 2026-07-01: Are you a runner too? AI Agent student-project submission _28322769880644527
- 2026-07-06: Fable 5 is about to change its pricing: what can you do on the last day? _28380615901526591
- 2026-07-01: What exactly is an AI Agent? _28318721007716081
- 2026-07-12: Issue 184: Midyear reflection, an AI-sounding-prose removal tool, Esor’s AI Second Brain — Raymond Weekly
- 2026-02-20: Zeabur use case: deploy AI Hub / Hermes Agent on a VPS; 50% discount coupon #2
Videos and Livestreams
- 2026-07-19: “Worthy of respect: a happy, ordinary person” — complete teaching article for the Super AI Individual member livestream, 2026-07-19; first public explanation of the course positioning (human work system vs. AI work system), five-year department-reorganization judgment, and refrigerator-manual analogy
- 2026-08-08: Issue 185: Half-month-late birthday, why I gave the flagship course to another platform and raised the price, Alley Guide housing map — Raymond Weekly
- 2026-08-16: Google is shutting down Gems; OpenAI also closed the path for personal GPT publishing _28859287870326056
- 2026-08-16: How to make a YouTube thumbnail with Codex: from AI draft and portrait cutout to final Figma design
- 2026-08-22: Why corporate AI adoption fails: from Cloudflare layoffs to AI organizational transformation, breaking down departments, and knowledge management
- 2026-08-30: Issue 186: Featured on the App Store, a trip without a laptop for the first time, an AI Agent course that keeps growing
- 2021-06-10: What is Markdown? An essential writing technique and a guide to immersive-editor tools
- 2026-09-08: Issue 187: Connecting AI does not automatically make you stronger; love life itself; video adapter and charger — Raymond Weekly
- 2026-09-09: Use a plain-language Skill to remove AI-sounding prose while analyzing tone and preserving meaning and Taiwanese usage
- 2026-09-23: Issue 188: To my 32-year-old self, six years of paid subscriptions, I fired myself — Raymond Weekly
Social Posts
(To be tagged using the Facebook post curation tool.)
Data
- 🔖 “Humans do not write the first draft; AI does not write the final draft.”
Related Concepts
Second Brain, Communication and Expression, Writing, Automation, Learning Methods, Externalizing Judgment, Shared Upstream + Personalized Entry, Tool Neutrality and Context Compact