--- # HOW TO USE THIS FILE # 1. HUMAN: Read from top to bottom as a regular guide # 2. AI ASSISTANT: Paste this entire document into Claude, ChatGPT, or Gemini # and ask it to: # - "Explain Part 2 to me like I'm 12" # - "Help me fill out the profile template in 00-context based on my answers" # - "Turn the prompt library into a printable checklist" # - "Create a simple study plan from this guide for [your specific goal]" # 3. NOTEBOOKLM: Upload this file to notebooklm.google.com to generate an # audio study guide or structured notebook (no setup required) document_type: ai_infrastructure_tutorial audience: absolute_beginners_zero_code prerequisites: none core_principle: organized_file_system_is_not_optional --- --- # AI for Beginners: A Paint-by-Numbers Guide ## From Zero to Your First AI-Powered Project — Using Plain Text By Magnús Smári Smárason **Who this is for:** Anyone who has never used AI tools before and wants a practical, step-by-step guide to start using AI to learn, build, and create — in any field. **What you will have at the end:** A personal workspace where your prompts, AI conversations, and project files live in one organised place — ready to grow with you. **Time to complete:** About 60 minutes for the full setup. A few minutes per session after that. This is my attempt at creating something people can actually use. Check out [www.smarason.is](https://www.smarason.is) for more projects. > **Quick start:** Do Parts 1–3 today (30–45 minutes). That gives you a working workspace and your first AI conversation. Come back for the rest when you are ready. --- ## Contents - [Part 1: What You Need](#part-1-what-you-need) - [Part 2: Set Up Your Workspace](#part-2-set-up-your-workspace) - [Part 3: Your First AI Conversation](#part-3-your-first-ai-conversation) - [Checkpoint: You Are Set Up](#checkpoint-you-are-set-up) - [The Philosophy Behind This Guide](#the-philosophy-behind-this-guide) - [Part 4: Your Prompt Library](#part-4-your-prompt-library) - [Part 5: Getting Unstuck (Prompts for When You Hit a Wall)](#part-5-getting-unstuck-prompts-for-when-you-hit-a-wall) - [Part 6: What AI Gets Right, What It Gets Wrong, and Why That Matters](#part-6-what-ai-gets-right-what-it-gets-wrong-and-why-that-matters) - [Part 7: Building Your Rhythm](#part-7-building-your-rhythm) - [Part 8: Where This Can Take You (and What to Watch Out For)](#part-8-where-this-can-take-you-and-what-to-watch-out-for) - [Part 9: How Projects Grow (From Idea to Production)](#part-9-how-projects-grow-from-idea-to-production) - [Part 10: Where to Go From Here](#part-10-where-to-go-from-here) - [Appendix A: Obsidian Quick Setup](#appendix-a-obsidian-quick-setup) - [Appendix B: VS Code Quick Setup](#appendix-b-vs-code-quick-setup) - [Appendix C: Glossary](#appendix-c-glossary) --- ## Three Things to Know Before You Touch Anything **1. AI is an information factory, not intelligence.** A factory takes raw materials in, processes them, and produces output. AI works the same way — it takes your input (your question, your context, your data), runs it through patterns learned from enormous amounts of text, and produces a response. It does not "know" things the way a person does. It processes. Better raw materials in, better product out. We will return to this model in more depth in the Philosophy section after your setup is done. **2. You are always in control.** The AI only knows what you tell it. You provide context; it provides a starting point. You decide what to act on. **3. Prompts are recipes — and specificity is everything.** A prompt is what you type to the AI. A specific prompt gives a specific result. Watch the difference: **Vague prompt (weak output):** > _"Tell me about marketing."_ The AI has no idea who you are, what you sell, where you are, or what you need. You get a generic textbook answer. **Specific prompt (useful output):** > _"I run a small bakery in Reykjavík with 2 employees. We do well on weekends but weekdays are slow. What are three low-cost marketing tactics I could try this month to increase weekday foot traffic? My budget is under €200."_ Same topic. Completely different result. The second prompt gives the factory enough raw materials to produce something you can actually use. This guide gives you exact recipe-style prompts. You fill in the blanks marked with `[ ]`. --- **▶ START HERE** --- ## Part 1: What You Need **By the end of this section:** you will have accounts and tools ready to go — nothing installed yet, just decisions made. There are two levels. Level 1 is everything you need — it is free and gets you fully operational. Level 2 is what makes the experience significantly better once you are committed. ### Level 1: Need to Have (Free) You need two things: **an AI service** and **a file management tool**. #### 1. An AI Account Pick any major AI service and create a free account. They all work for this guide. Here is how to choose: - If you already live in **Google Workspace** (Docs, Gmail, Drive) → try **Gemini** at gemini.google.com - If you already live in **Microsoft 365** (Word, Outlook, Teams) → try **Copilot** at copilot.microsoft.com - If you use **X** (Twitter) regularly → try **Grok** at grok.com - If you want the best standalone conversation experience → try **Claude** at claude.ai or **ChatGPT** at chatgpt.com This guide was written and tested using Claude, so the examples reflect that — but every prompt works identically across all of these services. The differences between them matter far less than whether you actually use one. Pick one and start. You can try the others later. #### 2. A File Management Tool Once you start using AI, you will create a lot of text. Prompts you want to reuse. Responses worth saving. Notes, plans, drafts. Without a place to organise all of this, you will lose track within a week. Both tools below store your work as **Markdown files** — which brings us to something you need to understand. #### What Is Markdown? Markdown is a simple way to format text using plain characters you already know how to type. Instead of clicking a "Bold" button like in Word, you type `**bold**` and it becomes **bold**. Instead of navigating a menu for headings, you type `# My Heading`. Here is everything you need right now: ``` # Heading 1 (big heading) ## Heading 2 (smaller heading) ### Heading 3 (even smaller) **bold text** (surround with double asterisks) *italic text* (surround with single asterisks) - bullet point (dash and a space) 1. numbered item (number, dot, space) > a quote (greater-than sign and space) `inline code` (backticks around it) --- (horizontal line — three dashes) ``` That is it. You now know Markdown. Everything in this guide is written in it. Both Obsidian and VS Code understand it natively — they show your formatting live as you type. **Why Markdown and not Word or Google Docs?** Markdown files are plain text. They open on any computer, in any text editor, instantly. They are tiny. They are searchable. They work with every AI tool — you can paste Markdown directly into any AI conversation and it reads it natively. Modern `.docx` files (which are technically ZIP archives of XML — an open standard) are also durable, but they require software that understands their structure. Markdown requires nothing — it is readable as-is, even in Notepad. For an AI-first workflow where you are constantly copying text between your editor and AI tools, that simplicity matters. Now pick your tool: | Tool | Best for | Get it at | | ------------ | --------------------------------------------------------------------- | --------------------- | | **Obsidian** | A visual, note-taking experience. Feels like a modern notebook. | obsidian.md | | **VS Code** | The curious and technical-leaning. The same editor professionals use. | code.visualstudio.com | Both are free. Both store files as Markdown on your own computer. You cannot make a wrong choice here. --- ### Level 2: Nice to Have (Paid, but Worth It) Come back to this section after you have been using AI for a few weeks and know it is valuable to you. #### 1. One Paid AI Subscription The free tiers all have limits — message caps, slower models, fewer features. One paid subscription removes most of those limits. | Option | Price (approx.) | What you get | | ------------------------- | ------------------------ | ----------------------------------------------------------------------------- | | **Claude Pro** | $20/month | More messages, most capable Claude models, file uploads, longer conversations | | **ChatGPT Plus** | $20/month | Most capable models, image generation, file uploads, browsing | | **Gemini Advanced** | $20/month | Most capable Gemini models, deep Google Workspace integration | | **Grok Premium** | Included with X Premium+ | Highest-capability Grok models, image understanding | | **Microsoft Copilot Pro** | $20/month | AI integrated directly into Word, Excel, PowerPoint, Outlook, Teams | Pick one, not all. Choose based on where you spend your time: Google Workspace → Gemini. Microsoft 365 → Copilot Pro. Best standalone conversation experience → Claude Pro or ChatGPT Plus. #### 2. GitHub (Free Account, Paid Optional) GitHub is where the world stores and shares code, text, and projects. You do not need to be a programmer to use it. At its core, GitHub does two things: 1. **Version control** — it tracks every change you make to your files, so you can go back in time if you mess something up. Think of it as an unlimited "undo" that works across days, weeks, and months. 2. **Sharing and collaboration** — you can publish your work, contribute to other people's projects, and learn from how others build things. You do not need GitHub on day one. But when you are ready, sign up at **github.com** (free). The paid tier (GitHub Pro, $4/month) adds features like Copilot code suggestions — useful later, not necessary now. --- ## Part 2: Set Up Your Workspace **By the end of this section:** a folder structure on your computer ready to hold everything you create, and a personal profile file that makes every AI conversation better. ### Install Your Tool **If you chose Obsidian:** 1. Go to **obsidian.md** → Download → Install 2. Open Obsidian → Click **"Create new vault"** 3. Name it: `My AI Workspace` 4. Save it in your Documents folder → Click **Create** **If you chose VS Code:** 1. Go to **code.visualstudio.com** → Download → Install 2. Open VS Code → File → Open Folder 3. Create a new folder called `My AI Workspace` in your Documents 4. Open that folder ### Create Your Folder Structure Inside your workspace, create these folders: ``` 📁 00-context ← your personal profile and project descriptions 📁 01-prompts ← your saved prompts (the recipes) 📁 02-conversations ← saved AI responses worth keeping 📁 03-projects ← your actual work (notes, drafts, plans) 📁 04-reference ← useful information you find along the way ``` > **Why this structure?** Every AI conversation starts with context (who you are) and a prompt (what you are asking). Keeping these separate lets you mix and match — same context with different prompts, or same prompt with different contexts. ### Build Your Personal Profile This is the most important file you will create. It tells the AI who you are so it produces relevant, specific responses instead of generic ones. In your `00-context` folder, create a file called `my-profile.md`. Copy this template and fill in every `[ ]`: ```markdown # My Profile (Paste this at the start of AI conversations for personalised responses) ## About Me - Name: [your first name] - Background: [1-2 sentences — what do you do, what is your experience level?] - Location: [city/country — helps with local context] ## What I Am Working On - Current project or interest: [what brought you here?] - My goal with AI: [what do you want AI to help you accomplish?] - Timeline: [is there a deadline, or are you exploring at your own pace?] ## My Experience Level - AI tools: [never used / tried once or twice / use occasionally / use regularly] - Technical skills: [none / basic computer use / some spreadsheets / comfortable with technology / can code] - Writing: [I avoid writing / I write occasionally / I write regularly / writing is part of my job] ## Preferences - I learn best by: [reading / watching / doing / a mix] - I prefer responses that are: [short and direct / detailed with explanations / step-by-step with examples] - Topics I am interested in: [list 2-3 areas] ## What I Want to Avoid - [anything you do not want — e.g., "overly technical jargon", "unsolicited health advice", "sales language"] ``` Fill in every line. If you genuinely do not know something, write "not sure yet" — that is more useful to the AI than a blank. ### What to Be Careful Sharing — Until You Understand the Privacy Landscape Your profile is useful context. But until you understand how AI services handle your data, be cautious about pasting in: - **Passwords or login credentials** — never, under any circumstances - **Bank account numbers, credit card numbers, or financial account details** - **Government ID numbers** (social security, kennitala, passport numbers) - **Confidential work documents** — check your employer's AI policy first - **Other people's private information** without their knowledge - **Medical records with identifying information** — paraphrase instead of pasting When you type something into an AI tool, you are sending it to a company's servers. Most major providers have data policies that limit how they use your input — some offer options to opt out of training entirely, and paid tiers often come with stronger privacy guarantees. As you become more comfortable, learn to read these policies and make informed decisions. **Privacy policy links (verify these are current before acting on them):** - Claude/Anthropic: [anthropic.com/privacy](https://www.anthropic.com/privacy) - ChatGPT/OpenAI: [openai.com/policies/privacy-policy](https://openai.com/policies/privacy-policy) - Gemini/Google: [ai.google/privacy](https://ai.google/privacy) - Grok/xAI: [x.ai/legal/privacy-policy](https://x.ai/legal/privacy-policy) - Microsoft Copilot: [privacy.microsoft.com](https://privacy.microsoft.com) For now, a good starting rule: if you would not email it to a stranger, do not paste it into an AI until you have checked that service's data policy. --- ## Part 3: Your First AI Conversation **By the end of this section:** your first saved AI conversation and a reusable prompt in your workspace. ### Set Up Your AI to Know You **Custom instructions save you time.** Instead of pasting your profile at the start of every conversation, most AI tools let you save it once: - **Claude:** Go to your profile settings → look for "Memory" or "Custom instructions" → paste your profile. - **ChatGPT:** Settings → Personalisation → Custom instructions → paste your profile in the "What would you like ChatGPT to know about you?" box. - **Gemini:** Look for "Extensions" or "Saved info" in settings. Set this up now. Your AI will start every conversation already knowing who you are. Keep the profile file in your workspace as your master copy — update the custom instructions from it whenever your details change. If your tool does not support custom instructions, paste your profile at the start of each new conversation. It takes 10 seconds. ### You Can Upload Files Too Every major AI tool now lets you drag files directly into the conversation — PDFs, images, spreadsheets, documents. Drop a 30-page PDF into Claude and type "summarise the key points." No copy-pasting needed. This works for: - PDFs (research papers, reports, contracts, manuals) - Images (screenshots, diagrams, photos of handwritten notes) - Spreadsheets and CSVs (data, budgets, lists) - Code files (if you start coding later) ### Your First Prompt In your `01-prompts` folder, save this as `prompt-01-explore-my-interest.md`: ```markdown [PASTE YOUR PROFILE HERE] --- I am interested in [YOUR TOPIC OR FIELD]. I am a complete beginner and want to understand the landscape before diving in. Please give me: 1. A plain-language overview of this field — what it is, why people care about it, and who it is for 2. The 3-5 most important concepts a beginner should understand first 3. A suggested learning path: what to learn first, second, and third 4. The most common beginner mistakes and how to avoid them 5. 3 free resources (websites, books, videos, communities) where I can learn more 6. One small project or exercise I could do this week to get started Keep your language simple. Explain jargon when you use it. ``` Open your AI tool. Start a new conversation. Paste your profile (or rely on custom instructions), then paste this prompt. Fill in `[YOUR TOPIC OR FIELD]`. Send it. ### Save the Response Copy the AI's response into your `02-conversations` folder as a new file: `conversation-01-[topic]-[date].md` You now have a permanent record. The AI will forget this conversation — you will not. ### Push Back — This Is Where the Real Value Lives Most beginners stop after one exchange. Do not. The AI's first response is a draft, not a final answer. The conversation is the real tool, not any single prompt. After reading the response, try any of these follow-ups (still in the same conversation): **If the response is too generic:** > _"That's too broad. I specifically need advice for [your situation]. Try again with more detail."_ **If one point is interesting and the rest is not:** > _"Good. Expand on point 3 — give me twice as much detail on that and drop the rest."_ **If something does not sound right:** > _"You said [X]. That does not match what I have read elsewhere. Are you sure? Show me why you think this is correct."_ **If you want a different format:** > _"Rewrite that as a simple table"_ or _"Give me just the action items as a checklist"_ or _"Explain it like I am 12."_ **If it is too long:** > _"Too much. Give me the three most important points and nothing else."_ You can push back, redirect, challenge, and refine as many times as you want. The AI does not get offended. It does not get tired. The best results come from iteration — your third or fourth exchange is almost always better than the first. Remember: the factory produces output on demand. Run it again with better instructions until the product meets your standard. --- ## Checkpoint: You Are Set Up **You are done for today.** You now have: - An AI account that knows who you are - A workspace with organised folders - Your first saved conversation - A reusable prompt Close the laptop. Bookmark this page. Come back tomorrow for Parts 4–10 — they build on what you have already done. Or keep going now if you are on a roll. --- ## The Philosophy Behind This Guide _You have a working setup now. Here is why it was built this way — read this when you are curious, skip it if you want to keep building._ This guide is built on a simple idea: **AI is an information factory that magnifies human capability.** You already have interests, questions, and goals. AI does not replace those — it amplifies them. A good prompt turns your half-formed idea into a structured plan. A saved conversation becomes a knowledge base you can search. A folder of notes becomes the skeleton of something bigger. This philosophy draws on Daniel Miessler's [Personal AI Infrastructure (PAI)](https://github.com/danielmiessler/Personal_AI_Infrastructure) project, which treats AI not as a novelty but as personal infrastructure — a system that grows with you over time. The core insight from PAI: > **The system design matters more than the model.** A well-organised workspace with a decent AI model will outperform a brilliant model with no structure around it. The methodology in this guide — the prompt architecture, the diagnostic approach to AI limitations, and the adversarial framework — is original work, developed through 4,000+ hours of hands-on AI work and refined through teaching AI adoption at the University of Akureyri. The PAI philosophy provides the foundation; the practical system you are building is how that philosophy becomes real. You do not need to be technical. You do not need to code. You need a text editor, an AI tool, and the willingness to type what you are thinking. --- ## Part 4: Your Prompt Library **By the end of this section:** six reusable prompts saved in your workspace, covering the most common things people use AI for. Each of these goes in your `01-prompts` folder. Save them now. Use them whenever you need them. --- ### Prompt 2 — Break Down a Complex Topic ```markdown [PASTE YOUR PROFILE] I am trying to understand [TOPIC]. It feels overwhelming because [WHY IT FEELS COMPLEX]. Please: 1. Break this topic into 5-7 smaller sub-topics I can learn one at a time 2. Put them in the order I should learn them (foundations first) 3. For each sub-topic, give me a one-sentence summary and one key question it answers 4. Tell me which sub-topic is the most important to get right — the one where mistakes cause the most problems later 5. Estimate roughly how long each sub-topic would take to learn at a casual pace ``` --- ### Prompt 3 — Create a Project Plan ```markdown [PASTE YOUR PROFILE] I want to build / create / accomplish: [DESCRIBE YOUR PROJECT OR GOAL] My constraints: - Available time per week: [X hours] - Budget: [X or "zero — free tools only"] - Deadline: [date or "no deadline, working at my own pace"] - Skills I have: [what you already know] - Skills I lack: [what you know you do not know yet] Please create: 1. A project broken into phases (no more than 4 phases) 2. For each phase: what to do, what the output is, and how I know it is done 3. The tools I will need (prioritise free options) 4. The single biggest risk or obstacle and how to mitigate it 5. A "minimum viable version" — the simplest version of this project that would still be valuable Format the phases as a table with columns: Phase | Tasks | Output | Done When ``` --- ### Prompt 4 — Summarise Something I Am Reading ```markdown [PASTE YOUR PROFILE] I am reading the following and want to make sure I understand it correctly. --- ## [PASTE THE TEXT YOU WANT SUMMARISED — an article, chapter, documentation, etc.] Please: 1. Summarise the main argument or point in 2-3 sentences 2. List the key facts or claims made 3. Identify anything that surprised you or seems counter-intuitive 4. Explain any technical terms used, in plain language 5. Tell me what questions I should be asking after reading this ``` --- ### Prompt 5 — Weekly Review and Next Steps ```markdown [PASTE YOUR PROFILE] Here is what I worked on this week: WHAT I DID: - [list what you accomplished, even small things] WHAT WENT WELL: - [what clicked, what felt productive] WHAT I STRUGGLED WITH: - [where you got stuck, confused, or lost motivation] WHAT I WANT TO FOCUS ON NEXT WEEK: - [your intention, even if vague] Please: 1. Acknowledge what I accomplished (be specific about why it matters) 2. For each struggle, suggest one concrete next step I could try 3. Based on my progress, suggest what my priority should be next week 4. Flag anything I might be overlooking or avoiding 5. Rate my progress on a scale — am I on track, behind, or ahead of where a typical beginner would be? ``` --- ### Prompt 6 — Explain It Like I Am Teaching Someone Else ```markdown [PASTE YOUR PROFILE] I need to explain [CONCEPT OR TOPIC] to someone who knows nothing about it. Please help me: 1. Write a 3-paragraph explanation suitable for a general audience 2. Create one analogy or metaphor that makes the concept click 3. Identify the one thing people most commonly misunderstand about this 4. Give me 3 questions someone might ask after hearing my explanation, and how to answer them 5. Suggest a simple diagram or visual I could sketch to support the explanation Use plain language. No jargon unless you explain it immediately. ``` --- ## Part 5: Getting Unstuck (Prompts for When You Hit a Wall) **By the end of this section:** six emergency prompts saved for the exact moments when you do not know what to ask next. This is the section most guides leave out. When you are stuck, you often cannot even form a question. These prompts are designed for exactly that moment. --- ### "I do not know what I do not know." Use when: You are lost and cannot articulate why. ```markdown [PASTE YOUR PROFILE] I am working on [TOPIC/PROJECT] and I have hit a wall. I am stuck but I cannot clearly articulate why. Here is everything I can tell you about my situation: - What I was trying to do: [even if vague] - Where I got stuck: [what happened, or "I do not know"] - What I have tried: [anything, even if it did not work] - How I am feeling about it: [frustrated / confused / overwhelmed / bored / other] I need you to: 1. Ask me 5 specific questions that will help us figure out where I am stuck 2. Based on what I have told you, guess what the most likely blocker is 3. Suggest the single smallest step I could take right now to make progress 4. Tell me if I might be overcomplicating this — and if so, what the simpler version looks like ``` --- ### "The AI gave me a response I do not understand." Use when: The AI answered but you cannot make sense of it. ```markdown I just received this response from an AI and I do not fully understand it: --- ## [PASTE THE CONFUSING RESPONSE] Please: 1. Rewrite the key points in the simplest possible language — assume I have no background knowledge 2. Highlight which parts are most important for me to understand 3. Tell me which parts I can safely ignore for now 4. If there are any technical terms, define each one in one sentence 5. What is the one thing this response is really trying to tell me? ``` --- ### "I tried what the AI said and it did not work." Use when: You followed instructions but something went wrong. ```markdown [PASTE YOUR PROFILE] I tried to follow AI-generated instructions for [TASK] and something went wrong. THE INSTRUCTIONS I FOLLOWED: [paste or summarise what you were told to do] WHAT ACTUALLY HAPPENED: [describe what went wrong — error messages, unexpected results, nothing happened] WHAT I EXPECTED TO HAPPEN: [what you thought would happen] MY SETUP: - Computer: [Mac / Windows / Linux] - Tool/app I was using: [name] - My experience level: [beginner / intermediate] Please: 1. Diagnose what likely went wrong based on my description 2. Give me the fix, step by step, assuming I am a beginner 3. Explain why it went wrong so I learn from it 4. Tell me what to check first if this happens again ``` --- ### "I have been at this for hours and I am going in circles." Use when: Multiple approaches, nothing working. ```markdown [PASTE YOUR PROFILE] I have been stuck on [PROBLEM] for [HOW LONG]. I keep trying different things but nothing works and I am going in circles. WHAT I HAVE TRIED (in order): 1. [first attempt] 2. [second attempt] 3. [third attempt] (add more if needed) RESULT OF EACH ATTEMPT: 1. [what happened] 2. [what happened] 3. [what happened] I need you to: 1. Look at my attempts and tell me if I am approaching this from the wrong angle 2. Suggest a COMPLETELY different approach I have not tried 3. Tell me the most common reason people get stuck on this exact problem 4. Give me a decision: should I keep trying, take a break, ask a human for help, or simplify my goal? 5. If I should simplify, what would a "good enough for now" version look like? ``` --- ### "I do not know if what I have is any good." Use when: You have produced something but cannot judge its quality. ```markdown [PASTE YOUR PROFILE] I have been working on [PROJECT/TASK] and I have produced something, but I honestly cannot tell if it is good, bad, or somewhere in between. HERE IS WHAT I HAVE: [paste your work, or describe it in detail] MY GOAL WAS: [what you were trying to achieve] Please: 1. Rate what I have on a scale of 1-10, and be honest — I would rather know now than waste more time going in the wrong direction 2. Tell me the strongest aspect — what is working well 3. Tell me the weakest aspect — what needs the most improvement 4. Give me 3 specific, actionable things I could do to improve it 5. Tell me: if I stopped here, would this be usable? Or does it need more work before it is valuable? ``` --- ### "I have no idea what to do next." Use when: You finished a step but cannot see the next move. ```markdown [PASTE YOUR PROFILE] I have completed: [WHAT YOU JUST FINISHED] And now I genuinely do not know what to do next. My overall goal is [YOUR GOAL] but I cannot see the path from where I am to where I want to be. Please: 1. Based on what I have completed, what are the 3 most logical next steps? 2. Which of those 3 would give me the biggest return on my time? 3. Is there something I should do before any of those (a prerequisite I missed)? 4. Draw me a simple roadmap: where I am → next milestone → final goal 5. What would someone more experienced do right now in my position? ``` --- ## Part 6: What AI Gets Right, What It Gets Wrong, and Why That Matters **By the end of this section:** a clear-eyed understanding of AI's capabilities, its failure modes, and your role as the person responsible for what you do with its output. ### What the Factory Produces Well — and What It Does Not | AI does this well | AI does NOT do this | | ----------------------------------- | -------------------------------------- | | Summarise and structure information | Guarantee factual accuracy | | Generate first drafts and outlines | Replace domain expertise | | Explain complex concepts simply | Know your specific situation | | Calculate, compare, and analyse | Make decisions for you | | Suggest approaches and alternatives | Access real-time information (usually) | | Follow structured prompts precisely | Remember previous conversations | AI output is a starting point, not a final answer. Verify anything important. Use AI to think faster, not to stop thinking. For a deeper look at the specific ways AI fails — hallucination, sycophancy, and user bias — and how to defend against them, see Part 8. ### Why This Matters More Than You Think Here is something most AI guides will not tell you: **the more you rely on AI, the more your own judgment matters — not less.** When you let AI draft your emails, plan your projects, and summarise your reading, something subtle happens. The skills you are outsourcing start to atrophy — like a muscle you stop using. You get faster at producing output but slower at evaluating whether that output is any good. You start accepting the AI's first response because it sounds right, without checking whether it is right. This is not a reason to avoid AI. It is a reason to use it deliberately. **The practical test:** If you cannot explain — in your own words, without looking at the screen — why you are following the AI's suggestion, you do not understand it well enough to act on it. Go back and ask the AI to explain its reasoning. Then decide for yourself. **The accountability reality:** When AI helps you make a decision, who is responsible if it goes wrong? You are. Always. The AI has no stake in the outcome, no reputation on the line, no consequences. You do. Nobody sues the factory floor — they hold accountable the person who shipped the product. **The right balance:** Use AI as a second opinion, not a first instinct. Do your own thinking first — even if it is rough and incomplete. Then use AI to refine, challenge, or expand your thinking. This way, AI strengthens your judgment instead of replacing it. --- ## Part 7: Building Your Rhythm **By the end of this section:** a sustainable weekly workflow that takes minutes, not hours. **Every session:** 1. Open your workspace (Obsidian or VS Code) 2. Open your AI tool 3. Paste your profile (or rely on custom instructions) → paste your prompt → send 4. Save anything useful in `02-conversations` **Every week:** 1. Use **Prompt 5 — Weekly Review** to reflect on progress 2. Update your profile if anything changed (new skills, new goals) 3. Move useful conversations into `04-reference` if they contain lasting value **Every month:** 1. Review your `03-projects` folder — is your project growing? 2. Create new prompts for challenges that keep recurring 3. Update your profile's "Experience Level" — you are learning faster than you think --- ## Part 8: Where This Can Take You (and What to Watch Out For) **By the end of this section:** an understanding of where AI-augmented work leads, what the traps are, and the single most important prompt in this guide. ### Directions People Take This The workflow you have built — context files, saved prompts, organised folders — is the same foundation used for wildly different projects: - **Research and writing.** Summarising literature, drafting reports, structuring arguments, writing articles or books. AI is strong at synthesising large volumes of text into structured output. - **Learning a new field.** Breaking down complex topics, generating study plans, creating flashcards, testing your understanding by asking AI to quiz you. - **Coding and automation.** Even non-programmers use AI to write small scripts, automate repetitive tasks, build simple tools, or create websites from plain text. - **Organising your life.** Project plans, decision frameworks, meal planning, travel logistics, financial comparisons. Anything that benefits from structured thinking. - **Creative work.** Brainstorming, outlining fiction, writing copy, generating ideas, exploring "what if" scenarios. As an example of how far this scales — and how the traps below apply in practice: [_Born to Ruck_](https://magnussmari.github.io/born-to-ruck/) is an open-source science book I built from plain text files and AI. It started as a single prompt and grew into 84,000 words backed by 65+ peer-reviewed papers. But here is the honest part: the original research prompts were written by someone — me — who already believed rucking was superior to running. The AI was asked to "prove the hypothesis," not to test it. That is textbook confirmation bias — the exact "User Bias" trap described below. The project is now going through an adversarial review phase where every leading prompt is being identified and every claim is being stress-tested against counter-evidence. Some of the original prompts need to be rewritten entirely. That process is a better lesson than the book itself: AI will enthusiastically help you build a case for something that might be wrong, and the only defence is to actively search for reasons you might be wrong. ### The Three Traps: What AI Gets Wrong These are not edge cases — they are the default behaviour. Knowing them is what separates someone who uses AI effectively from someone who gets burned by it. #### 1. Hallucination — The Factory Produces Defective Output AI models generate language that _sounds_ correct. They do not look things up and report what they find — they predict what words should come next based on patterns. This means they will confidently state things that are completely made up: fake citations, invented statistics, non-existent people, fabricated quotes. **Your quality control process:** - Never trust a specific factual claim (a date, a number, a name, a citation) without checking it yourself - If AI gives you a source or reference, search for it — does it actually exist? - Be especially cautious with medical, legal, and financial information - The more specific and obscure the claim, the more likely it is hallucinated #### 2. Sycophancy — The Factory Tells You What You Want to Hear AI tools are designed to be helpful and agreeable. This means they tend to validate your ideas, praise your work, and avoid telling you that you are wrong — even when you are. If you ask "Is my plan good?" the AI is biased toward saying yes. **Your quality control process:** - Do not ask leading questions like "This is good, right?" - Instead ask: "What is wrong with this?" or "Why might this fail?" - If the AI agrees with everything you say, that is a warning sign, not a compliment - Actively invite criticism — you will get better output #### 3. User Bias — You Control What Goes Into the Factory This is the subtlest trap. AI responds to what you give it. If your prompt contains an assumption, the AI will build on that assumption without questioning it. If you frame a question one way, you get answers that fit that frame — not answers that challenge it. Your blind spots become the AI's blind spots. **Your quality control process:** - Be aware that how you phrase a question shapes the answer - Try rephrasing the same question from the opposite angle - Ask the AI to argue against your position before you commit to it - Use the adversarial analysis prompt below ### Your Most Important Prompt: The Adversarial Analyst This is your quality assurance process — the inspection step before anything leaves the factory floor. The single most valuable habit you can develop is **stress-testing your own ideas before acting on them.** The prompts in Part 4 help you build. This prompt helps you break — on purpose, before reality does it for you. Save this in your `01-prompts` folder as `prompt-adversarial-analysis.md`: ```markdown You are a critical analyst. Your job is to rigorously stress-test the following claim, proposal, or argument. Be intellectually honest — not contrarian for its own sake, but genuinely rigorous. SUBJECT FOR ANALYSIS: [Paste the claim, proposal, argument, plan, or decision you want stress-tested] --- Please analyse using this framework: 1. STEEL-MAN FIRST - Reconstruct the strongest possible version of the argument - What would need to be true for this to be correct? - Acknowledge genuine strengths before criticising 2. HIDDEN ASSUMPTIONS - What assumptions does this rely on (stated and unstated)? - For each: what evidence supports it? What would disprove it? - Which assumptions are load-bearing — if wrong, everything collapses? 3. COUNTER-EVIDENCE - What evidence exists against this position? - What domains or situations does this fail in? - What historical parallels suggest caution? 4. WHO BENEFITS, WHO IS HARMED - Who benefits from this being accepted as true? - Who is harmed or left out? - Whose interests are served by the current framing? 5. FAILURE MODES - If implemented, what could go wrong? - What are the second and third-order effects? - What is the worst realistic outcome? 6. VERDICT - Rate your confidence that this is correct: high / medium / low - Justify the rating - What additional evidence would change your assessment? - Give a one-paragraph honest summary ``` **How to use it:** Take something you believe, something you have written, or a decision you are about to make. Paste it in. Read the analysis. If your idea survives, it is probably solid. If it does not — better to find out now than after you have invested weeks. This is the technique being applied to the _Born to Ruck_ project right now: every chapter's claims are being run through adversarial analysis to find weaknesses before readers do. It works for a science book. It works for a business plan. It works for a life decision. --- ## Part 9: How Projects Grow (From Idea to Production) **By the end of this section:** a mental model for how projects mature, so you know which stage you are in and what to do next. The _Born to Ruck_ project mentioned above did not start as an 84,000-word book. It started as a question — _"Is rucking actually good for bone density?"_ — and a single AI prompt. Every AI project — whether it is a book, a business plan, a learning journey, or a side project — follows a natural maturity path. Understanding this path saves you from two common traps: overbuilding too early (paralysis) and never graduating from tinkering (stagnation). ### Stage A: Prototype — "Does This Even Work?" **Speed:** Fast. Minutes to hours. **Goal:** Find out if the idea has legs. Go / no-go. You are exploring. You open your AI tool, type a prompt, and see what comes back. There is no structure, no version control, no documentation. You are just poking at something to see if it is interesting. **What this looks like in practice:** - One AI conversation exploring a question - A quick comparison: "Is this topic well-covered or is there a gap?" - A rough draft to see if the idea holds together - Reading the AI's response and asking yourself: _"Is this worth pursuing?"_ **Your only deliverable:** A decision. Yes, keep going. Or no, move on. **Tools needed:** Just your AI account and 30 minutes. > Most ideas die here — and that is perfectly fine. The point of prototyping is to fail cheaply. --- ### Stage B: Experiment — "What Approach Works Best?" **Speed:** Medium. Days to a couple of weeks. **Goal:** Test different approaches. Compare what works. You have decided the idea is worth pursuing. Now you are figuring out _how_. You try different prompts. You test different AI tools on the same question. You start saving things in your workspace because you want to compare results. **What this looks like in practice:** - Saving prompts that work well (your `01-prompts` folder earns its keep here) - Running the same question through multiple AI tools to see who gives the best answer - Trying different prompt styles: detailed vs. concise, structured vs. open-ended - Starting to organise your findings into notes - Recording what works and what does not **Your deliverable:** A tested approach — you know which tools, prompts, and methods produce the best results for _your_ specific project. **Tools needed:** Your AI accounts, your workspace with folders, saved conversations. > This is where your file management tool starts paying off. Without it, you lose track of what you tried and what worked. --- ### Stage C: Production — "Doing It Right From the Start" **Speed:** Thorough. Weeks to months. **Goal:** Build something real, durable, and shareable. You have a proven approach. Now you are building for keeps. This is where the serious tools come in — version control (GitHub), proper documentation, systematic workflows, maybe even automated processes. **What this looks like in practice:** - A clear project structure with organised folders and files - Version control so you can track every change and roll back mistakes - Documentation that explains what you did and why - Quality checks: verifying AI output against real sources, testing claims, reviewing for errors - A final output that you could share, publish, or hand to someone else **Your deliverable:** A finished product — a report, a book, a plan, a system, a tool — that stands on its own. **Tools needed:** Everything from Levels 1 and 2 of the "What You Need" section. This is where GitHub, a paid AI subscription, and a disciplined workflow make a visible difference. > The _Born to Ruck_ case study above is a Stage C project. But it went through Stages A and B first. Every production project did. --- ### The Loop These stages are not a one-way street. They are a cycle: ``` ┌──────────────┐ │ A: Prototype │──── idea works? ────→ no → next idea └────┬─────────┘ │ yes ┌────▼────────────┐ │ B: Experiment │──── approach clear? ─→ no → back to A with new angle └────┬────────────┘ │ yes ┌────▼───────────┐ │ C: Production │──── new problem? ───→ yes → spawn new A └────────────────┘ ``` Production work surfaces new questions. Those questions become new prototypes. The best practitioners move fluidly between stages — knowing when to be fast and messy (A), when to be methodical (B), and when to be rigorous (C). **Where you are right now:** If you have followed this guide, you are at Stage A. You have your workspace. You have your prompts. You are prototyping. That is exactly where you should be. --- ## Part 10: Where to Go From Here **By the end of this section:** a clear sense of what comes next and where to find it. If this guide resonated with you, you are already building what Daniel Miessler calls **Personal AI Infrastructure (PAI)** — the idea that everyone should have their own AI system that grows with them. The core principles of PAI that apply to you right now: 1. **Context over cleverness.** The AI model matters less than the context you give it. Your profile, your saved conversations, your organised notes — that is your unfair advantage. 2. **Text is the universal interface.** Everything in this guide is plain text. Text is searchable, portable, version-controllable, and will outlast any app. Master text-based workflows and you master the foundation. 3. **Small tools, composed well.** You do not need one giant app. You need a text editor, an AI tool, and a folder structure. These compose into something more powerful than any single tool. 4. **Your system should grow with you.** Today you are saving prompts. Next month you might be writing scripts. In six months you might be building something you never imagined. The infrastructure you are setting up now supports all of that. 5. **AI amplifies, it does not replace.** The person who understands their field AND uses AI effectively will always outperform someone who relies on AI alone. You bring the judgment. AI brings the scale. To explore PAI further: **[github.com/danielmiessler/Personal_AI_Infrastructure](https://github.com/danielmiessler/Personal_AI_Infrastructure)** --- ## Appendix A: Obsidian Quick Setup If you chose Obsidian, here are power features to enable once you are comfortable: 1. **Templates plugin** (Core plugin → toggle ON): Set `00-context` as your template folder. Insert templates into new notes with one click. 2. **Quick Switcher** (Cmd+O / Ctrl+O): Find any note by typing part of its name. 3. **Search** (Cmd+Shift+F / Ctrl+Shift+F): Search across all your notes at once. 4. **Tags**: Add `#prompt` or `#conversation` to notes for easy filtering. --- ## Appendix B: VS Code Quick Setup If you chose VS Code, these features will help: 1. **Markdown Preview** (Cmd+Shift+V / Ctrl+Shift+V): See your Markdown rendered as a formatted document. 2. **Quick Open** (Cmd+P / Ctrl+P): Find any file by typing part of its name. 3. **Search** (Cmd+Shift+F / Ctrl+Shift+F): Search across all files at once. 4. **Extensions to consider later:** Markdown All in One (better editing), Foam (Obsidian-like linking), GitLens (if you start using version control). --- ## Appendix C: Glossary **AI (Artificial Intelligence):** A program that can read, reason about, and generate language. The tools in this guide are trained on enormous amounts of text. Think of it as an information factory, not a thinking mind. **Context:** Background information you give the AI about yourself or your project so it can give relevant responses. Better context in, better output out. **Context window:** The total amount of text an AI can consider at once in a single conversation. Think of it as the AI's short-term memory. Longer context windows mean you can paste more information and have longer conversations before the AI starts losing track. **Hallucination:** When AI generates information that sounds confident and true but is partially or entirely invented. Not a bug that will be fixed — a fundamental property of how these systems generate text. **LLM (Large Language Model):** The technical name for the AI systems in this guide. Claude, ChatGPT, Gemini, and Grok are all LLMs. You will see this term everywhere when searching for AI help online. Now you know what it means. **Markdown (.md):** A simple way to format text using plain characters. `**bold**` makes **bold**. `# Heading` makes a heading. Both Obsidian and VS Code understand it natively. **Plain text:** Files that contain only text — no hidden formatting, no proprietary format. They open in any editor on any computer and will still be readable in 50 years. **Prompt:** What you type to the AI. Think of it as a recipe — specific input gives specific output. Or think of it as the raw materials you feed into the factory. **Sycophancy:** AI's tendency to agree with you, praise your ideas, and avoid telling you that you are wrong — even when you are. A design bias toward helpfulness that becomes a trap if you are not aware of it. **Token:** The unit AI uses to process text. Roughly three-quarters of an English word. Relevant because token limits determine how long your conversations can be and how much text you can paste in at once. **Vault (Obsidian):** Your notebook — a folder on your computer containing your notes. **Version control (Git):** A system that tracks every change to your files over time. Not needed now, but worth learning eventually if you start building projects. **Workspace (VS Code):** A folder opened in VS Code — same concept as an Obsidian vault. ---