The Elements of Artificial Intelligence

In 2025, I started writing a series called The Elements of Artificial Intelligence. It’s Strunk and White’s The Elements of Style, but for using AI. This page holds all my original, more detailed blog posts, as well as an evolving summary of the main lessons and takeaways.

If you like what you read, check out my home page and newsletter for more on what all I do. brady-gerber-the-elements-of-artifical-intelligence

Original blog posts:

The Elements of Artificial Intelligence

How to use AI well, or at least not poorly.

by Brady Gerber

Author’s note: I originally wanted to call this guide “The Element of AI,” but then I realized that “Elements of AI” is the name of a popular course by the University of Helsinki, which I have not taken yet. This guide was created with the assistance of Claude, ChatGPT, and Cursor. Thanks to Sam Brewer, who twisted my arm into trying Claude and Cursor. You were right.

Table of Contents

The Core Guide

  • Foreword
  • Introduction
  • Getting Started
  • I: Elementary Rules of AI Usage
  • II: Elementary Principles of AI Interaction
  • III: A Few Matters of AI Form
  • IV: AI Concepts Commonly Misunderstood
  • V: An Approach to AI Philosophy

Bonus Chapters

  • VI: When Not to Use AI
  • VII: Case Studies
  • VIII: Quick Reference Guide
  • IX: AI for Software Development
  • X: Advanced Techiques for Experienced AI Users
  • XI: Visual Examples and Templates

Foreword

If you only want a single takeaway from this guide, my favorite rule is No. 10: “Be a Person.”

You—not the AI—will get blamed for using AI irresponsibly.

Consider this scenario.

You ask AI to write an email to your boss. It sounds professional, so you send it. Later, you realize the AI included a completely fabricated statistic. Who will the boss blame for including this fake data in an email and pressing “send”?

Hint: It won’t be the AI.

Introduction

AI is wonderful, and it sucks.

AI is wonderful. It’s easy and accessible. It removes gatekeeping for anyone to learn or practice skills such as writing or coding. Experienced software engineers can now generate a working React component, debug a complex SQL query, and draft a technical specification in minutes instead of hours. You’ve been using AI longer than you realize too: autocomplete, search rankings, spam filters, GPS routes, and so on. What’s new in AI is the interface—and the marketing.

AI also sucks. It’s biased. It can be clunky. It confidently generates broken code, hallucinates API endpoints that don’t exist, and introduces subtle bugs that pass code review. It’s trained on Stack Overflow answers from 2019 and has no concept of your specific architecture, constraints, or business logic. There’s also a fear that AI will sooner or later replace you. The fear is not unwarranted. When it comes to the bottom line, even the most sympathetic companies will choose “good enough” over “great” if it saves them money, and AI is often “good enough” compared to experienced—”great”—professionals. And yes, tools that sound smart still need a lot of supervision.

The problem is not just the technology. The problem is that most people, including experienced developers, treat AI like a search engine.

Regardless, whether you like it or not, you need AI fluency now more than ever.

This guide is for people like you and me: people who don’t like how we all of a sudden need AI fluency.

That doesn’t mean learning about AI has to be boring or hard.

The Inspiration For This Guide

As the title hints, The Elements of Artificial Intelligence is modeled after William Strunk Jr. and E.B. White’s The Elements of Style, that famous little book which you can literally fit into your back pocket and which teaches readers how to write well. That book still holds up. Just like The Elements of Style, this guide is meant to be short, referable, and, most importantly, useful.

Strunk and White cared deeply about clarity, simplicity, and principle. So do I.

The Goal of This Guide (and What it Can’t Teach)

AI moves fast. This book won’t cover everything, nor should it. The goal is not mastery. The goal is to get started. We’re aiming for mindfulness. The goal is to use AI thoughtfully, safely, and creatively, even if the idea of using AI “creatively” makes your skin crawl. We want to work skeptically, systematically, and with appropriate safeguards. Whether you’re a student, writer, developer, or someone who never wants to touch a line of code, this guide will help you become a more knowledgeable—and mindful—AI user.

I can’t guarantee that AI won’t replace you, even after you read this guide. Perhaps AI won’t save you that much time or effort either. I won’t pretend or assume this will be the first or last AI guide you’ll read. There are some good guides out there, too. Still, it’s worth studying and understanding something that is only going to become more integrated into our lives. If AI usage is only going to grow, we might as well learn to understand and use it.

In 2025, AI is Indeed Like an Intern

This is still a popular metaphor that my colleagues and I use to describe basic AI. AI is like a smart intern who drinks infinite amounts of coffee, never sleeps, and is so excited to help you do a task that it’ll sometimes forget to fact-check its work, or literally make stuff up just to impress you. You should make peace with this intern. You should be forgiving of this intern. And you should remember what an intern can’t and shouldn’t do.

But remember: You should still hire and work with human interns, so that the following generations can get real-world experiences and enter the work force and better help you. 🙂

It’s time to get to work.

Getting Started

Choose Your AI Tool

As of this writing, most people start with ChatGPT (by OpenAI) or Claude (by Anthropic). Both offer free versions that are perfect for learning:

  • ChatGPT: Go to chat.openai.com, create a free account. The free version (GPT-3.5 as of this writing) works well for most tasks.
  • Claude: Go to claude.ai, create a free account. Known for longer, more nuanced conversations.
  • Other options: Google’s Gemini, Microsoft’s Copilot (built into Edge browser)

Cost expectations: Free versions handle most beginner needs. Paid versions ($20/month typically) offer faster responses, access during busy times, and advanced features.

Your First Interaction

  1. Sign up for your chosen platform
  2. Find the chat box (usually prominently displayed)
  3. Start simple: Type something like “Hello! Tell me I’m pretty.”
  4. Try a practical task: “Help me write a polite email declining a meeting invitation.”

What to Expect

  • Response time: Usually 5-30 seconds, depending on how much you ask
  • Length: AI often writes more than you expect—you can ask it to be shorter
  • Accuracy: Good for general help, but always double-check important facts
  • Interface: Most AI tools work like text messaging—just type and press enter

I. Elementary Rules of AI Usage

1. Don’t Lie

Be transparent about your AI use. You’re probably not good at hiding it. You might as well tell the truth and disclose AI assistance appropriately for your context. That’s why I admit in the Foreward that I used Claude, ChatGPT, and Cursor to help me draft and edit all the sections of this guide. I technically didn’t have to do it, but I like how it clears up any confusion or assumptions right away, so we can move on.

At the very least, don’t actively deceive people about your AI use. Don’t lie. It’s not worth it.

2. Verify Outputs Before Sharing or Putting to Use

“Plausible” is not “factual.”

  • Plausible: The Berlin Wall fell in 1988.
  • Factual: The Berlin Wall fell in 1989.

How to Verify AI Outputs

Facts and Figures
  • Cross-reference with established sources (official websites, news outlets, etc)
  • Use search engines to double-check dates, statistics, and historical claims
  • Be especially careful with recent events (AI training has cutoff dates)
Professional Content
  • Have a colleague review AI-generated emails, reports, or presentations
  • Test AI-suggested solutions in low-risk situations first
  • Check AI’s work against your company’s style guide or standards
Code or Technical Instructions
  • Test code in a safe environment before deploying
  • Run through instructions step-by-step before following them
  • Consult official documentation for technical procedures and code style consistency
  • Check for deprecated APIs, security vulnerabilities, performance issues
  • Test edge cases the AI likely didn’t consider
Red Flags to Watch For
  • Overly confident statements about controversial topics
  • Statistics without sources
  • Technical instructions that seem too simple
  • Historical dates or facts that “sound about right” but you’re unsure of

3. Provide Clear, Specific Prompts with Relevant Context

Have you heard the data expression “garbage in, garbage out”? Vague prompts produce vague results. Give AI the background it needs to help you with what you actually need.

  • Bad: Where should I eat in Chicago?
  • Good: I’m visiting Chicago for the first time this summer and love Indian food. Any recommendations near public transit? Budget: $100 max per meal.

4. Use Literal Examples

Show, don’t tell.

  • Bad: Help me write this email in a conversational tone.
  • Good: Help me write this email in a conversational tone, like this: “Here’s the thing about remote work—it’s not for everyone, but when it works, it really works.”

5. Break Complex Requests into Steps

Large, complicated requests often produce unfocused results. Divide your work into manageable phases.

  • Bad: Create a complete marketing strategy for my new app.
  • Good: First, help me identify three key user personas for my productivity app. Then, help me develop messaging for each persona.

6. Specify Your Desired Format

Your AI isn’t going to know otherwise.

  • Bad: Summarize this report.
  • Good: Summarize this report in three bullet points, each focusing on each department’s performance.

7. Set Constraints

Even if you don’t know exactly what you want, knowing what you don’t want helps your AI stay foucsed.

  • Bad: Explain quantum computing.
  • Good: Explain quantum computing in exactly 200 words, using no technical jargon, with one concrete analogy, speaking as if you’re talking to a five year old.

8. Iterate, Refine

Good writing is good editing. It’s the same with AI. Never accept the first response if you’re not satisfied. Follow-up prompts are your friend.

Some helpful refinement prompts:

  • Make this more concise.
  • Read my tailored resume text as if you’re a harsh yet experienced HR professional looking for a reason to say no.
  • Rewrite this memo for an older audience.

9. Understand Your AI’s Limitations

Again, think of AI like a very well-read intern who sometimes gets overconfident.

What AI Does Well

  • Creative drafting: Writing copy, brainstorming ideas, generating variations
  • Pattern recognition: Analyzing text, summarizing content, identifying themes
  • Busy work: Formatting, basic translations, routine correspondence
  • Learning assistance: Explaining concepts, creating practice questions
  • Data organization: Structuring information, creating outlines, categorizing items

What AI Struggles With

  • Current events: Most AI systems don’t know what happened yesterday or even last month
  • Real-time information: Stock prices, weather, sports scores, breaking news
  • Personal/private information: It can’t access your email, files, or personal accounts (though you should assume that it can and will; see the below “When NOT to use AI” section)
  • Complex math: Can make calculation errors, especially with multi-step problems
  • Nuanced judgment: Understanding context that requires life experience
  • Fact-checking itself: It can’t browse the internet to verify its own claims

AI vs Search Engines

As much as AI companies would like to make you believe otherwise, AI chatbots are not a direct replacement of internet search engines.

  • Use AI for: Creative data-related tasks, busy-work assistance, brainstorming, pattern recognition, explaining concepts, analyzing provided text
  • Use search engines for: Current events, real-time information, specific facts, recent developments, verified statistics

Remember How AI is Trained

Most LLMs have training cutoffs. They might suggest using a JavaScript framework from 2022 when there’s a better option released in 2024. Be mindful of what is overrepresented in AI training (popular frameworks, common patterns, tutorial-style code) and what might be underrepresented (company-specific patterns, recent updates, advanced optimizations, edge cases).

10. Be a Person

You—not the AI—will get blamed for using AI irresponsibly.

II. Elementary Principles of AI Interaction

11. Start Conversations with Context, not Commands

Begin with background information to help AI understand your situation and goals.

  • Bad: Write a press release.
  • Good: I’m the marketing director for a 50-person software company. We’re announcing a new integration with a new software that will help our customers sync data more easily. I need a press release for tech industry publications.

12. Build Conversations Progressively

Each exchange should build upon the last. Think of AI conversations as collaborative, back-and-forth work sessions, rather than isolated requests. Reference previous parts of your conversation and build complexity gradually.

13. Ask for Explanations, Not Just Answers

Understanding beats memorization. When AI gives you a solution, ask it why it works.

  • Bad: What’s the best pricing strategy?
  • Good: What are three pricing strategies for a new SaaS product? What factors would determine which one to choose?

14. Use AI for Brainstorming, Humans for Decisions

Generate options, then use your smart brain to choose the best option.

As mentioned in Rule No. 10: You will get blamed if you use bad AI-generated content. You might as well do it right the first time.

15. Document your Successful Prompts

Build a personal prompt library that you can copy and paste. Save your future self time.

How to Build Your Library

  1. Save what works: When a prompt gives you great results, copy it to a document (I use the Stickies app on my Mac)
  2. Note the context: Write down what situation it worked well for
  3. Create templates: Replace specific details with [brackets] so you can reuse them
  4. Organize by purpose: Group prompts by task type (writing, analysis, creative, etc)
  5. Update regularly: Refine prompts that almost worked, delete ones that consistently fail

16. Test AI Outputs with Real Scenarios

Ask yourself: Would this actually work in my situation? If not, explain that to your AI and explain what you need instead.

  • Example: This email template you made assumes that we have a formal relationship with clients, but our company culture is very casual. Can you adjust the tone?

17. Request Multiple Perspectives

Intentionally ask for counterarguments and alternative viewpoints. This helps you better understand the full scope of what you’re working on.

  • Example: Present three different perspectives on implementing AI in customer service—from management who is worried about next quarter’s earnings, jaded employees who think AI is a fad, and customers who think AI results in sloppy results.

18. Challenge AI Responses

AI systems don’t have feelings that you can hurt. Push back on responses that seem incomplete or questionable.

  • Example: What evidence supports this? What are the potential downsides? What assumptions are you making? How would someone disagree with this?

19. Structure Complex Projects in Phases

Break large projects into discrete phases: planning, research, drafting, refining. This prevents AI from trying to do everything at once. It gives you more control over the process, and your answers.

20. Know When to Stop and Switch to Human Expertise

AI has limits. When you need specialized knowledge, sensitive judgment, or accountability for important decisions, consult human experts. Remember: if AI trains on work created by human experts, it’s better to go directly to the source when you can. This goes back to our core principle of Rule No. 10: You remain responsible for the output.

Common Beginner Mistakes

  1. Treating AI like Google: Asking for current information instead of help with thinking and writing
  2. Being too vague: “Help me with my project” instead of specific, contextual requests
  3. Accepting first answers: Not iterating or refining when the output isn’t quite right
  4. Over-relying: Using AI for everything instead of building complementary skills
  5. Under-explaining context: Forgetting that AI doesn’t know your situation, company, or personal preferences

III. A Few Matters of AI Form

Citation and Attribution Practices

The level of stated attribution should match the level of AI contribution and your professional context.

  • Light editing: This text was reviewed with AI assistance
  • Substantial generation: This content was developed in collaboration with AI, then reviewed and edited
  • Research assistance: AI tools were used to gather initial research, which was then verified and supplemented

Privacy and Data Handling

Assume Zero Privacy

Every prompt you send to AI services is potentially:

  • Vulnerable to data breaches
  • Stored indefinitely by the AI company
  • Used for model training (unless you’re on enterprise plans with specific guarantees)
  • Accessible to employees for quality assurance

Never share with AI the Following

  • Production credentials, API keys, passwords
  • Real customer data, PII, or sensitive business information
  • Proprietary algorithms or competitive intelligence
  • Security vulnerability details from your systems
  • Internal system architecture diagrams

Enterprise Considerations

  • Use AI services with BAAs (Business Associate Agreements) for healthcare
  • Implement AI usage policies for your team
  • Consider on-premise LLM deployments for sensitive work
  • Monitor AI service terms of service changes

Effective Prompt Structure

Not everything needs to be written in paragraph form. Structure can improve clarity.

Example A

  • CONTEXT: [Background information]
  • TASK: [Specific request]
  • FORMAT: [Output requirements]
  • EXAMPLES: [Sample inputs/outputs]

Example B

  • ROLE: Act as a [specific role]
  • GOAL: Help me [specific objective]
  • CONSTRAINTS: Keep it under [X] words, avoid [Y], focus on [Z]

Backup and Fallback Strategies

When AI tools fail or are unavailable, you should still be able to do your work. Don’t become dependent on any single AI system. Maintain your underlying skills.

Integration with Existing Tools

  • Most AI tools work alongside your regular software—copy and paste between AI chat and Word/Google Docs/email
  • Some tools offer browser extensions or integrations, but the basic copy-paste method works everywhere
  • Keep AI conversations open in a separate browser tab while working on your main task

Remember Rule No. 10?

IV. AI Concepts Commonly Misunderstood

Artificial General Intelligence (AGI)

True AGI—AI that matches or exceeds human intelligence across all domains—does not currently exist. Today’s AI systems, however sophisticated, are specialized tools with narrow capabilities. They excel in specific areas but lack general understanding.

Think of current AI like a brilliant specialist doctor who only knows one field extremely well. True AGI would be like a doctor who’s equally expert in surgery, psychiatry, pediatrics, and research—all at once. We don’t have that yet. Today’s AI systems are more like having many different specialist doctors, each very good at their particular area.

AI vs. Machine Learning vs. Deep Learning

AI is the broad field of creating intelligent systems. Machine learning is a subset of AI that learns patterns from data. Deep learning is a subset of machine learning using neural networks with multiple layers. Most current AI tools use deep learning techniques.

  • AI: The whole hospital (any system that acts intelligently)
  • Machine Learning: The medical training program (systems that learn from examples)
  • Deep Learning: Advanced specialist training (complex learning using neural networks)

Hallucination

When AI generates false information presented with confidence. Think of it like a friend who tells stories so confidently that you believe them, even when they’re completely wrong about the details. AI doesn’t know it’s making things up—it’s just predicting what sounds right based on patterns it learned.

Author’s note: People have been presenting false information with confidence for centuries. Remember what I say in the introduction: You should be forgiving of this intern—and verify their work.

Bias in AI Systems

AI systems reflect and can amplify biases present in their training data like cultural, demographic, professional, and ideological biases. Imagine if you learned everything about cooking from only Italian cookbooks—you’d give great Italian food advice but might not know much about Thai or Mexican cuisine. AI has similar blind spots based on what it learned from.

See previous entry’s author’s note.

Large Language Models (LLMs)

AI systems trained on vast amounts of text to understand and generate human language. They predict likely next words based on patterns in training data. They don’t truly “understand” language in the human sense—they’re very sophisticated pattern matching systems.

Think of LLMs like autocomplete on steroids. Your phone’s keyboard predicts your next word; LLMs predict the next word in much more sophisticated ways after reading billions of books, articles, and websites. They’re incredibly good at predicting what should come next in a conversation, but they don’t “understand” language the way humans do.

Tokens

The basic units AI systems use to process text, roughly equivalent to 0.75 words per token (so 100 tokens ≈ 75 words). Think of tokens like how your phone plan has a data limit. AI has a “token limit”—how much text it can think about at once. Hit the limit, and it starts forgetting the beginning of your conversation.

Temperature

Parameters that control randomness in AI outputs, typically ranging from 0.1 to 2.0. Think of it like a creativity dial:

  • Low temperature (0.1-0.3): Like a careful accountant—consistent, predictable, conservative
  • High temperature (1.5-2.0): Like an improvisational jazz musician—creative, unpredictable, sometimes brilliant, sometimes nonsensical

Context Window

The amount of text an AI system can consider simultaneously. Think of it like short-term memory. A person might remember the last few minutes of conversation clearly, but struggle to recall details from three hours ago. AI has a similar limitation, but measured in tokens rather than time.

V. An Approach to AI Philosophy

Maintain Intellectual Honesty and Integrity

Don’t claim AI work as entirely your own, but also don’t diminish your genuine contributions. Maintain professional standards and give appropriate credit based on your context and AI’s actual contribution.

Preserve Human Agency and Judgment

Keep humans in control of important decisions. AI should inform and support human decision-making, not replace it, especially for choices that affect other people or have significant consequences.

Stay Curious About How AI Works

You don’t need to understand the technical details, but having a basic grasp of AI’s capabilities and limitations makes you a more effective user. Knowledge improves usage.

Question AI Outputs Actively, not Passively

Develop a healthy skepticism. Don’t accept AI responses at face value—engage with them critically. Skepticism is a feature, not a bug.

Use AI to Amplify Human Creativity, not Replace It

AI works best when it enhances and expands human imagination rather than substituting for it. Use AI to explore more possibilities, not to avoid creative thinking.

Consider the Environmental Impact of Your AI Use

AI systems require significant computational resources and energy. Consider whether your AI use adds genuine value. More AI isn’t always better AI.

Respect Privacy Boundaries

Protect personal information and respect others’ data rights when using AI tools. Remember that AI systems may retain or learn from information you provide.

Adapt as AI Evolves

Stay flexible and continue learning as AI technology advances. What works today may not work tomorrow. New capabilities will emerge regularly.

Remember: AI Serves Human Purposes

We use technology to improve our lives and solve problems. We don’t live to serve technology. AI should make your work more effective, not more complicated.

Practice Digital Minimalism

Use AI when it adds genuine value. More AI isn’t always, or even usually, better AI.

Cultivate Human Skills AI Can’t Replicate

Focus on capabilities that work well alongside AI: emotional intelligence, creative judgment, interpersonal communication, ethical reasoning, and strategic thinking.

Author’s Note: Not to sound doomsday, but for the sake of your skillset and sanity, learn a skill or hobby that doesn’t require an Internet connection or screen.

Recognize AI’s Cultural and Perspective Limitations

AI systems reflect the perspectives and biases of their training data and creators. Be aware of whose voices are included and whose might be missing, especially when working on diverse or sensitive topics.

Use AI to Learn, not to Avoid Learning

AI should enhance your thinking and knowledge, not replace the need to understand your field. Use AI as a learning tool, not a crutch.

AI Ethics and Responsibility

Transparency: Be clear about AI use in your work and decision-making processes.

Accountability: Take responsibility for AI-generated content and decisions.

Fairness: Consider how AI outputs might affect different groups of people.

Privacy: Protect personal information and respect data rights.

Bias awareness: Actively look for and address biases in AI outputs.

Human oversight: Maintain human control over critical decisions.

VI. When NOT to Use AI

Privacy and Security Concerns

Assume everything you tell AI will be stored, analyzed, and potentially exposed.

Never share:

  • Personal identifying information
  • Financial data or account credentials
  • Proprietary business information
  • Confidential client data
  • Medical or legal information
  • Passwords or API keys

Real-Time Decision Making

AI is not suitable for:

  • Emergency situations requiring immediate action
  • Time-critical decisions with high stakes
  • Situations where you need to act faster than you can type

Situations Requiring Perfect Accuracy

AI is probabilistic, not deterministic. Don’t use it for:

  • Financial calculations that must be exact
  • Medical diagnoses or treatment decisions
  • Legal advice or contract interpretation
  • Safety-critical system design
  • Any task where 99% accuracy isn’t good enough

Creative Work That Must Be Uniquely Human

Remember Rule No. 10. Avoid AI for:

  • Personal artistic expression
  • Work that represents your unique voice or brand
  • Content that needs to be distinctly human
  • Projects where authenticity is more important than efficiency

When You Need to Learn, Not Just Get Answers

AI can be counterproductive when:

  • You’re trying to understand fundamental concepts
  • You need to develop your own problem-solving skills
  • The learning process is more valuable than the result
  • You’re building foundational knowledge in a field

When Human Judgment and Intuition Matter Most

AI struggles with:

  • Nuanced social situations
  • Context-dependent decisions
  • Situations requiring emotional intelligence
  • Tasks that benefit from human creativity and intuition

Remember

AI is a tool, not a replacement for thinking. Your judgment about when to use AI is as important as how you use it. Sometimes, the best use of AI is to not use it.

VII. Case Studies

A few examples of when using AI helps, and when using AI backfires.

Case Study 1: The Lawyer Who Trusted AI Too Much

What happened: A lawyer used ChatGPT to research legal precedents for a court filing. The AI generated six fake court cases with plausible-sounding names and citations.

The mistake: The lawyer didn’t verify the cases existed before submitting them to court.

The consequence: The judge discovered the fake cases, fined the lawyer $5,000, and the case was dismissed.

The lesson: Always verify AI-generated information, especially when accuracy is critical.

How to avoid this: Use AI for initial research, then verify all citations and facts through authoritative sources before using them in professional contexts.

Case Study 2: The Marketing Team’s AI Success

What happened: A marketing team used AI to generate 50 different email subject lines for an A/B test, then used AI to analyze which ones would perform best based on industry data.

The approach: They tested the AI predictions against real performance data and found 80% accuracy.

The result: Their email open rates increased by 35% using AI-optimized subject lines.

The lesson: AI excels at pattern recognition and optimization when you validate its suggestions.

Key success factors: They used AI for ideation and analysis but validated results with real data before implementation.

Case Study 3: The Student’s Plagiarism Problem

What happened: A student used AI to help write an essay, then submitted it as entirely their own work.

The mistake: The student didn’t understand that using AI assistance requires disclosure according to their school’s policy.

The consequence: The student received a failing grade and was placed on academic probation.

The lesson: Understand your organization’s AI usage policies and be transparent about AI assistance.

How to handle this properly: Use AI for brainstorming and structure, but write the final content yourself, and disclose any AI assistance according to institutional policies.

Case Study 4: The Developer’s Debugging Breakthrough

What happened: A developer spent three days debugging a complex bug, then asked AI to analyze the code with the error message.

The approach: The AI identified the issue in 30 seconds and provided a working solution.

The result: The developer learned a new debugging technique and saved hours of work.

The lesson: AI can accelerate problem-solving when you provide clear context and specific error information.

Best practices demonstrated: The developer provided the exact error message, relevant code context, and was open to learning from the AI’s analysis.

Case Study 5: The Content Creator’s AI Dilemma

What happened: A content creator used AI to generate blog post ideas and outlines, but found that the AI-generated content sounded generic and lacked their unique voice.

The problem: The creator was using AI for the entire writing process instead of just ideation and structure.

The solution: They used AI for research, outlines, and initial drafts, then rewrote everything in their own voice and style.

The result: Content production increased by 40% while maintaining the creator’s authentic voice and audience engagement.

The lesson: Use AI for the parts of your process where you struggle, but maintain control over the final creative output.

Case Study 6: The Business Analyst’s Data Disaster

What happened: A business analyst used AI to analyze customer data and generate insights, but the AI made incorrect assumptions about the data structure.

The mistake: The analyst didn’t verify the AI’s understanding of the data before acting on its recommendations.

The consequence: The company made poor strategic decisions based on flawed analysis, resulting in lost revenue.

The lesson: Always verify AI’s understanding of your data and assumptions before making business decisions.

How to prevent this: Provide clear data context, verify AI’s understanding with sample data, and cross-check AI insights with traditional analysis methods.

VIII. Quick Reference Guide

Beginners

  1. Start simple: “Help me write a professional email.”
  2. Be specific: Include context, constraints, and desired format
  3. Verify everything: Fact-check AI outputs before using them
  4. Be honest: Don’t hide your AI usage

Intermediate Users

  1. Build conversations: Use follow-up prompts to refine outputs
  2. Document success: Save effective prompts for reuse
  3. Test scenarios: Ask “Would this work in my situation?”
  4. Challenge assumptions: Ask for evidence and counterarguments

Advanced Users

  1. Chain prompts: Break complex tasks into sequential steps
  2. Use examples: Provide sample inputs/outputs for better results
  3. Manage context: Start fresh conversations for new topics
  4. Combine tools: Use multiple AI systems for different tasks

Red Flags to Watch For

  • Overconfidence: AI sounds certain about uncertain topics
  • Outdated information: Claims about current events or recent developments
  • Inconsistent logic: Contradictory statements within the same response
  • Generic advice: Vague suggestions that don’t apply to your situation
  • Missing context: Responses that ignore important details you provided

When to Stop and Ask a Human

  • High-stakes decisions: Financial, medical, legal, or safety matters
  • Personal judgment calls: Situations requiring emotional intelligence
  • Creative authenticity: Work that represents your unique voice
  • Learning fundamentals: When you need to understand, not just get answers
  • Privacy concerns: Any situation involving sensitive information

IX. AI for Software Development

Code Review and Analysis

AI is great at pattern recognition. Use it to review code for common issues, suggest improvements, and explain complex logic.

  • Bad: Review this code
  • Good: Review this Python function for performance issues, security vulnerabilities, and adherence to PEP 8 standards. Focus on the database query optimization.

Debugging Assistance

AI can help identify potential issues, but remember: it’s analyzing patterns, not actually running your code.

  • Bad: Why isn’t this working?
  • Good: I’m getting a ‘TypeError: ‘NoneType’ object is not iterable’ error in this function. Here’s the full traceback and the function code. What’s the most likely cause?

Documentation Generation

AI can help create clear, consistent documentation, but always review for accuracy and completeness.

  • Bad: Write documentation for this function
  • Good: Generate JSDoc documentation for this JavaScript function that handles user authentication, including parameter types, return values, and usage examples.

Testing Strategy Development

AI can suggest test cases and help structure your testing approach.

  • Bad: Write tests for this
  • Good: Suggest unit tests for this React component. Focus on edge cases for the form validation logic and user interaction scenarios.

Architecture Discussions

Use AI to explore different architectural patterns and trade-offs.

  • Bad: What’s the best architecture?
  • Good: Compare microservices vs. monolithic architecture for a social media app with 10,000 users. Consider scalability, maintenance, and development speed.

Natural Language Programming

AI can translate requirements into code, but always verify the output.

  • Bad: Create a login system
  • Good: Create a secure login system using Python Flask with JWT tokens, password hashing, and rate limiting. Include input validation and error handling.

Code Translation and Migration

AI can help convert code between languages or frameworks.

  • Bad: Convert this to JavaScript
  • Good: Convert this Python class to TypeScript, maintaining the same interface and adding proper type annotations.

Understanding Legacy Code

AI can help explain complex or poorly documented code.

  • Bad: What does this do?
  • Good: Explain this legacy JavaScript function step by step. It appears to handle data transformation but the variable names are unclear.

Performance Optimization

AI can suggest improvements, but always benchmark before and after.

  • Bad: Make this faster
  • Good: Analyze this SQL query for performance issues. The table has 1M rows and this query takes 5 seconds. Suggest indexing strategies and query optimizations.

Security Review

AI can identify common security patterns, but don’t rely on it for comprehensive security analysis.

  • Bad: Check for security issues
  • Good: Review this authentication code for common security vulnerabilities like SQL injection, XSS, and session management issues.

Remember for Developers

  • AI is a tool, not a replacement: Use it to augment your skills, not replace them
  • Always test AI-generated code: Never deploy without thorough testing
  • Understand the output: Don’t copy-paste without understanding what the code does
  • Keep learning: AI can help you learn new languages/frameworks, but build your own understanding
  • Version control everything: Track changes when using AI assistance
  • Review for best practices: AI might not follow your team’s coding standards

X. Advanced Techniques for Experienced AI Users

Chain-of-Thought Prompting

Encourage AI to show its reasoning process by asking it to think step-by-step.

  • Bad: Solve this math problem
  • Good: Solve this math problem step by step, showing your reasoning at each stage.

Few-Shot Learning

Provide examples to help AI understand the pattern you want.

  • Bad: Write a product description
  • Good: Write a product description in the style of these two following examples. Example 1: Revolutionary wireless headphones with 30-hour battery life and crystal-clear sound. Example 2: Premium coffee maker featuring smart brewing technology and customizable settings.

System Prompts vs. User Prompts

  • System prompts: Set the AI’s role and behavior (e.g., “You are a helpful coding assistant”)
  • User prompts: Your actual requests
  • Best practice: Use system prompts to establish context, user prompts for specific tasks

RAG (Retrieval-Augmented Generation)

Combine AI with external knowledge sources for more accurate, up-to-date responses.

  • Use case: When you need current information or domain-specific knowledge
  • Implementation: Provide relevant documents or data alongside your prompt

Prompt Chaining

Break complex tasks into a series of connected prompts.

  1. First prompt: Analyze the problem
  2. Second prompt: Generate solutions based on analysis
  3. Third prompt: Evaluate and refine solutions

Context Management

  • Be mindful of token limits: Longer conversations cost more and may lose coherence
  • Summarize when needed: “Summarize our conversation so far” to reset context
  • Start fresh for new topics: Don’t mix unrelated tasks in the same conversation

Error Handling in Prompts

Anticipate and handle potential AI mistakes:

  • Bad: Write code for user authentication
  • Good: Write code for user authentication. Include error handling for invalid inputs, and explain any security considerations.

Iterative Refinement

Use AI’s responses as starting points, not final products:

  1. Generate initial output
  2. Identify gaps or issues
  3. Ask for specific improvements
  4. Repeat until satisfied

Remember for Advanced Users

  • Experiment with parameters: Try different temperatures, max tokens, etc.
  • Keep a prompt library: Document what works for your use cases
  • Stay updated: AI capabilities evolve rapidly
  • Combine tools: Use multiple AI systems for different tasks
  • Validate outputs: Always verify AI-generated content, especially for critical applications

XI. Visual Examples and Templates

Writing Assistance

CONTEXT: [Your background/role]
TASK: [What you need written]
TONE: [Professional, casual, academic, etc.]
LENGTH: [Word count or format]
EXAMPLES: [Similar content you like]

Problem Solving

PROBLEM: [Describe the issue]
CONTEXT: [Relevant background]
CONSTRAINTS: [Limitations or requirements]
DESIRED OUTCOME: [What success looks like]

Analysis

TOPIC: [What to analyze]
PERSPECTIVE: [From whose viewpoint]
CRITERIA: [What factors to consider]
FORMAT: [How to present]


Thanks for reading. I write about music, tech, and AI for a lot of great outlets and for my own website. Subscribe to my newsletter to stay in the loop on everything I do.