Chain-of-thoughtWhat is chain-of-thought?A prompting technique where you ask the AI to show its reasoning step by step before giving a final answer, improving accuracy on complex problems. Ask AI for more prompting

What it is: Asking the AI to think step by step and show its reasoning.

Why it works: Breaking problems into steps reduces errors and helps you verify the logic.

Without chain-of-thought

A farmer has 17 sheep and all but 9 die. How many are left?

AI might answer: 8 (subtracting 9 from 17)

With chain-of-thought

A farmer has 17 sheep and all but 9 die. How many are left?

Think through this step by step before giving your answer.

AI reasoning:

Step 1: "All but 9 die" means 9 sheep did NOT die
Step 2: The question asks how many are LEFT (alive)
Step 3: Therefore, 9 sheep are left

When to use chain-of-thought

The "magic phrase"

These phrases trigger chain-of-thought reasoning:

• "Let's think step by step"
• "Explain your reasoning"
• "Walk through this systematically"
• "Before answering, consider..."
• "Break this down into steps"

Few-shot learning

What it is: Teaching the AI a pattern by showing examples instead of explaining rules.

Why it works: Examples are concrete and unambiguous. The AI learns the pattern from data, not descriptions.

Pattern matching with examples

Convert dates from MM/DD/YYYY to "Month Day, Year" format:

03/15/2023 → March 15, 2023
12/25/2022 → December 25, 2022
07/04/1776 → July 4, 1776

Now convert: 11/30/2024

The AI learns:

• First two digits are month number
• Middle two digits are day
• Last four digits are year
• Month names are spelled out
• Format is "Month Day, Year"

Few-shot for code style

Convert these function names to camelCase following the pattern:

get_user_data → getUserData
update_user_profile → updateUserProfile
delete_old_records → deleteOldRecords

Now convert these:
fetch_customer_orders
validate_email_address
send_notification_email

How many examples?

Role prompting

What it is: Assigning a specific persona or role to the AI.

Why it works: Different roles have different knowledge bases, vocabularies, and perspectives.

Role prompt structure

Act as a [specific role]. [Context about the situation].

[Your request]

Examples of effective roles

Role prompting in action

Without role:

Review this code.

With role:

You are a senior JavaScript developer with 10 years of experience.
You specialize in React performance optimization.

Review this code for:
1. Performance issues
2. React best practices
3. Potential bugs
4. Readability improvements

Be thorough but constructive in your feedback.

Combining roles with constraints

You are a code reviewer at a strict company that values:
- Type safety above all else
- Minimal dependencies
- Comprehensive error handling

Review this function with those priorities in mind.

System prompts

What it is: Instructions that apply to the entire conversation, not just one message.

Why it works: Sets persistent behavior so you don't repeat instructions.

System vs user prompts

System promptWhat is system prompt?Hidden instructions set by the developer that shape how an AI assistant behaves throughout a conversation. Users don't see it, but it defines the AI's persona and constraints. Ask AI for more (set once):

You are a helpful coding assistant. Always:
1. Provide code in markdown code blocks
2. Explain complex concepts simply
3. Suggest best practices
4. Ask clarifying questions when needed

User prompts (ongoing):

How do I reverse an array in JavaScript?

When to use system prompts

Use system prompts for:

• Persistent formatting requirements
• Safety guidelines
• Personality/tone settings
• Domain expertise ("You are an expert in...")

APIWhat is api?A set of rules that lets one program talk to another, usually over the internet, by sending requests and getting responses. Ask AI for more example (OpenAI)

response = client.chat.completions.create(
    model="your-model-id",  # check docs for current models
    messages=[
        {"role": "system", "content": "You are a concise technical assistant. Always provide code examples."},
        {"role": "user", "content": "Explain closures in JavaScript"}
    ]
)

Web interface workaround

ChatGPT and Claude don't have a "system prompt" field, but you can simulate it:

For this conversation, I want you to act as a [role]. 
Always [behavior 1], [behavior 2], and [behavior 3].

Understood? (Wait for confirmation)

[Then proceed with your actual questions]

Iterative refinement

What it is: Treating prompting as a conversation, improving results through feedback.

Why it works: The AI can learn from your feedback and adjust its approach.

The iterative process

Step 1: Initial prompt
"Write a function to validate email addresses"

Step 2: Review and give feedback
"Good start, but I need it to also check that the domain 
 actually exists. Can you add that?"

Step 3: Further refinement
"Perfect. Now can you make it return specific error messages
 instead of just true/false?"

Step 4: Final polish
"Great! Can you add TypeScript types and JSDoc comments?"

Effective feedback phrases

When to start over vs iterate

Iterate when:

• The response is 70% of what you need
• You can pinpoint specific issues
• You want to explore variations

Start over when:

• The response is completely off
• You forgot crucial context
• You need a fundamentally different approach

Prompt chaining

What it is: Breaking a complex task into a sequence of simpler prompts, using the output of one as input to the next.

Why it works: Each step has focused context, reducing confusion and improving accuracy.

Example: Building a feature

Chain Step 1: Planning

I need to add user authentication to my React app.

Requirements:
- Login with email/password
- JWT token storage
- Protected routes

Create a plan with:
1. Components needed
2. State management approach
3. API integration points

Chain Step 2: Generate types/interfaces

Based on that plan, generate TypeScript interfaces for:
- User data
- Auth context
- API responses

Chain Step 3: Build the auth context

Now create the AuthContext using those types.
Include login, logout, and auth state.

Chain Step 4: Create the login form

Create a LoginForm component that uses AuthContext.
Include form validation with React Hook Form.

Benefits of prompt chaining

1. Better accuracy: Each step is simpler and clearer
2. Easier debugging: You can see where things go wrong
3. Human oversight: Review each step before continuing
4. Reusability: Save useful intermediate outputs

Prompt chaining patterns

Quick reference: technique selection

These advanced techniques aren't just tricks, they're ways of working with AI more effectively. Chain-of-thoughtWhat is chain-of-thought?A prompting technique where you ask the AI to show its reasoning step by step before giving a final answer, improving accuracy on complex problems. Ask AI for more improves accuracy, few-shot learning teaches patterns, roles bring expertise, and chaining manages complexity. Combine them as needed. The best prompters mix and match techniques based on the task at hand.