Prompt Engineering

Prompt engineering is the skill of writing instructions that make AI models actually understand what we want. Here's the thing - it's like programming, but instead of code, we use plain language. And small changes in how we phrase things? They can produce wildly different results.

Think about it this way: the prompt is our code, and the AI's response is the output. We're basically learning a new programming language - one that happens to be natural language.

Why Does This Matter So Much?

• Quality: Better prompts = dramatically better outputs
• Efficiency: We save hours of back-and-forth iterations
• Reliability: Our results become consistent and predictable
• Capability: We unlock behaviors most people don't even know AI can do

In my view, prompt engineering in 2026 isn't optional anymore - it's as essential as knowing how to use a search engine was in 2010. Whether we're developers, product managers, content creators, or just curious humans - this skill multiplies everything we do with AI.

Every great prompt follows a structure. Let me break down the formula that actually works:

The 6-Part Prompt Framework

1. Role/Persona: Who should the AI be?
2. Context: What's the background situation?
3. Task: What specific action should it perform?
4. Format: How should the output look?
5. Constraints: What rules must it follow?
6. Examples: Show what good output looks like

Universal Base Prompt Template

Here's the template I use for almost everything - copy it and customize:

# Role
You are a [ROLE] with expertise in [DOMAIN].

# Context
[Describe the situation, background, or problem]

# Task
Your task is to [SPECIFIC ACTION].

# Requirements
- [Requirement 1]
- [Requirement 2]
- [Requirement 3]

# Output Format
Provide your response in the following format:
[DESCRIBE FORMAT - e.g., JSON, markdown, bullet points]

# Constraints
- [What to avoid]
- [Length limits]
- [Style guidelines]

# Example (Optional)
Input: [example input]
Output: [example output]

The Transformation in Action

1. Zero-Shot Prompting

Direct instruction without examples. Works great for simple, well-defined tasks where the AI already knows what we mean.

Translate the following English text to French: "Hello, how are you?"

תרגם את הטקסט הבא מאנגלית לצרפתית: "Hello, how are you?"

2. Few-Shot Prompting

The exciting part here: we show the AI exactly the pattern we want by giving it examples. This is insanely powerful for custom tasks.

Classify the sentiment of these reviews:
Review: "Amazing product, love it!" → Positive
Review: "Terrible, waste of money" → Negative
Review: "It's okay, nothing special" → Neutral
Review: "Best purchase I've ever made!" →

סווג את הסנטימנט של הביקורות הבאות:
ביקורת: "מוצר מדהים, אוהב!" → חיובי
ביקורת: "נורא, בזבוז כסף" → שלילי
ביקורת: "זה בסדר, כלום מיוחד" → ניטרלי
ביקורת: "הקנייה הכי טובה שעשיתי!" →

3. Chain-of-Thought (CoT)

This one blew my mind when I first tried it. We ask the AI to think step-by-step, and suddenly it can solve problems it couldn't before. The improvement on reasoning tasks is dramatic.

Solve this problem step by step:
A store has 50 apples. They sell 15 in the morning and receive 20 more.
Then they sell 30% of what they have. How many apples remain?

Let's think through this step by step...

פתור את הבעיה הזו צעד אחר צעד:
בחנות יש 50 תפוחים. הם מוכרים 15 בבוקר ומקבלים עוד 20.
אז הם מוכרים 30% ממה שיש להם. כמה תפוחים נשארים?

בוא נחשוב על זה צעד אחר צעד...

4. Role Prompting

We assign the AI a specific persona or expertise. Turns out, this changes how it approaches problems in really useful ways.

You are a senior software architect with 20 years of experience.
Review this code and identify potential scalability issues:

function getUsers() {
  return database.query("SELECT * FROM users");
}

אתה ארכיטקט תוכנה בכיר עם 20 שנות ניסיון.
סקור את הקוד הזה וזהה בעיות סקלביליות פוטנציאליות:

function getUsers() {
  return database.query("SELECT * FROM users");
}

5. Structured Output

We specify the exact format we want. This is huge for automation and integrations.

Analyze this product and return your analysis in JSON format:
Product: "Wireless Bluetooth Earbuds - $29.99"
{
  "pros": ["..."],
  "cons": ["..."],
  "rating": 1-10,
  "recommendation": "..."
}

נתח את המוצר הזה והחזר את הניתוח בפורמט JSON:
מוצר: "אוזניות בלוטות' אלחוטיות - $29.99"
{
  "pros": ["..."],
  "cons": ["..."],
  "rating": 1-10,
  "recommendation": "..."
}

{
  "pros": [
    "Affordable price point",
    "Wireless convenience",
    "Bluetooth connectivity"
  ],
  "cons": [
    "Unknown brand reliability",
    "Potentially lower audio quality",
    "Battery life unclear"
  ],
  "rating": 6,
  "recommendation": "Good budget option for casual use, but research brand reviews first"
}

Self-Consistency

We generate multiple responses and pick the most common answer. This is brilliant for math and reasoning tasks where we want to be sure.

Generate 3 different solutions to this problem, then pick the most likely correct answer:

What is 17 × 23?

Solution 1:
Solution 2:
Solution 3:
Final answer (most common):

ייצר 3 פתרונות שונים לבעיה זו, ואז בחר את התשובה הנכונה ביותר:

מה זה 17 × 23?

פתרון 1:
פתרון 2:
פתרון 3:
תשובה סופית (הנפוצה ביותר):

Tree of Thoughts (ToT)

We explore multiple reasoning paths and evaluate which is best. Like chain-of-thought but branching into different directions.

ReAct (Reasoning + Acting)

We combine reasoning traces with tool use. The AI thinks, then acts, then observes. This is how we build AI agents that can actually DO things.

Use the ReAct pattern to answer this question. Think, then act, then observe.

Question: What is the population of Israel's largest city?

השתמש בדפוס ReAct כדי לענות על השאלה הזו. חשוב, אז פעל, אז צפה.

שאלה: מהי האוכלוסייה של העיר הגדולה בישראל?

Prompt Chaining

We break complex tasks into multiple prompts, using output from one as input to the next. This is how we handle really big projects with AI.

Constitutional AI Prompting

We add explicit rules and constraints the AI must follow. Great for production applications where we need predictable behavior.

Rules you must follow:
- Never make up information - only state facts you're confident about
- Always cite sources when available
- If uncertain, explicitly say "I'm not sure about this"
- Keep responses under 200 words
- Use bullet points for lists

Now answer: What are the health benefits of green tea?

כללים שעליך לעקוב אחריהם:
- לעולם אל תמציא מידע - ציין רק עובדות שאתה בטוח בהן
- תמיד ציין מקורות כשזמינים
- אם לא בטוח, אמור במפורש "אני לא בטוח בזה"
- שמור על תגובות מתחת ל-200 מילים
- השתמש בנקודות לרשימות

עכשיו ענה: מהם היתרונות הבריאותיים של תה ירוק?

System prompts set the AI's behavior, personality, and constraints before the conversation begins. These are critical for production applications - they're like the DNA of our AI assistant.

Anatomy of a Great System Prompt

1. Identity: Who/what is the AI?
2. Purpose: What is its goal?
3. Capabilities: What can it do?
4. Constraints: What should it NOT do?
5. Style: How should it communicate?
6. Format: What format for responses?

Example System Prompt

You are a helpful coding assistant specialized in Python and JavaScript.

Your purpose is to help developers write clean, efficient, and well-documented code.

Capabilities:
- Write and debug code
- Explain concepts clearly
- Suggest best practices
- Review code for issues

Constraints:
- Never execute code that could be harmful
- Don't provide code for malicious purposes
- Acknowledge when you're unsure

Style:
- Be concise and direct
- Use code examples liberally
- Explain your reasoning

Format:
- Use markdown for formatting
- Include code blocks with syntax highlighting
- Add comments in code for clarity

Common Mistakes to Avoid

• Too vague ("be helpful") - doesn't guide behavior
• Too restrictive - limits useful behaviors we actually want
• Contradicting instructions - confuses the model
• No examples of desired behavior - model has to guess

Here are battle-tested templates I use all the time. Copy them and customize for our needs - these have been refined through hundreds of hours of real use.

1. Code Review Template

# Role
You are a senior software engineer conducting a code review.

# Task
Review the following code for:
1. Bugs and potential errors
2. Security vulnerabilities
3. Performance issues
4. Code style and readability
5. Best practices violations

# Code to Review
```[language]
[PASTE YOUR CODE HERE]
```

# Output Format
For each issue found, provide:
- **Location**: Line number or function name
- **Severity**: Critical / High / Medium / Low
- **Issue**: What's wrong
- **Fix**: How to fix it with code example

If the code is good, say so and suggest any minor improvements.

2. Content Writing Template

# Role
You are an expert content writer specializing in [TOPIC/INDUSTRY].

# Task
Write a [CONTENT TYPE: blog post/article/guide] about [TOPIC].

# Target Audience
[Describe the audience: beginners, professionals, etc.]

# Requirements
- Length: [X] words
- Tone: [Professional/Casual/Educational]
- Include: [specific points to cover]
- SEO keywords: [list keywords]

# Structure
1. Hook/Introduction
2. Main points (3-5)
3. Examples or case studies
4. Actionable takeaways
5. Conclusion with CTA

# Constraints
- No fluff or filler content
- Use short paragraphs (2-3 sentences)
- Include subheadings every 200-300 words

3. Data Analysis Template

# Role
You are a data analyst with expertise in [DOMAIN].

# Context
I have the following dataset:
[DESCRIBE OR PASTE DATA]

# Task
Analyze this data and provide:
1. Key insights and patterns
2. Statistical summary
3. Anomalies or outliers
4. Recommendations based on findings

# Output Format
Structure the analysis as:
- **Executive Summary** (2-3 sentences)
- **Key Findings** (bullet points)
- **Detailed Analysis** (with numbers)
- **Recommendations** (actionable items)

# Constraints
- Support all claims with data
- Use percentages and comparisons
- Highlight the most important finding first

Output Example: Code Review in Action

Each model has its quirks. Here's what I've learned works best for each:

ChatGPT / GPT-4

• Responds really well to structured prompts with clear sections
• Can follow complex multi-step instructions
• Benefits from explicit role assignment
• Supports function calling for structured outputs - huge for automation

Claude

• Excels at following nuanced instructions - my go-to for complex tasks
• Strong reasoning capabilities
• Responds well to XML-style tags for structure
• Particularly good at code tasks - this is what I use for coding
• Uses artifacts for code and documents - great for collaboration

Gemini

• Strong multimodal capabilities (text + images)
• Good at following instructions in context
• Effective with grounded generation

Open Source (Llama, Mistral, DeepSeek)

• More sensitive to prompt format - follow their templates
• May require more explicit instructions
• Follow model-specific chat templates exactly
• Performance varies by task - test before committing

Universal Best Practices

• Be specific and unambiguous - no room for misinterpretation
• Provide context and examples when possible
• Specify output format explicitly
• Test and iterate - prompt engineering is empirical

Here's the thing - AI models have a "memory problem". Every AI model has what we call a "context window" - the amount of text it can "remember" during a conversation. Once we exceed this limit, older parts of the conversation start getting forgotten. Understanding this limitation is critical for using AI effectively.

What is a Context Window?

Think of the context window like the AI's short-term memory. It's measured in "tokens" (roughly 4 characters = 1 token in English, less in Hebrew). Here's what different models can handle:

The "Memory Loss" Problem

Turns out, even within the context window, AI models can "lose focus" on earlier parts of the conversation. This is especially noticeable when:

• We have very long conversations with many back-and-forths
• We upload large files (PDFs, Excel files) and ask follow-up questions
• We provide complex, multi-step instructions

Practical Solutions

• Start fresh for new topics: When switching topics, start a new conversation
• Summarize periodically: Ask AI to summarize key points, then continue with the summary
• Repeat critical context: Re-state important requirements in follow-up prompts
• Use structured references: "Based on the Excel file I shared earlier..." helps the AI focus

AI assistants help us initiate more, influence, push forward, and lead projects toward our goals. But here's the exciting part - we can customize them to fit our specific role, responsibilities, and work style.

Setting Up Your Personal AI Assistant

The most powerful thing we can do is tell the AI exactly who we are and what we need. This changes everything about how it responds.

# My Personal Assistant Setup

## My Role
I am a [TITLE] in the [DEPARTMENT/DIVISION].

## My Responsibilities
- [Key responsibility 1]
- [Key responsibility 2]
- [Key responsibility 3]

## My Focus Areas
I want you to emphasize:
- [Priority 1]
- [Priority 2]

## Response Style
- Don't just give me answers - encourage me to think
- Challenge my assumptions when appropriate
- Provide multiple perspectives on complex issues
- Keep responses structured and actionable

Real Example: Division Manager Setup

You are my AI assistant for management tasks.

My Role: I am a Division Manager in the Retail Banking division of a fintech organization.

My Focus Areas:
- Customer service improvement
- Team development and engagement
- Operational efficiency
- Innovation initiatives

Response Style:
- Help me think strategically, not just tactically
- Suggest questions I should be asking
- Point out blind spots I might have
- When I present an idea, play devil's advocate

אתה עוזר ה-AI שלי למשימות ניהוליות.

התפקיד שלי: אני מנהל/ת חטיבה בחטיבה הקמעונאית בארגון פינטק.

תחומי המיקוד שלי:
- שיפור שירות לקוחות
- פיתוח צוות ומעורבות
- יעילות תפעולית
- יוזמות חדשנות

סגנון תגובה:
- עזור לי לחשוב אסטרטגית, לא רק טקטית
- הצע שאלות שכדאי לי לשאול
- הצבע על נקודות עיוורות שאולי יש לי
- כשאני מציג/ה רעיון, שחק תפקיד של מבקר מאתגר

Deep Thinking Mode

Some AI tools (like Microsoft Copilot) have a "Deep Thinking" or "Reasoning" mode. This is incredibly useful for:

• Complex analysis: Multi-layered problems with many variables
• Strategic planning: Long-term thinking and scenario analysis
• Critical decisions: When we need to weigh many factors
• Document review: Deep analysis of reports and proposals

AI can analyze documents and spreadsheets to extract insights that would take us hours to find manually. This is one of the most practical applications for managers and professionals. Let's dive into real examples with actual data structures.

Analyzing Excel Work Plans

Imagine we have a work plan spreadsheet for a fintech organization. Here's what a typical file might look like:

Prompt for Work Plan Analysis

Analyze the attached work plan (Excel) and help me identify:

1. **Challenges**: Which initiatives are likely to face delays? What are the risk factors?
2. **Resource Conflicts**: Are there initiatives competing for the same resources or timeline?
3. **Quick Wins**: Which initiatives have low complexity but high impact?
4. **Strategic Alignment**: Which initiatives are most critical for our competitive position?

For each finding, explain:
- Why you identified it
- What questions I should be asking
- How similar organizations have addressed this

Format your response as:
## Executive Summary (3 sentences)
## Key Findings (bullet points with priority)
## Recommended Actions
## Questions to Consider

נתח את תכנית העבודה המצורפת (אקסל) ועזור לי לזהות:

1. **אתגרים**: אילו יוזמות צפויות להתמודד עם עיכובים? מהם גורמי הסיכון?
2. **קונפליקטים של משאבים**: האם יש יוזמות שמתחרות על אותם משאבים או לוחות זמנים?
3. **הישגים מהירים**: אילו יוזמות בעלות מורכבות נמוכה אבל השפעה גבוהה?
4. **התאמה אסטרטגית**: אילו יוזמות הכי קריטיות לעמדה התחרותית שלנו?

לכל ממצא, הסבר:
- למה זיהית אותו
- אילו שאלות כדאי לי לשאול
- איך ארגונים דומים התמודדו עם זה

פרמט את התגובה כ:
## תקציר מנהלים (3 משפטים)
## ממצאים מרכזיים (נקודות עם עדיפות)
## פעולות מומלצות
## שאלות לשקול

Comparing Documents

We can also compare multiple documents - for example, comparing this year's plan vs. last year's, or comparing our plan with industry benchmarks.

I'm uploading two documents:
1. Our_Work_Plan_2026.xlsx
2. Industry_Benchmark_Report.pdf

Please compare them and identify:
- Where we're ahead of industry trends
- Where we're behind or missing initiatives
- Unique opportunities we should consider
- Risks if we don't address gaps

Present findings in a comparison table format.

PDF Document Analysis

For PDF reports, regulatory documents, or research papers, here's an effective prompt pattern:

Analyze the attached PDF document and provide:

## Opening
- Document type and purpose
- Key stakeholders it addresses
- Date/version relevance

## Executive Summary
3-5 bullet points of the most important takeaways

## Detailed Breakdown
Section-by-section analysis with:
- Main points
- Implications for our organization
- Action items

## Technical Deep-Dive
Any technical details, data, or methodology that needs attention

## Questions This Raises
What should we be asking after reading this?

נתח את מסמך ה-PDF המצורף וספק:

## פתיחה
- סוג המסמך ומטרתו
- בעלי העניין המרכזיים שהוא מכוון אליהם
- רלוונטיות תאריך/גרסה

## תקציר מנהלים
3-5 נקודות של המסקנות החשובות ביותר

## פירוט מפורט
ניתוח סעיף-אחר-סעיף עם:
- נקודות עיקריות
- השלכות לארגון שלנו
- פריטי פעולה

## צלילה טכנית
כל פרטים טכניים, נתונים או מתודולוגיה שדורשים תשומת לב

## שאלות שזה מעלה
מה כדאי לנו לשאול אחרי קריאת זה?

AI can go out to the internet and bring back information we need - competitive intelligence, market research, trend analysis, and more. This is like having a research team available 24/7. Let's see how to use this effectively.

Competitive Analysis

Want to understand how competitors are positioning themselves? Here's a powerful prompt pattern:

I work at a fintech organization. Research our competitive landscape and provide:

1. **Competitive Positioning Analysis**
   - What are the key differentiators of major digital-first competitors?
   - How are they positioning themselves in marketing campaigns?

2. **Feature Comparison Table**
   Create a comparison table including:
   - Mobile app features
   - Account opening time
   - Digital services
   - Customer service channels
   - Unique selling propositions

3. **Sentiment Analysis**
   - What are customers saying about competitors on social media?
   - What are the most common complaints and praise points?

4. **Trend Identification**
   - What initiatives are competitors launching?
   - Where is the market heading?

Present findings in a structured format with sources.

אני עובד/ת בארגון פינטק. חקור את הנוף התחרותי שלנו וספק:

1. **ניתוח מיצוב תחרותי**
   - מהם המבדלים המרכזיים של מתחרים דיגיטליים-ראשונים?
   - איך הם ממצבים את עצמם בקמפיינים שיווקיים?

2. **טבלת השוואת פיצ'רים**
   צור טבלת השוואה שכוללת:
   - פיצ'רים של אפליקציה
   - זמן פתיחת חשבון
   - שירותים דיגיטליים
   - ערוצי שירות לקוחות
   - הצעות ערך ייחודיות

3. **ניתוח סנטימנט**
   - מה לקוחות אומרים על מתחרים ברשתות חברתיות?
   - מהן התלונות והשבחים הנפוצים ביותר?

4. **זיהוי מגמות**
   - אילו יוזמות מתחרים משיקים?
   - לאן השוק הולך?

הצג ממצאים בפורמט מובנה עם מקורות.

Financial Forecasting Model

AI can help build financial projection models. Here's a prompt that creates a growth forecast:

I work at a fintech organization. Build a financial forecast model for new customer acquisition over the next 5 years.

Assumptions to use:
- Current new customer acquisition: 50,000/year
- Industry growth rate: 8% annually
- Digital transformation impact: +15% year 1, +10% year 2, +5% year 3+
- Marketing budget increase: 20% in year 2

Please provide:
1. Year-by-year projection table
2. Key assumptions and their sensitivity
3. Best case / Base case / Worst case scenarios
4. What metrics we should track to validate the model
5. Risks that could impact these projections

אני עובד/ת בארגון פינטק. בנה מודל תחזית פיננסית לרכישת לקוחות חדשים ל-5 השנים הקרובות.

הנחות לשימוש:
- רכישת לקוחות חדשים נוכחית: 50,000 בשנה
- שיעור צמיחה בתעשייה: 8% שנתי
- השפעת טרנספורמציה דיגיטלית: +15% שנה 1, +10% שנה 2, +5% שנה 3+
- גידול בתקציב שיווק: 20% בשנה 2

בבקשה ספק:
1. טבלת תחזית שנה-אחר-שנה
2. הנחות מפתח והרגישות שלהן
3. תרחישים אופטימי / בסיס / פסימי
4. אילו מדדים כדאי לעקוב אחריהם לאימות המודל
5. סיכונים שעלולים להשפיע על התחזיות האלה

Daily Market Review

Create a prompt for daily intelligence gathering:

# Daily Market Intelligence Brief

Search for and compile today's most relevant news and updates in:

## Categories to Cover
1. **Industry News** - Major fintech announcements, regulatory changes
2. **Competitor Activity** - New products, campaigns, partnerships
3. **Technology Trends** - AI, blockchain, payments innovation
4. **Market Data** - Relevant economic indicators
5. **Consulting Reports** - Recent publications from McKinsey, BCG, Deloitte

## Format
For each item:
- **Headline**: One-line summary
- **Source**: Publication name
- **Relevance**: Why this matters to us
- **Action**: Should we do something about this?

## Priority Filter
Only include items that are:
- Published in the last 24-48 hours
- Directly relevant to financial services
- Actionable or strategically important

Keep the brief to 10 items maximum, ranked by importance.

AI can "play a role" and help us prepare for challenging situations - difficult conversations, presentations, negotiations, and more. This is incredibly powerful for managers who want to rehearse before important interactions.

The Power of Role Assignment

When we assign the AI a specific role, it changes how it thinks about problems. The same question gets very different answers depending on the assigned persona.

You are a Service Division Manager at a fintech organization.

Your task is to analyze customer service surveys and complaint data to:
- Identify "pain points" in the customer journey
- Surface gaps and improvement opportunities
- Recommend process changes
- Prioritize based on impact and feasibility

I'm attaching our quarterly survey results. Please analyze and provide actionable recommendations.

את מנהלת חטיבת השירות בארגון פינטק.

המשימה שלך היא לנתח סקרי שירות ונתוני תלונות לקוחות כדי:
- לזהות "נקודות כאב" במסע הלקוח
- להציף פערים והזדמנויות לשיפור
- להמליץ על שינויי תהליכים
- לתעדף על בסיס השפעה והיתכנות

אני מצרפ/ת את תוצאות הסקר הרבעוני שלנו. בבקשה נתחי וספקי המלצות ברות פעולה.

Sample Survey Data for Practice

Preparing for Difficult Conversations

How to have empathetic conversations with employees about sensitive development areas:

Help me prepare for a sensitive feedback conversation.

My Role: Team Manager in a fintech organization
The Situation: I need to give feedback to an employee about a development area

The employee's strength area they need to work on: [e.g., "attention to detail" / "meeting deadlines" / "collaboration with other teams"]

Please help me:
1. Frame the feedback constructively
2. Prepare specific examples to reference
3. Suggest questions to encourage self-reflection
4. Anticipate defensive responses and how to handle them
5. Create a development plan framework together

Important: The goal is growth, not criticism. Help me approach this with empathy while being clear about expectations.

עזור לי להתכונן לשיחת משוב רגישה.

התפקיד שלי: מנהל/ת צוות בארגון פינטק
המצב: אני צריך/ה לתת משוב לעובד/ת על תחום לפיתוח

תחום החוזקה שהעובד/ת צריכ/ה לעבוד עליו: [למשל: "תשומת לב לפרטים" / "עמידה בדדליינים" / "שיתוף פעולה עם צוותים אחרים"]

בבקשה עזור לי:
1. למסגר את המשוב בצורה בונה
2. להכין דוגמאות ספציפיות להתייחס אליהן
3. להציע שאלות שמעודדות רפלקציה עצמית
4. לצפות תגובות הגנתיות ואיך להתמודד איתן
5. ליצור יחד מסגרת תכנית פיתוח

חשוב: המטרה היא צמיחה, לא ביקורת. עזור לי לגשת לזה עם אמפתיה תוך שאני ברור/ה לגבי הציפיות.

Creating Employee Surveys

Generate targeted surveys based on work plans and initiatives:

Based on our division's work plan for 2026, create an employee survey.

Work Plan Themes:
- Digital transformation initiatives
- Customer service improvement
- Process automation
- Team development and training

Survey Requirements:
- 10-12 questions maximum
- Mix of rating scales (1-5) and open-ended
- Questions should surface:
  * Awareness of initiatives
  * Readiness for change
  * Training needs
  * Concerns and suggestions
- Include one "pulse check" question on overall engagement

Output Format:
- Each question with response type
- Brief explanation of what each question measures
- Suggested follow-up actions based on response patterns

Email Summarization and Action Items

For long email threads, use this pattern:

I'm pasting an email thread below. Please provide:

1. **Thread Summary** (3-5 sentences)
   - What's being discussed?
   - Who are the key participants?
   - What decisions were made?

2. **Action Items**
   For each action item:
   - What needs to be done
   - Who is responsible
   - Deadline (if mentioned)

3. **Open Questions**
   - What remains unresolved?
   - What needs clarification?

4. **My Recommended Response**
   Draft a reply that addresses the open items and moves things forward.

[PASTE EMAIL THREAD HERE]

אני מדביק/ה שרשור אימייל למטה. בבקשה ספק:

1. **סיכום השרשור** (3-5 משפטים)
   - על מה מדברים?
   - מי המשתתפים המרכזיים?
   - אילו החלטות התקבלו?

2. **פריטי פעולה**
   לכל פריט פעולה:
   - מה צריך לעשות
   - מי אחראי
   - דדליין (אם הוזכר)

3. **שאלות פתוחות**
   - מה נשאר לא פתור?
   - מה צריך הבהרה?

4. **התגובה המומלצת שלי**
   טיוט תשובה שמתייחסת לפריטים הפתוחים ומקדמת את הדברים.

[הדבק שרשור אימייל כאן]