Welcome to the AI Prompt Masterclass!

"Even the finest tool is nothing without a wise helmsman at the wheel."

Who Should Take This Course

  • Individuals with basic experience using AI chat tools.
  • Interested to craft precise, effective, and sophisticated prompts for professional or creative tasks.
  • Suitable if you want to move beyond simple questions.
  • Keen in advanced, systematic applications covered in later modules.

How It Works

  • Read the instructional text in each module.
  • Use the text boxes provided to draft prompts based on your learned concepts.
  • Click the 'Copy' icons to copy specific sections, or click 'Copy All' to copy the full prompt.
  • Paste your prompt into an AI tool (e.g., Gemini, ChatGPT, Claude) to test your work.

Tips for Success

  • Save and organise your most effective prompts.
  • Build a personal prompt library and turn your new skills into a powerful, reusable asset.
  • Think of these methods as your foundation. You should customise and build upon these techniques to develop your style as you explore further.

Module 1: Core Concepts & The C-R-A-F-T+T Framework

What is an AI Language Model?

Think of an AI like ChatGPT, Gemini or Claude as an incredibly knowledgeable and fast assistant. Technically, it's a Large Language Model (LLM).

  • It's a Prediction Machine: An LLM doesn't "think" or "understand" like a human. It's a sophisticated pattern-matcher. When you give it a prompt, it calculates the most probable sequence of words to form a response. For example, if you type "The cat sat on the...", the AI predicts that "mat" is statistically more likely to follow than "rocket ship", based on the vast amount of text it has processed.
  • The Prompt is Everything: Because the AI operates on probability, the quality of your instruction is paramount. A vague prompt gives the AI too many possibilities, leading to a generic guess. A clear, detailed prompt narrows the possibilities, guiding the AI to a specific and useful response. This is the difference between "garbage in, garbage out" and "brilliance in, brilliance out".
  • That's why vague prompts like "make it better" are the equivalent of telling a junior designer "make it intuitive" and then walking away. You'll end up with something generic, safe, and soulless, not because the AI "missed the brief" or lacked intelligence, but because no brief was provided. The system defaults to its most average response without specific goals, a defined voice, and detailed context. You may never realise your output is "average" because you haven't seen a better one. The model's performance depends entirely on the quality of your prompt- the blueprint you give it. Although future AI might "think independently," for now, every detail in your prompt can compound and significantly influence the output positively or negatively.

Understanding AI's Thought Process and Enhancement

AI operates differently from humans; it specialises in recognising patterns and predicting words based on the context provided. The effectiveness of AI responses is heavily influenced by the quality of your prompts, with approximately 80% of strong responses stemming from well-crafted questions.

If your initial prompt falls short, don't hesitate to tweak it. Think of prompt writing as a back-and-forth conversation that can gradually improve. Remember that practice will sharpen your communication skills with AI, so continue to experiment for better outcomes.

The C-R-A-F-T+T Framework

A powerful prompt is a comprehensive brief. Use this framework to ensure you provide all the necessary information. Let's look at some examples:

  • Context: The background information.

    Weak: "Write about our new product."

    Strong: "Our company, ADVANCRT, is launching a new sustainable packaging solution called 'EcoPack' next month. It's made from 100% recycled materials and is fully compostable. Our primary competitor has just released a similar, but more expensive, product."

  • Role: Who should the AI be?

    Weak: "Sound professional."

    Strong: "Act as a seasoned B2B technology journalist writing an objective, critical overview of a new piece of software."

  • Audience: Who is the final output for?

    Weak: "Write for our customers."

    Strong: "The target audience is marketing managers in mid-sized consumer packaged goods companies. They are familiar with basic digital marketing concepts but are likely new to AI applications in logistics."

  • Format: How should the output look?

    Weak: "Give me some ideas."

    Strong: "Provide the output as a Markdown table with three columns: 'Idea', 'Target Audience', and 'Potential Headline'."

  • Task: The specific action verb.

    Weak: "Do something with this text."

    Strong: "Summarise the key findings of the attached market research report into a 300-word executive summary and then extract the top three most surprising statistics as bullet points."

  • Tone: The emotional quality or style.

    Weak: "Make it engaging."

    Strong: "Adopt an informative yet optimistic tone, avoiding corporate jargon and clichés. The style should be authoritative but approachable."

Ethical Considerations in Prompting

As we incorporate AI into our practices, it's essential to align with ethical standards and maintain a human-centric approach. Prompts play a crucial role in upholding these principles. When creating prompts and evaluating AI outputs, consider the following:

  • Bias Mitigation: AI models trained on biased data can perpetuate stereotypes. Therefore, ask for diverse representation and avoid stereotypes. For instance, you might say: "Generate three social media post ideas for International Women's Day that are inclusive and empower individuals across various fields."
  • Truthfulness: AI can sometimes 'hallucinate' or produce inaccurate information. To mitigate this, request source citations and verify facts across different models and with human review. For example: 'Draft an article on X, noting data points that require fact-checking."
  • Originality: Since AI learns from existing data, it’s important to review and rephrase outputs to add a personal touch and ensure uniqueness.
  • Transparency: Decide when to disclose AI involvement to your audience to maintain trust.

Ultimately, prompts are our first line of defense and key to guiding AI responsibly, ensuring that content aligns with ethical and human-centered values. Thoughtful, proactive prompting supports our commitment to ethical AI use.

Key Takeaways from Module 1

  • Effective prompts are more than just questions; they are instructions built from core components like Role, Task, Context, and Format.
  • Prompting is an iterative conversation; practice and refinement are key to getting good results.
  • Ethical considerations like bias, truthfulness, and originality must be part of the prompting process.
  • "Garbage in, garbage out" and "Brilliance in, brilliance out"

Activity 1: Your First Structured Prompt

Worked Example: Let's Build a Prompt Together

Before you try it yourself, let's apply the concepts from this module. Our goal is to ask the AI to write a professional email.

1. Define the Role: 'Act as a senior project manager.'
2. Set the Task: 'Write an email to a client.'
3. Provide Context: 'The client's project is delayed by one week due to a technical issue. We need to inform them professionally and maintain their confidence.'
4. Specify the Format: 'The email should be concise, empathetic, and clear.'

Resulting Prompt:
Act as a senior project manager. Write an email to a client informing them that their project is delayed by one week due to an unexpected technical issue that is now resolved. The email should be concise, empathetic, and clear, and it should reassure them that we are prioritising their project.

Activity: Now, You Build One!

Think of a common, repetitive task you do at work. Using the C-R-A-F-T+T framework, draft a detailed prompt for this task in the template below.

From Art to Science: Bridging Modules 1 and 2

In Module 1, we learned the 'Art of Conversation', using core elements like Role, Task, and Context. Now, we explore the 'Science' behind it. Think of it this way:

  • The Role you assign is an application of the Persona component.
  • The Task you define is a Precise Task Instruction.
  • The background you provide is the Context.

This module breaks down these ideas into a more structured, scientific framework that experts use to get reliable results every time.

Module 2: Prompt Engineering vs. Context Engineering

Key Prompt Components: What the Experts Use

While prompts can be simple, professional-grade prompts are built from key ingredients. Here’s what they are and how they're used:

1. The Persona

What it is: Telling the AI who to be.

Real-World Application: Duolingo's AI likely uses a 'Friendly, encouraging language tutor' persona in its prompts to generate supportive feedback for learners.

2. The Specific Task

What it is: Telling the AI exactly what to do.

Real-World Application: A support chatbot's prompt might include the task: 'Analyse the user's query and classify it into one of five categories: Billing, Technical, Sales...'.

A Quick Note: While the Task (what to do) and the Persona (who to be) are presented as separate components, they work together. The Persona influences the style and tone in which the Task is executed. They are two separate but complementary instructions for the AI.

3. Context & Constraints

What it is: Giving the AI the background information and rules it must follow.

Real-World Application: A legal AI assistant would be given context (case law documents) and constraints ('Do not provide legal advice, only summarise the findings') in its prompt.

Key System Techniques

Beyond the basic components, experts use specific techniques to manage information flow and structure complex requests.

Retrieval-Augmented Generation (RAG)

What it is: This is a powerful first step into true context engineering. It involves grounding the model's response in a trusted, external knowledge base. The system first retrieves relevant documents (like your company's internal wiki or product specs) and then adds that information to the prompt so the AI can generate an answer based on those facts.

Real-World Application: A customer service bot uses RAG to retrieve the latest return policy from the company's knowledge base before answering a customer's question about a refund.

Prompt Chaining

What it is: Prompt chaining is the technique of using the output from one prompt as an input for the next. This creates a workflow where the AI builds on its own work, breaking a complex task into manageable steps.

Real-World Application:

  1. Prompt 1: 'Brainstorm 10 names for a new coffee shop.'
  2. Prompt 2: 'Take the 3rd name from the list above, "The Daily Grind," and write a short slogan for it.'

Key Takeaways from Module 2

  • Professional prompts are built from core components like Persona, Task, and Context.
  • Advanced systems use techniques like RAG to ground AI in facts and Prompt Chaining to handle complex workflows.
  • The "art" of a good conversation is powered by the "science" of a well-structured prompt.

Activity 2: Mapping Your Context Sources

Worked Example: Mapping Context for a Marketing Email

Let's say our task is to write a marketing email. Here’s how we would map the context sources before writing the prompt.

System Instructions: 'Always use the ADVANCRT brand voice: professional, innovative, and customer-centric. Never mention competitors by name.'
Conversation History: 'N/A for this first email.'
Retrieved Documents (RAG): 'The official product one-pager for "EcoPack" (product-specs.pdf).'
Tool Definitions: 'N/A'
Desired Output Structure: 'A standard email format: Subject Line, Salutation, Body, Closing.'

Activity: Map Your Own Context!

Look at the task you chose in Activity 1. Now, think like a context engineer. What information, beyond the prompt itself, would the AI need to perform this task perfectly every time? Create a "Context Map" by listing the answers to these questions:

What are the absolute, high-level rules?

Is there relevant info from past interactions?

What external documents would be useful?

Does the AI need tools?

Do you need a specific format like JSON for another system to read?

Module 3: Techniques for Superior Reasoning

Now that we have the architect's mindset, let's stock our toolkit with advanced techniques. These are instructions that help the AI 'think' more effectively, moving from simple answers to reasoned conclusions.

  • Chain-of-Thought (CoT): The simplest yet most powerful technique. You explicitly ask the AI to "think step by step" or "show its reasoning." This forces a more logical, less rushed process, dramatically reducing errors in complex tasks.

    Marketing Example: "We ran a social media campaign with a budget of £5,000. It generated 200,000 impressions, 5,000 clicks, and 100 conversions. Let's think step by step to calculate the Cost Per Click (CPC) and Cost Per Acquisition (CPA)."

    This guides the AI to first calculate CPC (£5000 / 5000 clicks = £1), then CPA (£5000 / 100 conversions = £50), ensuring accuracy.

  • Decomposition: You instruct the AI to break a complex task into a sequence of simpler sub-tasks. This is ideal for project planning or multi-stage problems.

    Marketing Example: "I need to create a content strategy for our new 'EcoPack' product launch. Decompose this into the following steps: 1. Brainstorm three blog post titles targeting eco-conscious consumers. 2. For each title, write a brief two-sentence summary. 3. Draft a sample LinkedIn post announcing the first blog article."

  • Self-Critique: You can instruct the model to review and improve its own answer. For example: "After generating the answer, review it for accuracy and clarity. If there are any issues, refine it before presenting the final version."

Key Takeaways from Module 3

  • You can guide the AI's reasoning process, not just its final output.
  • Chain-of-Thought ("think step by step") improves accuracy on logical and mathematical tasks.
  • Decomposition helps the AI tackle large, complex projects by breaking them into smaller parts.

Module 4: Techniques for Clarity & Control

Next, let's look at techniques to make your instructions crystal clear and control the output format, ensuring you get exactly what you need.

  • Few-Shot Prompting: Don't just tell the AI what you want; show it. Provide 2-3 high-quality examples of the desired input and output. This is one of the most effective ways to guide the model on format, style, and nuance.

    Marketing Example: "Rewrite the following technical feature into a customer-centric benefit statement. Follow the examples provided.

    ---
    **Example 1:**
    Feature: 'Our software uses a quad-core processing algorithm.'
    Benefit: 'Experience lightning-fast performance that lets you get your work done in half the time.'
    **Example 2:**
    Feature: 'The product is made from aerospace-grade aluminium.'
    Benefit: 'Enjoy a durable, lightweight design that's built to last, wherever you go.'
    ---

    Now, rewrite this feature: 'Our new CRM integrates with over 50 third-party applications.'"

  • System 2 Attention (S2A): An instruction that tells the model to first carefully analyse the provided context and extract the most relevant facts before generating a response. This is named after the concept of 'System 2' thinking in psychology—slow, deliberate, and analytical.

    Marketing Example: "I've pasted 20 customer reviews below. Before you write a summary, first extract the top 5 most frequently mentioned positive themes and the top 3 most frequently mentioned negative themes. Then, use those themes to write a balanced summary of customer sentiment."

  • Using Delimiters: Use clear separators like XML tags (<document>, </document>) or Markdown headers (###Instruction###) to logically separate different parts of your prompt. This helps the AI understand which part is the context, which is the instruction, and which are the examples.

    Example Structure:
    ###CONTEXT###
    [Paste your background information here]
    ###EXAMPLES###
    [Paste your few-shot examples here]
    ###TASK###
    [State your specific task here]

Key Takeaways from Module 4

  • Showing the AI examples (Few-Shot Prompting) is often more effective than telling it what to do.
  • You can instruct the AI on how to process information (S2A) before it answers.
  • Using clear separators (Delimiters) makes your prompts easier for the AI to understand.

Activity 3: Upgrading Your Prompt with Advanced Techniques

Worked Example: Upgrading a Prompt

Let's take our simple email prompt from Activity 1 and upgrade it with a Few-Shot Example and Delimiters.

###CONTEXT###
The client's project is delayed by one week due to an unexpected technical issue that is now resolved. We need to inform them professionally and maintain their confidence.

###EXAMPLE###
Subject: Quick Update on Project Phoenix
Body: Hi [Client Name], Just a quick note to let you know we've had to adjust our timeline slightly. We're now looking at delivery on [New Date]. We hit a small technical snag, but the team has already resolved it. We're fully focused on your project and will keep you updated.

###TASK###
Act as a senior project manager. Write an email to the client based on the context and matching the style of the example.

Activity: Upgrade Your Own Prompt!

Take your prompt from Activity 1. Now, upgrade it. Incorporate at least two of the advanced techniques we just discussed.

Module 5: Context Engineering: Building For Scale

You've just learned powerful techniques to craft expert prompts. So, why dedicate entire modules to organising them? Because for any professional, consistency and efficiency are key.

  • For Marketers: Ensure every ad copy request maintains the exact brand voice.
  • For Recruiters: Generate tailored outreach messages that follow company policy and tone.
  • For Developers: Create a library of prompts for generating boilerplate code, writing documentation, or explaining complex snippets in a consistent style.

The next modules shift from crafting single prompts to building your personal system. This is how you scale your new skills.

Key Course Concepts

  1. Custom Chatbots: Your Specialist AI
    Analogy: Think of a custom chatbot as a highly trained specialist doctor. While a general AI is like a general practitioner, you create a specialist AI for one expert purpose.
    What It Is: A personalised version of an AI that you tailor with specific instructions, extra knowledge (like uploaded files), or a completely redesigned user experience.
    Why It Matters: Custom chatbots are excellent for scaling a specific process or creating an AI expert that can handle a single task with high accuracy.
  2. Context Engineering: Setting the Stage for the AI
    Analogy: Imagine you're directing a play. Context is the entire stage set, the lighting, and the actor's backstory. The prompt is just the actor's single line. A good stage (context) makes the line powerful and meaningful.
    What It Is: The process of designing and managing the background information an AI uses. It's about giving the model a "memory" and the right environment for your interaction.
    Why It Matters: Good context ensures the AI understands the world it's operating in, leading to more accurate and relevant responses.
  3. The Relationship: Context is the Foundation
    Analogy: Think of building a house. Context is the strong, wide foundation. The prompt is the frame of the house built on top. You cannot build a stable house on a weak foundation.
    What It Is: The relationship is layered. Context is the broad foundation of knowledge, and the prompt is the specific instruction that works within it. Good context filters out irrelevant information, allowing prompts to be short and powerful.
    Why It Matters: When the context is strong (for example, by providing a specific report), a simple prompt like "Summarise the key findings" works far better than a long, detailed one.
  4. Augmenting Your Workflow: Your AI Partner
    Analogy: It’s the difference between using a basic calculator and creating custom spreadsheet functions. A calculator helps, but custom functions solve your unique problems instantly and give you a professional edge.
    What It Is: The practice of moving beyond just chatting with an AI and building task-specific assets (like mini-assistants or custom bots) for your daily work.
    Why It Matters: This approach turns you from a passive user into an active creator, building intelligent tools that provide a significant advantage in your tasks.

We've come a long way. Now, let's put it all together into a production-ready framework. The definitive template organises all possible components into four logical phases, creating a robust structure for any complex task. Below, we explain the purpose of each component and its position in the flow.

Phase 1: The Constitution & Core Identity

(Goal: Establish the absolute rules, the actor's identity, and the overall objective. This is the non-negotiable foundation.)

  • 1. AI Safety Protocol: Why first? This is the supreme law. It frames the entire operation and prevents override.
  • 2. Persona: Why second? After the rules, you define the actor. The Persona must adhere to the Constitution.
  • 3. Tone & Style: Why third? This refines the Persona, flowing from general identity to specific voice.
  • 4. Mission / Objective: Why fourth? With identity set, you now give the AI its ultimate purpose.

Phase 2: The Dossier: Information & Examples

(Goal: Provide the AI with all the necessary data, variables, and examples required to execute the mission.)

  • 5. Context: Why here? The AI has its briefing; now it needs the "intelligence dossier" of background info.
  • 6. Task Inputs: Why after Context? This narrows from general background to the specific variables for this run.
  • 7. Few-Shot Examples: Why last in this phase? After all raw info, you show a perfect example of success to solidify understanding.

Phase 3: The Blueprint: Execution & Reasoning

(Goal: Instruct the AI on how to think and act. This is the engine room of the prompt.)

  • 8. Logical Frameworks (CoT, etc.): Why first in this phase? This is the "Standard Operating Procedure" for reasoning.
  • 9. Program-Aided Language (PALs): A specialised tool. This tells the AI to switch to code when it needs perfect logic or maths.
  • 10. Reusable Modules: Another specialised tool. If the AI sees a trigger (e.g., "run an audit"), it activates a detailed, pre-written procedure.
  • 11. Interactive Execution Protocol: A crucial "Go/No-Go" gate. It forces the AI to pause and ask for clarification if a request is ambiguous, preventing wasted effort.

Phase 4: The Inspection: Output & Verification

(Goal: Define the final product and enforce a strict quality control check before delivery.)

  • 12. Output Specification: Why first in this phase? Before verifying its work, the AI needs a clear blueprint of the final product.
  • 13. Constraints & Guardrails: These are the rules used to audit the formatted output (e.g., word count).
  • 14. Verification Process: Why last? This is the final action: the AI performs a quality check against all preceding rules before presenting the answer.

Example Template Applications

Example 1: The Marketing Campaign Template

This template helps you quickly generate creative ideas for a campaign.

Persona: Act as a senior marketing strategist with expertise in [Target Demographic].
Task: Generate 5 distinct marketing campaign ideas for our new product, [Product Name], which is a [Product Description].
Context: The campaign's primary goal is [Campaign Goal, e.g., brand awareness, lead generation]. The key channels are [Marketing Channels]. The tone should be [Tone, e.g., witty, professional, inspiring].
Format: Present the ideas in a bulleted list. For each idea, provide a catchy slogan.
Example 2: The Bug Report Summary Template

This template helps developers get clear, summarised bug reports.

Persona: Act as a Senior QA Engineer.
Task: Summarise the following user-submitted bug report into a structured format. Identify the likely root cause and suggest a next step for the development team.
Context: User Report: "[Paste User Bug Report Here]". The application is [Application Name], version [Version Number].
Format: 
- **Summary:** [1-sentence summary of the issue]
- **Replication Steps:** [Bulleted list of steps]
- **Expected vs. Actual Behaviour:** [Clear comparison]
- **Potential Root Cause:** [Your analysis]
- **Suggested Next Step:** [e.g., 'Investigate API endpoint X', 'Review component Y']

Key Takeaways from Module 5

  • A structured, multi-phase template turns a simple prompt into a robust, professional instruction set.
  • Organising your prompt into phases (Constitution, Dossier, Blueprint, Inspection) improves clarity and reliability.
  • Well-designed templates are reusable assets that scale your expertise across a team.

Activity 4: Building a Production-Ready Prompt

Worked Example: A Production-Ready Prompt

Here is a prompt for a marketing task, built using the four-phase structure. Notice how it includes a safety protocol, a clear mission, context, a logical framework, and a verification process.

### AI TRUTHFULNESS & SAFETY PROTOCOL
You must adhere to the principle of verifiable truth. Do not invent statistics or user testimonials.

### PERSONA
You are an expert social media marketing manager for a B2C brand in the sustainable products space.

### MISSION
Your primary objective is to draft three engaging and distinct Instagram post captions for our new 'EcoPack' product.

### CONTEXT
The product is a set of reusable, compostable food storage bags. The target audience is environmentally conscious millennials aged 25-40.

### LOGICAL FRAMEWORK
You MUST follow this thinking process:
1.  Analyse the context to identify the key benefits: sustainability and convenience.
2.  For each of the three captions, focus on a slightly different angle (e.g., one on waste reduction, one on style, one on practicality).
3.  Write the captions.

### OUTPUT FORMAT
Your final response must be a list of three captions. Each caption must be under 150 characters and include at least two relevant hashtags.

### VERIFICATION PROCESS
After generating the drafts, review them to ensure they align with the persona and mission.

Activity: Build Your Own Production Prompt

We're going to do one final revision of your prompt. This time, use the full 14-point template structure. You don't need to use all 14 points, but try to incorporate the key phases. This is your master template.

Module 6: Meta-Prompting: Creating a Prompt Refinement Assistant

Now that you've mastered the art of crafting prompts, let's explore an advanced, "meta" concept: creating a specialised AI assistant whose sole purpose is to help you build and refine other prompts. Think of this as your personal "Prompt Manager" or "Gem Manager".

The goal is to create a set of instructions so powerful that you can give this AI a rough idea or a weak prompt, and it will guide you through a structured process to transform it into a high-performance, reusable prompt template.

Activity 5: Build Your Prompt Manager

Worked Example: The Goal

The goal is to build a prompt that turns your AI into a prompt-building expert. When you give it a simple prompt like "write a blog post about AI," it should not write the post. Instead, it should start asking you clarifying questions to help you build a better prompt, like this:

YOU: "Help me write a prompt to create a blog post about AI."

YOUR NEW ASSISTANT: "Of course. To create the best prompt for you, I need a few more details:
1. **Audience and Tone:** Who is this blog post for? What style should it have (e.g., technical, for beginners, witty)?
2. **Scope and Constraints:** Are there any specific topics within AI to focus on or avoid?
3. **Format and Structure:** Should it have a specific word count or sections?"

Activity: Build Your Manager

Instead of just giving you a template, let's build your Prompt Manager step-by-step. This will help you understand the purpose of each component. Complete each text area below to construct the full set of instructions for your new assistant.

Step 1: The Constitution

This is the non-negotiable foundation. It ensures every prompt your assistant creates is safe and reliable. This part is critical, so we've provided it for you. Your task is simply to ensure it's the very first part of your prompt.

Step 2: Core Identity & Mission

Now, define who your assistant is and what its primary goal is. This gives it a clear purpose.

Step 3: The Process

This is the most important part. Define the step-by-step process your assistant MUST follow. This includes analysing the user's prompt, asking clarifying questions, rebuilding it, presenting it, and justifying the changes.

Conclusion: Wrap-Up & Your Path Forward

Incredible work! You've gone from the basics of AI to designing complex, production-level prompts. Let's recap the journey.

Key Takeaways

  • 1From Writer to Architect: We moved from simply writing prompts (Prompt Engineering) to designing the entire information ecosystem (Context Engineering).
  • 2A Toolkit of Techniques: You now have a powerful set of advanced techniques like Chain-of-Thought and Few-Shot examples to control the AI's reasoning and output.
  • 3Structure is Everything: A well-structured prompt, like the definitive 14-point template, is the key to creating reliable, scalable, and reusable AI solutions.

Actionable Strategies for Improvement

Your learning doesn't stop here. The world of AI is moving incredibly fast. Here is a practical roadmap for your next steps:

  1. Master the Foundations: Advanced Prompt Structuring. Perfect the core instruction. Consistently use clear delimiters like <document> tags, assign a clear persona, provide high-quality few-shot examples, and use Chain-of-Thought for reasoning.
  2. Adopt an Architectural Mindset: Map Your Context Sources. Shift from thinking about a single prompt to designing an information flow. For any task, plan all potential sources of context: static rules, user queries, chat history, external documents (RAG), and required tools.
  3. Implement a Foundational RAG Pipeline. Retrieval-Augmented Generation is a powerful first step into true context engineering. This involves grounding the model's response in a trusted, external knowledge base, ensuring it uses factual, up-to-date information.
  4. Actively Manage the Context Window. Do not treat the context window as a passive dumping ground. Prioritise information, placing the most critical instructions at the very beginning and end to avoid the "lost in the middle" problem. Use summarisation techniques for long documents to preserve detail while reducing token count.
  5. Build Agentic Workflows with Isolated Context. For complex, multi-step tasks, break them down. Use prompt chaining (where the output of one prompt becomes the input for the next) or design multi-agent systems with specialised agents (e.g., a "planning agent," a "research agent"), each with its own optimised context.

Thank you for your incredible focus and making it up till here. You are now a prompt architect, ready to build the future of how you work with AI!