If you’ve been using AI tools like ChatGPT or Claude, you’ve probably experienced this frustration: every time you start a new conversation, you have to re-explain who you are, what you need, and the context for your work. It’s like introducing yourself to the same colleague every single day.
This is the fundamental limitation of prompt engineering – and it’s why the field is rapidly evolving toward something more powerful: context engineering.
The Problem with Prompt Engineering Alone
Prompt engineering has been the dominant approach to working with AI: craft the perfect instructions, get better results. But here’s the challenge:
Every conversation starts from zero. You must re-specify your context, preferences, constraints, and requirements with every single prompt.
If you’re a teacher in South Africa working with CAPS curriculum, this might look like:
- “You are a South African teacher”
- “Use CAPS curriculum standards”
- “Create materials for Grade 7”
- “Consider multilingual classrooms”
- “Use South African examples and context”
- “Align with DBE guidelines”
Research shows that 80% of prompt engineering effort goes into re-establishing basic context, not describing the actual task. This is inefficient, inconsistent, and doesn’t scale.
What is Context Engineering?
Definition (2025):
“Context engineering is the practice of designing systems that decide what information an AI model sees before it generates a response.”
In simpler terms: instead of telling the AI what to do every time, you build a persistent knowledge system that automatically provides relevant context for every interaction.
The Fundamental Shift
| Prompt Engineering | Context Engineering |
| Manual, repetitive, temporary | Automated, reusable, permanent |
| “Tell AI what to do every time” | “Give AI WHO you are, WHAT your world is” |
| Works for one task | Works across all tasks |
| Doesn’t learn or improve | Gets better with use |
The Five Core Techniques
Context engineering is built on five foundational techniques. Master these, and you can build AI systems that truly understand your needs.
1. System Prompts & Instructions
What it is: Define the AI’s persistent role, objectives, and scope.
Example: Instead of saying “Create a lesson plan” every time, your system knows: “You are a CAPS-aligned educator for South African Foundation Phase. Your purpose is to create standards-based materials that respect multilingual classrooms and South African cultural context.”
Platforms: Claude Projects (custom instructions), ChatGPT Custom Instructions, Gemini GEMs, system messages in n8n
2. Structured Inputs & Outputs
What it is: Specify exact formats with required fields.
Example: Every lesson plan automatically includes: grade_level, CAPS_standard, learning_objectives, assessment_strategy, differentiation_notes, resource_list. No need to ask for these fields each time.
Benefit: Consistency across all outputs, easier to integrate with other tools
3. Dynamic Context Elements
What it is: Automatically inject current information (dates, standards, schedules).
Example: The system knows it’s Term 2, Week 6, and automatically references the appropriate CAPS content for that week. It knows upcoming holidays and adjusts planning accordingly.
Technical note: Often implemented with variables, APIs, or database queries
4. Historical Context Management
What it is: Provide AI access to previous outputs, revisions, and learning patterns.
Example: The AI references lesson plans it created previously, sees which activities worked well (based on your feedback), identifies patterns in student confusion, and improves future outputs based on what succeeded.
This is the “memory” layer: Systems that learn and improve over time
5. RAG & Caching Strategies
What it is: Retrieval-Augmented Generation – pull relevant knowledge from databases, cache frequently-used context.
Example: Instead of pasting the entire CAPS document every time, the system retrieves only the relevant standard when needed. Common context (like your teaching philosophy) is cached for instant access.
This is what makes context engineering scalable: Handle massive knowledge bases efficiently
Practical Implementation: Where to Start
You don’t need to be a developer to implement context engineering. Here’s how to start with tools you already have access to:
Beginner Level: Custom Instructions
Platform: ChatGPT Custom Instructions, Claude.ai (projects)
What you do: Write a comprehensive profile about your role, context, and preferences once. Save it.
Example for educators:
“I am a Foundation Phase teacher in South Africa teaching Grades 1-3. All my work must align with CAPS curriculum standards. I teach in a multilingual classroom (English, isiZulu, Afrikaans). I need materials that are practical for 30+ students with limited resources. Always include differentiation strategies and use South African cultural context.”
Time investment: 10 minutes | Impact: Massive improvement in output relevance
Intermediate Level: Project Knowledge Bases
Platform: Claude Projects, Custom GPTs, Gemini GEMs
What you do: Upload key documents (CAPS curriculum, ATP documents, school policies), write detailed instructions, and save as a reusable project.
What you get: AI with deep knowledge of your specific curriculum, standards, and requirements. No more uploading documents each session.
Time investment: 1-2 hours setup | Impact: Professional-grade, consistent outputs
Advanced Level: Custom Applications
Platform: Custom tools using APIs (Claude API, OpenAI API), LangChain, LlamaIndex
What you do: Build specialized applications with databases, RAG systems, and automated workflows.
Example: An automated lesson planning system that knows the current week of the ATP, retrieves relevant standards, references your previous successful lessons, and generates complete lesson plans in your preferred format.
Time investment: Development project | Impact: Fully automated, scalable systems
Real-World Example: CAPS Lesson Planner
Let me share a concrete example of context engineering in action for South African educators:
The System:
- Technique 1 (System Prompt): “You are a CAPS-aligned lesson planner for South African teachers. All content must meet DBE standards and respect multilingual classrooms.”
- Technique 2 (Structured Output): Every lesson follows the same format: Learning objectives, CAPS alignment, activities (beginning/middle/end), assessment, differentiation, resources.
- Technique 3 (Dynamic Context): System knows current term, week, and automatically selects appropriate ATP topic.
- Technique 4 (Historical Context): References previous lesson plans, teacher feedback, and successful activities.
- Technique 5 (RAG): Retrieves relevant CAPS standards, ATP content, and resource libraries as needed.
The result? Teachers can request “Create a lesson plan” and receive a fully contextualized, standards-aligned, culturally appropriate lesson plan in seconds – without re-explaining their context.
Why Context Engineering Matters Now
We’re moving from a world where AI is a tool you instruct, to a world where AI is a system that understands your world.
Context engineering is the bridge between these two paradigms. It’s how we move from:
- Manual → Automated
- Repetitive → Reusable
- Generic → Personalized
- Individual → Scalable
For educators, this means AI systems that truly understand CAPS curriculum, South African context, multilingual classrooms, and resource constraints – without needing to explain these every single time.
For professionals in any field, it means AI systems that know your industry standards, organizational context, and professional requirements – persistently and reliably.
Getting Started: Your Action Plan
Week 1: Foundation
- Choose one AI tool you use regularly (ChatGPT, Claude, Gemini)
- Write comprehensive custom instructions about your role, context, and needs
- Test with 3-5 typical tasks
- Refine based on results
Month 1: Structured Systems
- Create a project/GEM with your key documents
- Define output structures for your common tasks
- Build a small knowledge base (curriculum docs, templates, examples)
- Share with a colleague and refine
Quarter 1: Advanced Implementation
- Add historical context (save successful outputs, track what works)
- Implement dynamic elements (current dates, schedules, standards)
- Explore RAG tools if comfortable with technology
- Measure impact: time saved, quality improvement, consistency gains
The Bottom Line
Context engineering is not only a technical skill but it’s a fundamental shift in how we work with AI. Instead of treating AI as a tool that needs constant instruction, we build systems that understand our world, our work, and our needs.
The educators, professionals, and organizations that master context engineering will have AI systems that don’t just follow instructions and they anticipate needs, maintain consistency, and improve over time.
Start building your context system today.
The AI tools you use tomorrow will be as smart as the context you build today.
Resources for Further Learning
- Anthropic: Context Engineering for AI Agents (official documentation)
- Prompt Engineering Guide: promptingguide.ai – comprehensive resource
- Claude Projects: Implement context engineering with knowledge bases and custom instructions
- Custom GPTs: Build specialized AI assistants with persistent context
- Gemini GEMs: Create custom AI experts with specialized knowledge
Have questions about implementing context engineering in your work? Reach out – I’d love to hear about your use cases and challenges.