A · INSTRUCTIONS
System Prompts
Application-level instructions that guide behavior.
Prompt Design
How task, constraints, examples, and expected response are communicated.
06 · Prompting & System Design
What Should the Model See — and What Should It Return?
This module explains how AI applications shape instructions, assemble context, and define predictable outputs around a model.
Understand the difference between writing a prompt, designing context, setting system-level instructions, and defining output structure.
The Big Picture · context and request stack
SYSTEM PROMPT
USER PROMPT
EXAMPLES
RETRIEVED CONTEXT
LLM
↓
STRUCTURED OUTPUT
↓
APPLICATION
↓
STRUCTURED OUTPUT
↓
APPLICATION
HISTORY
TOOL RESULTS
STATE
OUTPUT SCHEMA
Prompt Design focuses on how instructions are written. Context Engineering is broader: it decides the complete information environment supplied to the model.
Knowledge Structure
B · CONTEXT
Context Engineering
Selecting and organizing the information the model receives.
C · OUTPUT CONTRACT
Structured Outputs
Responses constrained to a predictable schema.
One Example
Invoice review in four layers
USER REQUEST
“Check this invoice and return the result.”
SYSTEM PROMPTYou are an invoice review assistant.
PROMPT DESIGN + CONTEXT
Check required fields; include invoice content, validation rules, policy, and prior tool results.
STRUCTURED OUTPUT{ "status": "...", "issues": [...] }
Key Distinctions
System Prompt≠User PromptApplication instruction vs individual request
Prompt Design≠Context EngineeringWriting the request vs assembling the full environment
Context Engineering≠Context WindowContent selection vs available capacity
Structured Output≠Ordinary proseSchema contract vs free-form response
Recommended Learning Path
This is a learning path, not a request pipeline.
01Prompt Design
Express the task.
02System Prompts
Set persistent instructions.
03Context Engineering
Assemble the environment.
04Structured Outputs
Make responses predictable.
Remember the structure
Prompt design shapes the request, context engineering shapes what the model can see, system prompts set application-level instructions, and structured outputs define what the application receives.