Skip to main content

Lesson 5: Prompt vs. Context Engineering

Topics Covered
  • Prompt Engineering: The craft of wording instructions.
  • Context Engineering: The system-level discipline of assembling data, tools, and memory.
  • Agentic Components: Memory, state management, hybrid RAG, and tools.

Most people know Prompt Engineering: crafting the input text to steer a model. But Context Engineering is the broader discipline: programmatically assembling everything the LLM needs during inference—prompts, retrieved documents, memory, and tools—to deliver accurate responses.

1. Prompt Engineering Techniques​

This is part art, part science. Key techniques include:

  • Role Assignment: Telling the LLM who to be (e.g., "You are a senior Python developer"). The model adopts the vocabulary and concerns of that persona.
  • Few-Shot Examples: "Show, don't just tell." Providing 2-3 input/output pairs helps the model grasp the exact format style (e.g., specific JSON fields).
  • Chain of Thought (CoT): Asking the model to "explain your reasoning" or "think step-by-step." This prevents jumping to conclusions on complex tasks.
  • Constraint Setting: Explicitly defining boundaries (e.g., "Limit response to 100 words" or "Only use provided context") to prevent tangents.

2. Context Engineering Components​

Context engineering builds dynamic, agentic systems that orchestrate the environment.

Memory​

  • Short-term: Summarizing long conversations to stay within the context window.
  • Long-term: Using Vector Databases to retrieve user preferences, past trips, or learned patterns.

State Management​

Tracks where the agent is in a multi-step process (e.g., Flight booked? -> Yes. Hotel booked? -> Pending). This ensures context isn't lost mid-task.

RAG (Retrieval Augmented Generation)​

Connects agents to dynamic knowledge sources.

  • Uses Hybrid Search: Combines semantic (vector) and keyword matching.
  • Contextual Slicing: Instead of dumping an entire "Travel Policy" document into the prompt, it extracts only the relevant sections (e.g., expense limits for Paris) to avoid noise.

Tools​

LLMs can't natively check databases or call APIs. Tools bridge that gap.

  • Function: Queries SQL databases, fetches live prices, or deploys infrastructure.
  • Definition: Context engineering involves defining clear tool descriptions so the LLM knows when and how to use them.

The Sythesis​

  • Prompt Engineering gives you better questions.
  • Context Engineering gives you better systems.

A final production prompt might be 20% static instructions (the question) and 80% dynamic content (populated from memory, state, and RAG).