Why Your LLM Is Trash (And How To Fix It With Context Engineering)
ToxSec | A practical guide to getting better, more accurate results from any AI with context engineering
TL;DR: Your Quick Guide to Better AI Results
Be Specific: Vague questions get vague answers. Tell the AI exactly what you want, in detail.
Provide the Data: Don’t assume the AI knows what you’re talking about. Paste in the text, data, or examples it needs to do the job.
Give it a Role: Tell the AI to act as an expert, a copywriter, or any persona. This dramatically improves the tone and quality of the response.
Use Advanced Tools for Big Jobs: For complex tasks, use techniques like RAG (connecting to live data) or Chaining (making multiple AIs work together in a workflow).
0x00: What is Context Engineering (And Why Does It Matter)?
I remember the first time I asked an AI to do something I thought was complex. I wanted it to analyze a customer complaint and suggest a solution based on our company’s policies. The result was… disappointing. It was generic, missed the point, and felt completely useless.
The truth is, the problem wasn’t the AI. It was me.
Welcome to the world of Context Engineering, the single most important skill for getting high-quality results from any AI. It’s how you turn a generic chatbot into a specialized assistant that actually understands you.
The Problem: Why Does My AI Sound So Generic?
The core of the issue is that AI models don’t think; they predict the next most likely word based on the data you give them. If you give them generic data, you get generic results.
There’s a classic saying in software: “Garbage In, Garbage Out.” This is the golden rule of working with AI.
What is the “Context Window”?
First, you need to understand the context window. Think of it as the AI’s short-term memory. It includes everything from your current conversation—your questions, its previous answers, and any data you’ve given it. But this memory is finite and measured in tokens (pieces of words). Once the conversation exceeds the limit, the oldest information gets pushed out.
0x01: How Can I Instantly Improve My Prompts?
Getting better answers starts with a few core principles.
Rule 1: Are Your Questions Specific Enough?
A detailed and unambiguous prompt will always outperform a vague one.
Vague Prompt: “Tell me about electric cars.”
Specific Prompt: “Compare the 2025 Tesla Model 3 Long Range with the 2025 Ford Mustang Mach-E Premium. I want to know the battery range in miles, the 0-60 mph time, and the starting MSRP in the US.”
Rule 2: Are You Providing the Right Data?
Never assume an LLM knows about a specific article or email. The best way to ground an AI in facts is to provide the information directly.
Instead of: “Help me answer this customer email.”
Try: “Here is our return policy and the customer’s email below. Draft a friendly response that explains why they can’t return a final sale item: [paste policy and email here]”
Rule 3: Have You Given the AI a Role to Play?
Assigning a persona tells the model how to frame its response.
Act as a friendly and encouraging personal trainer. A client is struggling with motivation. Write a short, encouraging text message I can send them.
Rule 4: Is Your Prompt Easy for an AI to Read?
Use simple formatting like headers or delimiters (###) to create clear, separate sections.
###ROLE###
You are a financial analyst.
###ARTICLE TEXT###
[Paste the full text of a financial news article here...]
###YOUR TASK###
Based *only* on the article I provided, what are the top three risks mentioned?
0x02: How Do I Tackle Complex Problems with AI?
Once you’re comfortable with the basics, you can use advanced strategies.
Technique 1: How Can I Show the AI What I Want? (Few-Shot Learning)
With few-shot learning, you provide the model with a few examples of the desired input/output format. This shows the AI the pattern you want it to follow.
I need you to classify these support tickets as “Urgent,” “High,” or “Medium.”
Ticket: “My entire website is down and I can’t access anything!”
Priority: Urgent
Ticket: “I can’t figure out how to export my monthly report.”
Priority: High
---
Now, classify this one:
Ticket: “The login page is giving me a 404 error and none of my users can get in.”
Priority:
Technique 2: How Do I Connect My AI to External Data? (RAG)
Retrieval-Augmented Generation (RAG) connects an LLM to an external knowledge base, like your company’s documents or the live internet. It retrieves relevant info and adds it to your prompt, overcoming outdated knowledge and allowing the AI to use private data securely.
Technique 3: Can I Make Multiple AIs Work Together? (Chaining)
Some problems are too big for one prompt. The solution is to break them down into a workflow, where the output from one specialized AI prompt (or “agent”) becomes the input for the next.
0x03: What Does the Future of AI Interaction Look Like?
Context engineering is evolving quickly. Tools will begin to automate prompting, context windows will become massive, and we’ll start using multimodal context (images, audio, video). But even then, the core skill will remain the same: prioritizing the right information to get the right answer.
Your Turn
Mastering context is a journey, but it starts with these simple steps. It’s the difference between fighting with your tools and building a true partnership with them.
What’s the most frustrating AI answer you’ve ever received? Share it in the comments below!
Frequently Asked Questions (FAQ)
Q1: What’s the difference between prompt engineering and context engineering? A: Prompt engineering is about crafting the perfect single question. Context engineering is broader; it’s about managing the entire conversation and data flow to the AI, including using tools like RAG and chaining for complex tasks.
Q2: How big is a typical context window? A: It varies greatly by model, from a few thousand tokens (a few pages of text) to over a million tokens (several books). This technology is improving rapidly.
Q3: Can I use these techniques with any AI model? A: Yes! The fundamental principles—specificity, providing data, assigning a role, and clear formatting—will improve your results with virtually any Large Language Model.
Q4: What’s the single most important thing I can do to get better AI answers? A: Be more specific and provide the necessary information directly in the prompt. Don’t make the AI guess. Give it all the facts it needs to succeed.
Want to keep reading and better understand GenAI? Read ToxSec’s GenAI Crash Course!