Risk Level: 🟡 Caution Advised
Guide Entry
CONTEXT WINDOW (n.): The amount of text an AI can "remember" at once. Like RAM, but for conversation. When you exceed it, early information falls off the edge into the void, taking your carefully specified requirements with it.
The Problem
AI models have limited context windows:
- GPT-4: ~128k tokens
- Claude: ~200k tokens
- Others: Varies
Sounds like a lot until:
- Your codebase is 50 files
- Your conversation is 100 messages
- Your paste is 1000 lines
Then: context overflow.
Symptoms of Context Problems
The Forgotten Requirement
Message 1: "We're using TypeScript strict mode" Message 50: AI generates untyped JavaScript
The Style Drift
Early: Clean, consistent code matching your patterns Later: Generic code that doesn't fit your codebase
The Contradiction
AI: "As I mentioned earlier..." You: You mentioned the opposite earlier.
The Repetition Request
AI: "Could you remind me what framework you're using?" You: I told you three times.
Survival Strategies
Start Fresh Often
Long conversations accumulate errors. New chat = clean slate.
Rule of thumb:
- Simple task: 1 conversation
- Medium task: 2-3 conversations
- Complex task: Multiple focused conversations
Restate Key Constraints
Periodically remind the AI of critical requirements:
"Quick reminder before we continue:
- TypeScript strict mode
- No new dependencies
- Follow existing patterns"
Use CLAUDE.md / System Context
Put project constants in a file the AI always sees:
# Project Context
- Language: TypeScript 5.0
- Framework: Next.js 14
- Style: Functional components with hooks
- Testing: Jest + React Testing Library
Chunk Large Requests
Instead of:
"Here's my entire codebase, refactor it"
Do:
"Here's file 1, refactor this"
"Here's file 2, refactor this using the same patterns"
"Here's file 3..."
Summarize Before Continuing
After complex discussions:
"Let's summarize what we've decided:
1. Using approach X
2. With constraint Y
3. Excluding Z
Confirm this is correct before we continue."
The Street Rule
"The AI's memory is shorter than you think. Repeat important things."
Move to Make
Check your last long conversation:
- Find where the AI "forgot" something
- Identify what context would have helped
- Plan where to repeat key constraints next time