A humorous yet practical guide to AI-assisted development. DON'T PANIC.
View the Project on GitHub HermeticOrmus/hitchhikers-guide-to-vibe-engineering
Risk Level: 🟡 Caution Advised
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.
AI models have limited context windows:
Sounds like a lot until:
Then: context overflow.
Message 1: “We’re using TypeScript strict mode” Message 50: AI generates untyped JavaScript
Early: Clean, consistent code matching your patterns Later: Generic code that doesn’t fit your codebase
AI: “As I mentioned earlier…” You: You mentioned the opposite earlier.
AI: “Could you remind me what framework you’re using?” You: I told you three times.
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
Periodically remind the AI of critical requirements:
"Quick reminder before we continue:
- TypeScript strict mode
- No new dependencies
- Follow existing patterns"
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
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..."
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 AI’s memory is shorter than you think. Repeat important things.”
Check your last long conversation: