Index First | Artizan Software

See the Difference

Same codebase question. One dumps everything. One lets AI navigate.

Without Index

You: "Here's my entire codebase. How do I add user authentication?" [15,000 lines of code pasted]

AI wades through irrelevant utility functions, configuration files, and unrelated features. The answer references the wrong files or misses key integration points buried in the middle.

Too much context. Important details get lost.

With Index

You: "Here's my project structure. Which files should I look at to add user authentication?"

- UserController: handles user endpoints
- AuthService: manages tokens and sessions
- DatabaseContext: data access layer
- Middleware/: request pipeline handlers

AI: "For authentication, I'd need to see AuthService, UserController, and your Middleware folder."

You send just those files. AI gives precise, grounded advice.

Focused context. Accurate answers.

Why This Works

AI has a limited attention budget. When you dump everything into a prompt, important information competes with irrelevant noise. Details in the middle get the least attention. The signal drowns in the flood.

An index solves this. Short descriptions let AI identify what matters before seeing the full content. It navigates to relevance instead of searching through everything. Less context, better focus, sharper answers.

What Makes a Good Index

Brief descriptions: One sentence explaining what each item contains or does
Clear names: "AuthService: manages login tokens" not just "auth.js"
Logical grouping: Related items together so AI sees the structure
Complete coverage: Everything AI might need to choose from

Index Formats That Work

Project outline: List each module with a one-line purpose
Architecture doc: Components with descriptions, inputs, and outputs
Table of contents: Document sections with brief summaries
API manifest: Endpoints with what each one handles

The Technique

Don't dump everything into context. Send an index first. Ask AI what it needs. Then provide only that. Two focused exchanges beat one overloaded prompt.

When to Use This

• Working with large codebases or document collections
• Asking questions that only need part of your material
• Getting vague or inaccurate answers from big context dumps
• Building systems that retrieve information for AI
• Any time AI seems to miss important details