The Hidden Cost of Bad Context: How to Stop Burning Tokens

The "More is Better" Fallacy

When developers first get Cursor, they discover the @Codebase tag. They love it. They use it for everything.

• "How do I center a div?" -> @Codebase
• "Fix this typo." -> @Codebase

This is bad for two reasons:

1. It confuses the AI: When you include 500 files for a CSS question, the AI has to sift through your backend logic, database schema, and tests just to find the CSS file. It increases the chance of hallucination.
2. It's slow: Processing 100k tokens takes time. You are waiting 10 seconds for an answer that should take 1 second.

The "Flashlight" Method

Treat your context like a flashlight in a dark room. You don't want to turn on the stadium lights; you want to shine a beam on the workbench.

The Rule: Only tag files that are *directly* involved in the change, plus one layer of "neighbors."

Example: adding a new column to a database table.

• Tag: schema.prisma (The DB definition)
• Tag: /app/api/route.ts (The API that uses it)
• Tag: /components/Table.tsx (The UI that displays it)
• Do NOT Tag: The auth system, the footer component, the util library.

The Power of .cursorignore

Just like .gitignore, you can create a .cursorignore file. Use this to ban files that waste tokens but provide no value.

# .cursorignore
package-lock.json
yarn.lock
/node_modules
/dist
.next
*.svg
**/*.test.tsx (Unless I am specifically writing tests)

This ensures that even if you accidentally use @Codebase, the AI won't waste brainpower reading your 50,000 line lockfile.

Conclusion

Context curation is a skill. The senior engineer of the AI age isn't just a code writer; they are a Context Curator. Master this, and your AI tools will feel twice as fast and twice as smart.