How do you share just the relevant part of a large file with an AI to save tokens?

Question

Accepted Answer

Pasting an entire 2,000-line file wastes tokens and *hurts* answer quality. The goal is to give the AI **exactly the context it needs to reason about your problem** — and nothing more.

## What to send

- **The specific function** you're asking about, plus the **types and signatures** it depends on.
- **A one-line summary** of the surrounding context instead of the surrounding code: "This runs inside an Express request handler; `db` is a Postgres pool."
- **File paths** so the AI can refer to structure without you pasting it.
- **Strip noise** — dead code, commented-out blocks, unrelated functions, verbose logging.

## Before / after

```text
# Bad: paste the whole 800-line user-service.ts
#   → AI wades through unrelated code, may anchor on the wrong function

# Good: send only what matters
//  file: src/services/user-service.ts
//  context: called from POST /users; `db` is a Postgres pool (pg)
type User = { id: number; email: string };

async function createUser(email: string): Promise<User> {
  const { rows } = await db.query(           // <-- the function I need help with
    "INSERT INTO users (email) VALUES ($1) RETURNING id, email",
    [email],
  );
  return rows[0];
}
//  Question: how do I handle a duplicate-email conflict here?
```

The second version gives the AI the function, its return type (`User`), and a one-line note about `db` — enough to answer precisely, with no irrelevant code to distract it.

## Why less context is better

It's not only about cost. Extra code is **distraction**: the model may latch onto an unrelated function, mirror a bug elsewhere in the file, or dilute its attention across thousands of irrelevant tokens. A focused prompt produces a focused, accurate answer.

## Why it matters

Token budgets are finite and large dumps degrade both speed and quality. Sending the minimal relevant slice — the target function, its types, a context summary, and file paths — saves tokens *and* sharpens the answer by removing distractions. Curating context is a core skill of working effectively with AI: you're not hiding information, you're directing attention to what matters.