DEV Community

Cover image for Prompting Like a Pro – How to Talk to AI
Manoj Mishra
Manoj Mishra

Posted on

Prompting Like a Pro – How to Talk to AI

#ai #llm #tutorial #productivity

Introduction

The way you phrase your request determines the quality of AI’s output. Most developers treat AI like a search engine—asking vague questions and accepting the first answer. In this series post, we’ll explore five common prompting mistakes and show you how to fix them with concrete examples.

Mistake 1: Vague Debugging Requests

Description: Developer provides no context while asking AI to fix an error, forcing the AI to guess.

Realistic Scenario: Production API failing after deployment. Developer pastes only the error message without any surrounding code or stack trace.

❌ Wrong Prompt:

Fix this error: "NullPointerException"
Enter fullscreen mode Exit fullscreen mode

⚠️ Why it is wrong:
No stack trace, code context, or environment details to locate the actual null reference.

✅ Better Prompt:

Fix the NullPointerException in the below Spring Boot service method.

Environment: Spring Boot 3.2, Java 17, running on Kubernetes
Error stack trace:
java.lang.NullPointerException at com.example.UserService.getUserById(UserService.java:42)
at com.example.UserController.getUser(UserController.java:18)

Code snippet:
public User getUserById(String id) {
return userRepository.findById(id).get();
}

Expected behavior: Return a default User object or throw a custom NotFoundException when user doesn't exist.
Enter fullscreen mode Exit fullscreen mode

💡 What changed: Added stack trace pinpointing exact line, code context, and expected behavior so AI can suggest proper null handling.

Mistake 2: Asking for Code Without Context

Description: Requesting implementation without providing existing codebase patterns, leading to inconsistent style and architecture.

Realistic Scenario: Junior developer asks AI to implement authentication without knowing the project uses JWT with specific custom claims format.

❌ Wrong Prompt:

Write me authentication for a Spring Boot app
Enter fullscreen mode Exit fullscreen mode

⚠️ Why it is wrong: No information about existing auth mechanism, user store, or security requirements.

✅ Better Prompt:

Implement authentication for Spring Boot 3.2 using our existing pattern.

Our codebase currently uses:

JWT with custom claims: userId, tenantId, roles

User details loaded from PostgreSQL via Spring Security

OAuth2 client for Google login

Custom AuthenticationProvider for username/password

Add a new endpoint /api/auth/login that accepts email/password, validates against database, and returns JWT with our standard claims format. Follow existing package structure under com.example.auth.
Enter fullscreen mode Exit fullscreen mode

💡 What changed: Provided existing patterns, tech stack details, and specific requirements to generate consistent code.

Mistake 3: No Iteration After First Response

Description: Accepting first AI response without iterating to refine, test, or adapt to edge cases.

Realistic Scenario: Developer needs to parse large CSV files. AI gives initial solution that works for small files but fails with memory issues in production.

❌ Wrong Prompt:

How to parse a CSV file in Python?
Enter fullscreen mode Exit fullscreen mode

Developer copies first solution without further questions.

⚠️ Why it is wrong: AI provided a basic solution that may not handle large files, encoding issues, or malformed data.

✅ Better Prompt (Iterative Approach):

Round 1: "Write Python code to parse a 5GB CSV file efficiently. Use generators and avoid loading entire file into memory."

Round 2 (after review): "Now add error handling for malformed rows, skip corrupted lines, and log errors to a file."

Round 3: "Optimize for speed using pandas with chunksize. Include progress bar for processing 10M rows."
Enter fullscreen mode Exit fullscreen mode

💡 What changed: Iterative refinement addressed scale, error handling, and performance progressively.

Mistake 4: Ignoring AI’s Clarifying Questions

Description: AI asks for clarification but developer proceeds with incomplete information, resulting in incorrect output.

Realistic Scenario: AI asks which database technology is being used for a query optimization request. Developer ignores and uses generic advice.

❌ Wrong Prompt:

Optimize this query: SELECT * FROM orders WHERE customer_id = 123
Enter fullscreen mode Exit fullscreen mode

AI: "Which database are you using?"
Developer: (ignores) "Just optimize it"

⚠️ Why it is wrong: Optimization strategies differ vastly between PostgreSQL, MySQL, MongoDB, etc. Generic advice may even degrade performance.

✅ Better Prompt:

Optimize this query for PostgreSQL 15:

SELECT * FROM orders WHERE customer_id = 123;

Current execution plan shows sequential scan on orders table (500K rows). customer_id has low cardinality (10K distinct values). Write the appropriate index and rewritten query if needed.
Enter fullscreen mode Exit fullscreen mode

💡 What changed: Answered AI's implicit questions upfront—database type, current performance, data distribution.

Mistake 5: Using Outdated Examples in Prompts

Description: Using deprecated library versions or syntax in prompts, leading to AI generating outdated code.

Realistic Scenario: Developer asks for React code using class components in 2024, missing modern patterns.

❌ Wrong Prompt:

Write a React component with state and lifecycle methods
Enter fullscreen mode Exit fullscreen mode

⚠️ Why it is wrong: AI may generate class components with deprecated lifecycle methods instead of modern hooks.

✅ Better Prompt:

Write a React 18 functional component using TypeScript that:

Fetches user data from /api/users on mount

Shows loading state

Uses React Query for caching

Handles errors with ErrorBoundary

Follows project pattern: hooks in separate files under src/hooks/
Enter fullscreen mode Exit fullscreen mode

💡 What changed: Specified modern React version, TypeScript, preferred libraries, and project structure.

Summary & Best Practices

  • Always provide context: stack traces, code snippets, environment details, expected behavior.
  • Be specific about versions and libraries: avoid ambiguous “latest” or “modern”.
  • Iterate: treat AI as a pair programmer, not a one‑shot code generator.
  • Answer AI’s questions proactively to save time.
  • Keep prompts up‑to‑date with your actual tech stack.

With these techniques, you’ll turn AI from a generic code machine into a precise, context‑aware assistant.

Top comments (0)