​
Overview

Effective prompting is the foundation of AI-first crypto development. This guide covers proven approaches and recommended patterns that help generate production-quality code and handle complex blockchain scenarios efficiently.

​
Core approaches

​
Context layering

Build context progressively from general to specific: 
"I'm building a consumer crypto wallet using CDP embedded wallets and Next.js.

The app uses CDP Web SDK and OnchainKit components. Here's my current structure: [paste code]

Add a send money feature with email notifications for recipients without wallets."

​
Error-driven development

Use errors as learning opportunities: 
"I'm getting this error: [exact error message]

Code: [paste code]
Goal: [what I was trying to achieve]

Fix this and explain why it happened so I can avoid similar issues."
Key elements:
  • Exact error messages (don’t paraphrase)
  • Complete code context
  • Clear intent
  • Ask for explanations

​
Comparative analysis

Get AI to compare approaches: 
"I need to implement [functionality]. I'm considering:

Approach A: [description]
Approach B: [description]

Which would you recommend for [app type] and why? Consider security, UX, and complexity."

​
Crypto-specific patterns

​
Wallet integration

"Integrate [wallet type] with my [app description].

Requirements:
- [user flows]
- [security needs]
- [network support]

Handle: connection states, network switching, error recovery, mobile compatibility."

​
Transaction flows

"Design a transaction flow for [action].

User journey: [steps]
Constraints: [gas limits, network]
Security: [validation requirements]

Include user feedback, error handling, and status tracking."

​
Build in stages

Start simple, then improve: First, get it working: 
"Create basic [feature] for CDP app. Focus on core functionality."
Then, make it better: 
"Improve this implementation: [paste code]
Add error handling, loading states, validation."
Finally, polish it up: 
"Optimize for production: [paste code]
Focus on performance, security, testing."

​
Ask for different perspectives

Get focused feedback by asking the AI to wear different “hats”: For architecture decisions: “Focus only on high-level design decisions and system integration.” For implementation: “Write clean, performant code. Assume architecture is decided.” For security review: “Review only for vulnerabilities and compliance issues.”

​
Best practices

​
Do

  • Be specific with requirements and context
  • Include exact error messages and code
  • Ask for explanations, not just solutions
  • Build complexity incrementally

​
Don’t

  • Use vague requests like “make it better”
  • Skip project context
  • Accept code without review
  • Rush through complex features
Apply these techniques in: