Prompt Engineering for Voice Coders: How to Talk to AI Like a Senior Dev

The AI assistant doesn't care about your feelings. It doesn't understand context unless you provide it. And it will happily give you terrible code if you ask for it badly.

Voice coding amplifies both the benefits and the problems of AI assistance. When you're speaking, you tend to be vaguer, more natural, less precise. Sometimes that's great—it captures intent. Sometimes it produces garbage.

Let's talk about how to prompt effectively when you're speaking rather than typing.

The Three Laws of Voice Prompting

1. Context Before Request

Bad: "Write a login function."

Better: "I'm building a React app with TypeScript. We use JWT for auth and store tokens in httpOnly cookies. Write a login function that posts to /api/auth/login and handles the response."

When speaking, we often skip setup and jump to the request. Force yourself to establish context first. Spend 10 seconds on background to save minutes on clarification.

2. Constraints Are Your Friends

AI can generate infinite variations. Without constraints, you get generic solutions that don't fit your project.

Good constraints to speak:

• "Match the style of our existing codebase"
• "Keep it under 50 lines"
• "Don't use any external dependencies"
• "Make it work without JavaScript enabled"
• "Optimize for readability over performance"

3. Iterate Vocally

The first output is rarely perfect. That's fine—voice makes iteration cheap.

"That's close, but extract the validation into its own function."

"Can you add error handling for network failures?"

"Actually, let's use async/await instead of promises."

Each refinement gets you closer to what you need.

Magic Phrases That Actually Help

Through experimentation, I've found certain phrases consistently improve AI output:

"Think step by step" - For complex problems, this cues the AI to break down the problem rather than jumping to a solution.

"Consider edge cases like..." - Explicitly naming edge cases gets them handled in the code.

"In production, this would need to handle..." - Frames the request as production code, not toy examples.

"A senior developer would..." - Surprisingly effective at improving code quality. Role-setting works.

"Before writing code, list the steps" - Forces planning before implementation.

Common Voice Prompting Mistakes

The Vague Request

"Help me with the authentication."

What kind of help? Debugging? New implementation? Review? The AI will guess, often wrong.

The Assumption Bomb

"Fix the issue with the user service."

What issue? The AI doesn't see your error message. It doesn't know your codebase. Speak the specifics.

The Kitchen Sink

"Create a complete user management system with registration, login, password reset, email verification, role-based permissions, and admin dashboard."

Too much at once. Break it down. One piece at a time.

The Invisible Context

"Make it work like the other one."

The AI didn't see the other one. You have to describe what "like the other one" means.

A Template That Works

When I'm voice prompting something significant, I follow this structure:

1. Context: "I'm working on [project type] using [tech stack]."
2. Goal: "I need to [specific outcome]."
3. Constraints: "It should [requirements and limitations]."
4. Style: "Match the pattern of [existing code or convention]."

Example: "I'm working on a Node.js API using Express and TypeScript. I need to add rate limiting to the public endpoints. It should use Redis for distributed limiting across instances. Match the middleware pattern we use for authentication."

That prompt will get you 90% of the way there on the first try.

The Meta-Skill

Prompting well is about clearly knowing what you want. If your prompt is vague, it's often because your understanding is vague.

Use voice prompting as a forcing function for clarity. If you can't speak what you need, you probably haven't thought it through. The moment of articulation is often where the real design work happens.