What is voice-first AI development?

Voice-first AI development is an approach where developers interact with their IDE through natural conversation rather than typing. Amazon Nova 2 Sonic enables this by listening to developer discussions, asking clarifying questions, and generating steering files and code suggestions in real-time. This mimics how developers naturally collaborate by talking through problems, but with AI actively participating in the conversation.

How does Nova Sonic differ from traditional AI coding assistants?

Nova Sonic introduces three key innovations: 1) Asynchronous processing that continues accepting input while tools run, 2) Barge-in capability allowing developers to interrupt and redirect the AI mid-response, and 3) Simultaneous multi-tool handling. Unlike traditional assistants that wait for complete prompts, Nova Sonic maintains an ongoing conversation flow that better matches human collaboration patterns.

What are Kiro steering files?

Kiro steering files (product.md, tech.md, and structure.md) provide essential context for AI-assisted development. They capture the project's purpose, technical specifications, and architecture decisions in markdown format. The demo showed how Nova Sonic can generate these files through conversation rather than manual creation, reducing setup time from hours to minutes while ensuring alignment between product goals and technical implementation.

Can this workflow handle complex technical decisions?

Yes, the demo demonstrated Nova Sonic handling nuanced technical discussions about frontend frameworks (Next.js 14 with React 18), backend architecture (Node.js 24 with Express 5.2), database choices (DynamoDB with single-table design), and even coding conventions (camelCase vs PascalCase). The AI captured these decisions in the steering files along with best practice recommendations like proper error handling and API design patterns.

How does the system maintain state during development?

The architecture uses a state.json file that tracks all decisions, open questions, and resolved items. This allows the system to recover seamlessly if the connection drops and provides developers with a complete audit trail of their conversation-driven development process. The demo showed how previously discussed items like success metrics and naming conventions were automatically marked as resolved in this state file.

What types of projects benefit most from this approach?

Voice-first development excels for greenfield projects where requirements are being defined (like the fitness photo journal in the demo), complex systems requiring alignment between multiple technical decisions, and projects where rapid iteration is valuable. The approach reduces the friction between idea formulation and implementation, particularly for teams that naturally brainstorm through conversation.

How does this compare to traditional documentation processes?

Traditional documentation often becomes outdated because it's created separately from the development process. This conversational approach embeds documentation creation directly into the development workflow, ensuring steering files always reflect current decisions. The demo showed how Nova Sonic captured technical details that developers might omit in manual documentation, like specific version requirements and architectural anti-patterns to avoid.

How can GrowwStacks help implement voice-first AI development?

GrowwStacks specializes in implementing conversational AI development workflows tailored to your team's needs. We can configure Nova Sonic with your preferred tools and frameworks, establish optimal steering file templates, and integrate the system with your existing development environment. Our solutions typically reduce initial project setup time by 60-80% while improving documentation quality and technical alignment.

Voice AI AI Agents Developer Tools

February 20, 2026 9 min read AI Development

Voice-First AI Development: Building Conversational Coding Agents with Nova Sonic & Kiro

Most developers waste hours manually translating conversations into documentation and code. Amazon Nova 2 Sonic with Kiro turns natural discussions into steering files and implementation in real-time, cutting project setup from days to minutes while capturing technical decisions most teams forget to document.

Voice-first AI development with Nova Sonic and Kiro creating steering files from conversation

The Voice-First Development Revolution

Developers naturally solve problems through conversation - whiteboarding sessions, pair programming discussions, and design meetings. Yet traditional IDEs force this knowledge into typed prompts and manual documentation, losing nuance and wasting time. The demo shows a better way: Nova Sonic listens to unfiltered developer conversations and actively participates by asking clarifying questions, just like a skilled technical teammate would.

At 2:15 in the video, you'll see the pivotal moment where Nova Sonic interrupts a vague technical decision with "What are the core user journeys you envision?" This single question transforms an abstract discussion into concrete requirements captured in the steering files. The AI doesn't just transcribe - it elevates the conversation quality.

85% of technical debt originates from undocumented decisions made during early development conversations. Voice-first AI captures these critical choices as they happen, baked directly into your project's foundational files.

Nova Sonic's Breakthrough Capabilities

Traditional AI coding assistants force developers into an unnatural start-stop rhythm. Nova Sonic introduces three game-changing capabilities demonstrated between 4:30-6:45 in the video:

1. Asynchronous Processing

Continues accepting input while tools execute, maintaining conversation flow. When discussing backend architecture at 5:20, the team adds new constraints about error handling while previous suggestions are still processing.

2. Barge-In Functionality

Allows natural interruptions to correct or redirect the AI. At 6:10, the developer interjects to specify "Node.js 24 and Express 5.2" mid-explanation, and Nova Sonic seamlessly incorporates this precision.

3. Multi-Tool Parallelism

Handles simultaneous documentation updates, code generation, and architecture decisions. Notice at 7:30 how it manages repository structure while simultaneously capturing API design conventions.

Conversational velocity increases 3× compared to traditional AI assistants because developers don't have to wait for completion before providing additional context or corrections.

Watch the Full Tutorial

See the complete 18-minute demonstration of Nova Sonic and Kiro in action, from initial conversation to fully generated steering files and running application. Pay special attention to the 12:45 mark where the team reviews the automatically generated tech.md file containing all their technical decisions with surprising depth and accuracy.

Full tutorial: Building voice-first AI agents with Nova Sonic and Kiro

Ready to Transform Your Development Process?

Every day without voice-first AI means lost hours translating conversations into code. Our team will implement Nova Sonic with your preferred tools and workflows in under 2 weeks. Book a free consultation to see how we can cut your project setup time by 60-80%.

Book Free Consultation → Read More Articles