Spec-driven development with Kiro (Interview)

What It Covers

AWS launches Kiro, an AI coding environment using spec-driven development where engineers create specifications in markdown rather than typing code directly. Deepak Singh explains how this approach mirrors senior engineer workflows and addresses limitations of chat-based coding assistants.

Key Questions Answered

• •Spec-Driven Workflow: Kiro converts problems into three markdown artifacts: requirements in EOS format, design documents, and task lists. Engineers modify specifications rather than code directly, preserving agent context and enabling better collaboration. The agent generates all implementation code from these specifications automatically.
• •Context Preservation Strategy: Manual code edits destroy agent context and break synchronization in long sessions. Kiro maintains context through specifications, steering files, MCP servers, and hooks that execute automated tasks when files change. This approach prevents the context loss that creates user anxiety in other tools.
• •Enterprise Adoption Metrics: Eighty percent of Amazon developers now use AI tools regularly for code development, with some projects reaching ninety percent AI-generated code. Teams conduct sprint planning every four days instead of two weeks because they complete backlogs faster using agentic development systems.
• •Pricing Model Evolution: Kiro uses credit-based pricing at twenty, forty, and two hundred dollars monthly for two thousand, four thousand, and ten thousand credits respectively. The auto agent consumes credits thirty percent slower than Sonnet four. Real-time usage visualization helps users understand consumption patterns across prompts and tool usage.
• •Technical Architecture Requirements: Current models need one more generation before fully autonomous application development becomes reliable. Kiro integrates neurosymbolic AI techniques from the Hydro project to verify correctness of distributed systems mathematically, reducing dependence on human verification for complex implementations.