Descriptions:
Al Harris, Principal Engineer at Amazon, presents Kiro—Amazon’s agentic IDE built around spec-driven development—just days after its general availability launch. The talk explains the methodology Kiro enforces and demonstrates the full SDLC workflow the tool is designed to compress.
Kiro’s central thesis is that vibe coding, while powerful, offloads too much structure onto the developer’s prompts and judgment. Spec-driven development instead synthesizes a prompt into formal requirements using the EARS (Easy Approach to Requirements Syntax) format—a structured natural language notation that maps directly to verifiable system properties. With the GA release, those EARS requirements can now be automatically translated into property-based tests (using frameworks like Hypothesis for Python or fast-check for Node), giving teams an automated correctness harness derived from their initial intent rather than hand-written afterward. Harris emphasizes a deliberate long-term goal: use LLMs for as little as possible over time, replacing them with classical automated-reasoning techniques wherever determinism and reliability matter.
The live demo walks through building a LangGraph-based agent with S3 checkpointing: Kiro ingests a prompt, refines requirements, produces a design document, extracts correctness properties, and generates tests—all in a tight inner loop. Engineers moving beyond solo vibe-coding to team-scale, auditable AI development will find Kiro’s approach a direct response to code-quality and maintainability concerns that have emerged as agentic coding matures.
📺 Source: AI Engineer · Published January 09, 2026
🏷️ Format: Keynote Launch
