Summary
Erik Hanchett, senior developer advocate at AWS with 15+ years of software development experience, explains spec-driven development — a workflow where structured markdown specification documents are authored before any code is written — and demonstrates how it maps onto modern AI coding assistants. The central analogy frames LLMs as eager AI interns who need explicit guidance; without a spec, they tend to go off the rails on complex, multi-step projects.
The session introduces Kiro, AWS’s newly launched AI coding environment available at kiro.dev, which bakes spec-driven development into both an IDE and a CLI. Kiro’s “spec mode” takes a natural language prompt and generates three artifacts in sequence: user requirements, a design document, and a task-by-task implementation plan — all before writing a single line of code. Hanchett notes Kiro went viral on release, accumulating tens of thousands of downloads fast enough that AWS had to temporarily gate access; it is now publicly available. He also covers GitHub’s open-source Spec It tool as an alternative for teams using other coding assistants.
Best practices discussed include keeping “steering docs” (agents.md or claude.md files) in a Goldilocks zone of context — specific enough to guide behavior but not so dense they degrade output — and using skills files for reusable task-specific instructions. Hanchett argues spec-driven development applies equally to greenfield projects and existing codebases with years of history.
📺 Source: AI Engineer · Published June 28, 2026
🏷️ Format: Tutorial Demo
