Haim Michael

Spec-Driven Development (SDD) fundamentally changes how software is conceived, implemented, and evolved—especially in an era where AI agents are no longer passive assistants, but active collaborators. In this talk, I will share my hands-on experience implementing Spec-Driven Development using Amazon’s Kiro IDE, focusing on how AI agents are orchestrated throughout the development lifecycle to deliver production-ready software with clarity, traceability, and speed.

The session is grounded in real projects in which the Kiro IDE was used to drive development from explicit specifications rather than ad hoc prompts or informal requirements. I will demonstrate how structured specifications—written upfront and continuously refined—become the single source of truth that guides implementation, validation, documentation, and testing. Rather than “coding first and fixing later,” the workflow ensures that what we build consistently matches what we intended to build.

A central theme of the talk is agent orchestration. I will show how Kiro IDE enables developers to define and integrate custom AI agents directly into the IDE, each with a focused responsibility. Examples include agents that autonomously:

Generate and evolve automated tests based on specifications

Produce and maintain technical documentation aligned with the current spec.

Validate implementation decisions against the original requirements.

Assist with controlled refactoring while preserving spec compliance.

The audience will see that creating new agents is far simpler than commonly assumed. I will walk through the process of defining new hooks that lead to new autonomous agents running in our IDE in accordance with the specifications. These agents operate as first-class citizens inside the IDE, collaborating with both the developer and other agents.

The talk also provides a concise yet comprehensive overview of the Spec-Driven Development cycle as implemented in the Kiro IDE. This includes:

Translating intent into structured, machine-readable specifications

Using the spec as a steering document for agent behavior

Iterative implementation guided by continuous spec validation

Automatic generation of tests and documentation as part of the same flow

Maintaining long-term alignment between requirements, code, and artifacts

What makes this workflow unique is not just the use of AI, but the discipline imposed by the spec itself. AI agents are not free-form code generators; they are constrained, directed, and evaluated through explicit specifications. This results in systems that are easier to reason about, review, and evolve—especially in teams and long-lived codebases.

This session is aimed at developers, architects, and engineering leaders who want to move beyond prompt-driven coding toward a more reliable, scalable, and auditable development model. Attendees will leave with a clear mental model of agent orchestration in an IDE, practical examples they can apply immediately, and a deeper understanding of why Spec-Driven Development is emerging as a cornerstone methodology in the AI era.