Deep dives into specification-driven development, AI-native engineering, and the future of building software.
OpenSpec, Spec Kit, Kiro, and Augment Code all promise spec-driven development. But they disagree on something fundamental: how AI agents should interact with specifications. This split defines everything about how they work.
Most spec-driven tools use one AI agent wearing different hats. BMAD goes all in on multi-agent collaboration: an Analyst, Product Manager, Architect, Developer, QA, and Scrum Master, all working together. Here's when that matters and when it doesn't.
Both tools structure AI-agent workflows around specifications. But OpenSpec is a workflow engine with a philosophy. Spec Kit is a starter kit with templates. The differences run deeper than you might expect.
In the age of AI-generated code, the value isn't in the lines you write. It's in the specs you design, the architecture you choose, and the taste you bring to the process. Code is an implementation detail.
AI coding agents are fast, impressive, and wrong about half the time. Spec-driven development helps, but verifying that code actually matches the spec is still a leap of faith. IIC closes the gap with assertion integrity.
The software industry has a documentation problem. We write docs after the fact. We maintain wikis nobody reads. Meanwhile, AI coding assistants generate code at breakneck speed, but without a compass, speed is just chaos.