Living Spec

Traditional spec-driven development writes requirements up front, then code is written against them, and the two often drift apart. A living spec is designed to close that gap during agentic execution. When multiple agents work in parallel on different parts of a task, the living spec acts as shared ground truth: each agent reads it to understand scope and writes back to it as work completes, so the document always describes the current state of what is built and what remains.

The concept gained traction alongside multi-agent coding tools like Augment Intent, where a Coordinator agent generates the initial living spec from a developer's prompt, Implementor agents update it as they complete tasks, and a Verifier agent uses it to check that results match the original intent. The key property is that the spec is machine-readable and machine-writable, not just a document for humans.

For builders, a living spec solves a real coordination problem: when five agents are running in parallel, how do you know what each one is doing and whether the whole thing still adds up? The living spec is the answer. It is also the natural artifact to hand a human reviewer at the end, because it tells the story of what was planned, what changed, and why.