Build specs through automated multi-agent verification and review loops

A developer writes a spec, then lets an implementation engine build it with review rounds from different agents and models, triaging issues and looping until the code meets spec.

Done with a custom-built or unnamed agent

What they did

The developer defines the product behavior and discusses the technical approach up front, then captures it in a spec. That spec goes to a durable implementation engine, which builds it. Automated verification runs, and different agent types on different models review the code. Issues are grouped into clusters and checked for false positives before being fixed. The developer reviews only code organization and structure.

How it went

The loop repeats until the code matches the spec. The commenter gives no numbers or timings, only the view that single-pass builds, even large ones, come out full of issues.

Worth knowing

The source says the method has no secret: models need repetition and multiple review passes, and it doesn't describe the engine itself or how to set one up.

Source: HN · Oct 16, 2026