Towards Spec-Driven Test Automation: Part 1

Modern AI coding agents often generate both code and its corresponding tests from the same specification, which undermines the principle of independent verification. This practice allows ambiguities in the original specification to be interpreted identically in both the implementation and the tests, resulting in a test suite that cannot truly validate the code. To address this, a new spec-driven workflow separates duties by using two distinct AI agents. One agent builds the software based only on the requirements, while a separate agent generates tests from specific acceptance criteria, which the coding agent never sees. This automated separation of concerns, enforced by a new open-source tool, brings the rigor of safety-critical V&V processes to general software development.