Microsoft opened an open‑source tool that lets engineers write AI test cases in plain English.
The framework, called Adaptive Spec‑driven Scoring for Evaluation and Regression Testing, debuted on Tuesday. It accepts text prompts that describe expected model behavior and automatically generates evaluation suites. The code is published on GitHub under an MIT license, and Microsoft provides sample specs for large language models and vision systems.
For developers, the main gain is speed. Instead of hand‑coding test inputs and expected outputs, a short description can be turned into a runnable test suite in minutes. That lowers the barrier to systematic regression testing, which has been a pain point as models grow in size and complexity. Teams can catch drift or unintended responses earlier, potentially saving weeks of debugging.
The tool joins a growing catalog of Microsoft‑backed open‑source AI utilities, but it’s still early days; real‑world adoption will depend on how well the generated tests mirror nuanced user expectations.