PI: Prof. Nauman bin Ali, Department of Software Engineering, Blekinge Institute of Technology
co-PI: Assoc. Prof. Emelie Engström, Department of Computer Science, Lund University
AI coding agents increasingly write code and tests, and make architectural decisions autonomously, yet their non-deterministic output of varying quality demands human oversight and deterministic control. This shift toward agentic development compresses the process of discovering and validating requirements into ephemeral developer–AI exchanges where assumptions about expected behavior are made but never verified against product requirements. i2T (Intent-to-Tests) capitalizes on developer–GenAI conversations to address this challenge by extracting developer intent, aligning it with existing requirements artifacts, and producing human-readable Gherkin specifications that stakeholders can inspect and challenge. When confidence in the extracted intent is insufficient, or conflicts between intent and requirements cannot be resolved, an evidence card routes the case to human review rather than silently propagating uncertainty. From those validated specifications, automated acceptance tests are generated, providing deterministic guardrails whose pass/fail logic is fully AI-independent, verifying the code and every subsequent change against product requirements.
Project number: H1