A new AI agent called Instinct has been opened to a first group of testers, who describe it as promising. Early reports also point to issues that are already becoming apparent during these initial trials. Details of what the agent is designed to do and the nature of the problems remain limited in the report.
A newly posted arXiv paper introduces Consort, a framework that governs AI coding agents through written specifications, enforced testing, and work on isolated copies of live databases. It frames such scaffolding as a control mechanism for agents whose output can vary unpredictably, ensuring code passes verification before it is merged. The work follows the rapid growth of spec-first tooling like GitHub Spec Kit, obra/superpowers, and BMAD.
A new arXiv paper introduces FaultLens, a method for constructing small behavioral test suites that validate automatically generated operational programs. It targets the gap between sparse hand-written examples, which can overlook boundary and interaction faults, and exhaustive regression suites, which are costly. The approach aims to catch faults efficiently without running full regression coverage.