How I AI - Automated AI Interactive Review

This article describes an experiment where AI agents, powered by a goal-oriented prompt in Codex, autonomously interact with a deployed web application to find defects. The process integrates with GitHub workflows: stories and defects are tracked as GitHub issues, code changes are reviewed via pull requests, and the AI generates readable logs, defect reports, replication videos, and summaries. The author emphasizes using human-oriented tools (GitHub, CI pipelines) for AI-to-human communication, aiming to enhance team processes rather than replace them. While not yet calling it full testing, the approach automates interactive review and defect discovery, with all sessions documented in a public repository.