The complexity of IoT device configuration can compromise security through misconfiguration and cross-app interface threats. Previous efforts have utilized Large Language Models (LLMs) for natural language policy translation but often lack automated mechanisms for systematic conflict resolution. We propose to fill this gap with a multi-agent, conflict-aware framework for the openHAB platform. This research preview introduces AgentHAB, which reengineers the authoring process. It consists of a Policy Generation Agent with a Validator Agent. The generator uses retrieval-augmented generation (RAG) from a curated corpus of openHAB grammar to draft rules. The validator agent then iteratively checks these rules against formal syntax and static constraints, forcing the generator to self-repair. Our initial research concludes: (1) the validation and refinement loop significantly improves syntactic validity over a single-shot baseline; (2) the pipeline is robust to paraphrased user requests; and (3) context retrieval is critical for correct generation. This work presents a “proof-of-concept” and a clear research plan for achieving safe, context-aware rule generation for the smart home. This work aims to spark discussion on reliable agentic reengineering for IoT policy validation.