Smart home automation is rapidly expanding, yet existing platforms face critical usability and safety limitations. AutoIoT requires users to photograph their physical IoT devices for system configuration, creating friction through image capture rather than natural text input. RuleEngine systems generate automation rules but lack post-generation conflict checking, allowing contradictory rules that cause unpredictable device behavior and safety risks. This paper presents an AI-powered automation framework addressing both challenges through natural language processing and mandatory semantic conflict detection. The system employs a multi-layered architecture: (1) a Retrieval-Augmented Generation layer accessing SmartThings device capabilities, (2) a Gemini-powered agentic AI that autonomously selects and orchestrates specialized tools, and (3) a Pinecone vector-based semantic search layer performing conflict detection before deployment. The workflow translates plain language requests into SmartThings-compatible JSON rules and validates them against existing automations. We present a proof-of-concept implementation demonstrating detection of three conflict categories: direct contradictions, temporal illogicality, and condition overlaps. When conflicts are detected, the system provides explainable recommendations with resolution strategies. This approach eliminates image-based device setup barriers while adding critical safety validation absent in traditional rule engines, enabling safer and more accessible smart home automation.