Agentic AI systems automate complex workflows but require extensive manual tuning. This paper presents a framework for autonomously optimizing Agentic AI solutions across industries, such as NLG-driven enterprise applications. It employs agents for Refinement, Execution, Evaluation, Modification, and Documentation, using iterative feedback loops powered by an LLM (Llama 3.2-3B). The system optimizes configurations without human input by autonomously generating and testing hypotheses, enhancing scalability and adaptability. Case studies demonstrate a significant boost in output quality, relevance, and actionability. Data, including original and evolved agent codes and outputs, are open-sourced.