Iteratively improving search quality through repeated experimentation has historically required specialized teams. As a result, organizations could apply this process to only a limited number of applications. Agentic applications increase the demand for high-quality search because their ability to complete tasks reliably often depends on the context they retrieve. In this talk, we present an agentic optimization loop that measures search quality based on agent telemetry, builds hypotheses for improvements, runs experiments, evaluates the results, and chooses the next iteration. We cover the technical foundations and platform capabilities the loop requires, along with implementation options for OpenSearch. You walk away with architecture patterns you can deploy to your own environment.