Klemensas Mecejus will present a practitioner-focused session on testing AI systems after they leave the demo environment. The talk explains how enterprise and government teams can build quality gates for agentic workflows, retrieval systems, and model-powered decision support, covering test data design, acceptance criteria, human-in-the-loop checks, model risk governance, security and privacy reviews, monitoring, incident response, and value measurement. Drawing on AI71’s deployment experience across government, construction, defense, telecom, and regulated enterprise sectors, the session gives quality assurance (QA), test automation, and engineering leaders a concrete checklist for making AI reliable enough to operate in production workflows.