Large language model (LLM) agents offer huge potential to boost productivity and have attracted significant attention across academia and industry. However, engineering agentic systems poses significant challenges due to their inherent autonomy, non-deterministic behaviour, and increasing concerns around AI safety. In this talk, I will first present our responsible AI engineering approach for addressing system-level challenges, e.g., how to assess and mitigate AI risks throughout the entire lifecycle of AI systems using Responsible AI Question Bank and Responsible AI Pattern Catalogue. I will then highlight the unique characteristics of agentic systems and share our recent research progress in agent engineering, including the Agent Design Pattern Catalogue, the AgentArcEval method, and the Swiss Cheese Model for agent safety. Finally, I will examine the intersection of RE and agents, covering both how agents can support RE activities and how RE can be adapted to guide the development of safe agents. I will conclude key open challenges and future research opportunities at the intersection of RE and agents.