Executive summary This paper examines the development and functionality of AI agents – and the implications of their use – amid rapid advances in large language and multimodal models. Defined as autonomous systems that sense and act upon their environment to achieve goals, artificial intelligence (AI) agents are being deployed in a wide range of roles in different industries. This requires the adaptation of governance frameworks to ensure responsible adoption. AI agents, comprising components such as sensors and effectors, have evolved from rule- based systems to advanced models capable of complex decision-making and independent operation. Enabled by breakthroughs in deep learning, reinforcement learning and the transformer architecture,1 AI agents span applications from workflow automation to personal assistants. This progression now encompasses more sophisticated utility-based AI agents that incorporate memory, planning and tool integration, broadening their capabilities and relevance. The benefits of AI agents include productivity gains, specialized support and improved efficiency in sectors such as healthcare, customer service and education. However, AI agents also present novel risks, including potential misalignment,2 along with ethical concerns about transparency and accountability. Future advances in the area are likely to involve multi-agent systems (MAS), where AI agents collaborate to address complex challenges such as urban traffic management. More advanced systems introduce new demands for interoperability and communication standards to function effectively, while these protocols still need to be debated and agreed upon by a wider community. This paper highlights the need for robust governance, ethical guidelines and a cross-sectoral consensus to integrate AI agents safely into society. As more advanced AI agents continue to proliferate, it is imperative that their transformative potential remains balanced with essential safety, security and governance considerations. Navigating the AI Frontier: A Primer on the Evolution and Impact of AI Agents 4

Navigating the AI Frontier 2024