Speakers

Title: Agent Programming in the Generative AI Era

Abstract:

Historically, agent programming and agent-oriented programming languages have played an essential role in the engineering of agent and multi-agent systems across diverse domains, from software agents to robotics. Their main contribution has been to provide proper levels of abstraction for designing and developing agents, as well as rigorous theoretical foundations and tools to detect and prevent errors, verify behaviours and interactions. This role, however, has been challenged over the past decade. Today, the tremendous achievements in generative AI have led to the emergence of agentic AI frameworks and technologies that enable the development of LLM-based autonomous agents and multi-agent systems,  apparently without the need for explicit abstractions, architectures, or programming languages. In just a few years, these technologies have achieved widespread diffusion and adoption, making 2025 the “year of AI agents.” In this invited talk, I will explore this evolving landscape and argue for novel approaches to engineering agents and multi-agent systems that, on the one hand, preserve the benefits of “Good Old-Fashioned Agent Programming” and “Good Old-Fashioned Agent-Oriented Software Engineering,” and, on the other hand, effectively harness the new capabilities brought by generative AI.

University of Southampton, UK.

Title: Neuro-Symbolic Decision-Making in Autonomous Agents

Abstract: Neuro-Symbolic AI brings together the strengths of machine learning, such as neural networks and deep learning, with the interpretability and structure of symbolic AI, including rule-based reasoning and cognitive agent models. Recent advances in both domains have revitalised this area, enabling progress in demanding applications like robotics and autonomous driving. In this talk, I will explore how these applications motivate the integration of learning and reasoning, outline a dual-system perspective on neuro-symbolic decision-making, and present recent work combining Belief-Desire-Intention (BDI) agents with pre-trained machine learning models to achieve joint high-level control of self-driving vehicles in the CARLA simulator.