Artificial intelligence is undergoing a profound transformation. For much of its history, AI systems functioned primarily as analytical models: they classified images, predicted outcomes, or detected patterns in data. Today, however, a new generation of systems is emerging. AI agents built on large language models can plan tasks, access tools, interact with data environments, and coordinate actions across digital systems.\r\nIn this shift, AI is moving from model to actor.\r\nThis transformation has far-reaching implications. As AI systems begin to act autonomously in digital infrastructures—interacting with humans, organizations, and other agents—they become embedded in complex social systems rather than merely analysing them.\r\nRecent developments in AI agent architectures illustrate this transition. By combining large language models with memory, planning mechanisms, and tool access, AI agents are able to pursue goals, interact with data sources, and execute sequences of actions across digital environments. These capabilities blur the traditional boundary between analytical models and operational decision systems.\r\nRather than viewing AI purely as a technological capability, we may need to understand it as a new class of actors embedded in socio-technical systems. This perspective opens a new research frontier for computational social science: studying and shaping societies in which human and artificial agents increasingly interact, collaborate, and co-evolve.\r\nFrom this perspective, the challenge is no longer only to evaluate models, but to understand and govern agent behaviour, interactions, and decision environments. As organizations begin to deploy AI agents within workflows, platforms, and information infrastructures, questions of transparency, accountability, and coordination become central.\r\nThe talk argues that computational social science—and particularly traditions such as agent-based modelling—offers valuable conceptual tools for analysing this emerging landscape. Concepts such as interaction rules, bounded rationality, emergence, and multi-agent dynamics may become essential for understanding the behaviour of AI agents operating within complex social systems.