Recently, a new trend has gained momentum in which generative AI agents are enabled to independently perform tasks for users. This topic brief focuses on large action models (LAMs), which have a so-called neurosymbolic model architecture. This is a combination of symbolic and neural AI, which enables a new type of control of digital applications. In technical terms, LAMs mediate between human intention and digital execution, much like operating systems mediate between hardware and software. Neurosymbolic AI models are used in LAM-based agents. Only a few LAM-based agents are currently available on the market as apps or stand-alone devices. They are characterised by a simplified form of interaction for the human user and should be able to autonomously execute various chains of action in the digital space. In the future, it can be assumed that people will issue voice commands to their mobile devices to complete various tasks and then the integrated AI agents will scroll, tap and click themselves. It remains to be seen how data protection and the protection of privacy can be guaranteed if AI agents, as everyday assistants, create comprehensive data profiles of users by performing their tasks. In view of the speed at which AI systems are being further developed, it can be assumed that the AI agents currently known mark the beginning of a change in human interaction with digital technologies. The dynamic development of autonomous AI agents makes it difficult to assess the impact on individual problem-solving skills and on people's personal freedom of choice.