Artificial intelligence (AI) is coming, also in the nuclear industry. Three years ago AI and nuclear was not mentioned in the same sentence, while now even regulators embrace the technology. At the last Regulatory Information Conference by the U.S. NRC in Washington DC in March, a human-AI session stated that «AI and autonomous technologies redefine operational boundaries». What are these boundaries, and are the same questions and challenges relevant for all of them? Will AI be possible in the loop, and in which loop? This paper will outline some of the operational boundaries and discuss dimensions that need to be addressed in some of them. Human-AI teaming can be seen in relation to Human-automation, and this will also be discussed. Classic dimensions in human-AI teaming, such as explainability, trust and the role of humans will be discussed. Then a main question is how can we do safety analysis of such human-technology joint systems. What does this implicate for PRA? Can we use traditional HRA methods for the human part? Can we analyse the AI part for itself, or should the human-AI teaming be analysed as one system? Which dimensions will be relevant for the various types of collaboration systems? The paper will outline a framework for addressing the right type of questions to various parts.