<p>Coalition formation in Agentic AI systems refers to the grouping of autonomous agents performing tasks with no or less human intervention. A set of disjoint coalitions forms a coalition structure. In hedonic games, agents’ preferences over coalition structures are constructed based on their total utility to be in coalition with the other agents. In this paper, we define a decentralized preference system where agents have a positive impact score for the tasks. Instead of total utility, agents’ preferences depend on the total impact score in their coalition formed for a task. On the other hand, the individual stability of a coalition structure is a state which is immune to an agent’s move that increases its total utility without decreasing for any other agents in the coalition where it moved. We define majority stability in the decentralized preference system, where a majority-stable coalition structure is immune to a movement that increases or keeps the same total impact score of the majority of agents in the two coalitions involved in the move, irrespective of the changes for the moving agent. The majority stability in a coalition structure also applies to the trading of two agents. We characterize a majority stable coalition structure, show its limitations under the decentralized preference system, and define two fair variants of it. A set of polynomial-time algorithms is proposed to verify whether a coalition structure is majority stable or one of its variants. Some analysis of the experimental results with synthetic instances is presented.</p>