A trend in the academic field is agglomerations among scholars to generate knowledge with a disruptive influence on science and technology; however, the benefits have not been fully substantiated. This paper analyzes over 660,000 papers on artificial intelligence published from 1961 to 2023. We propose a method to calculate the innovative capacity of disruptive knowledge based on the similarity of historical, current, and future keywords, finding that scholars who commence their scientific endeavors earlier possess a heightened capability for disruptive knowledge innovation as Dkc index. The analysis reveals that multiagglomeration scholars have the highest average number of publications and citations, followed by agglomeration-flow scholars. Moreover, a larger agglomeration results in a lower ability to disrupt and consolidate knowledge innovation. Multiagglomeration and agglomeration-flow scholars harm disruptive/consolidative innovations. However, as the agglomeration effect intensifies, these two types of scholars from the disruptive perspective and multiagglomeration scholars from the consolidation perspective have a diminishing marginal effect on innovation capacity. The agglomeration size acts as a partial intermediary in the Multi→Size→Dkc index from the dual perspective and as a full mediator in the Flow→Size→Dkc index from the disruptive perspective, but only with a direct effect from the consolidative perspective.