The paper claims that traditional postphenomenology has not fully acknowledged the significance of breakdowns and anomalies in technology, as it is overly influenced by Heidegger's tool analysis and the dichotomy between ready-to-hand and present-at-hand. This approach means that classical postphenomenology prioritizes the aspects of usage and functionality, dismissing a broken artifact as mere detritus irrelevant to technological mediation. Therefore, to truly grasp the role of breakdowns and anomalies within technological mediation, a shift in the conceptual framework is necessary. Consequently, the paper suggests improving the concept of technological mediation with the idea of immunization, as developed by Sloterdijk and Esposito. This shift in perspective allows us to a) move beyond the Heideggerian dichotomy, b) reassess the preeminence of usage and functionality, and c) recognize the boundaries of technological mediation. The paper refers to some of these boundaries as "anti-mediations," which are cases of unbalanced or excessive mediation. It posits that in the realm of AI, breakdowns and anomalies are not just setbacks but invaluable moments that safeguard against the pitfalls of anti-mediation. This viewpoint significantly enriches our comprehension of AI, spotlighting its role not merely as a tool but as a dynamic participant in shaping the balance between effective mediation and overreach. By recognizing the constructive role of errors and disruptions in AI systems, we gain profound insights into how AI can evolve and be steered towards fostering more meaningful human-technology interactions, steering clear of the risks associated with anti-mediation.