IoT deployments often face premature obsolescence due to evolving protocols, fragmented software stacks, hardware heterogeneity, and aging machine-learning models, leading to costly replacements. This talk explores whether agentic AI — combining large language models with autonomous agents — can sustain IoT systems by enabling automated lifecycle management that adapts code, models, and configurations as requirements, environments, and hardware evolve. It highlights feasibility and usability aspects of running language models on edge devices while addressing open questions around collaborative edge–cloud intelligence, adaptive deployments, and model aging, with a view toward the longevity and sustainability of future IoT infrastructures.

Commercial IoT products make the news all the time—and not always in the way we’d hope. From dramatic device failures to frustrated customers and the occasional data-privacy horror story, these issues don’t exactly help build trust in the IoT ecosystem. The challenge is clear: the IoT is wonderfully diverse and open, but that same freedom lets every actor choose their own path—sometimes leading straight into trouble. Longevity suffers, users suffer, and the field as a whole struggles to keep pace. In this open forum, we invite everyone who cares about making IoT systems last—researchers, practitioners, makers, and the simply curious. Let’s dig into the challenges together, share what’s going wrong (and right!), and explore how we can steer the future of IoT toward durability, reliability, and trust. Which research problems can we solve to make a real difference for IoT longevity? Can we jumpstart a community that makes a change?