What if the devices around us could not only make intelligent decisions but also learn and evolve on their own? As computing becomes ubiquitous, the “computing everywhere” paradigm is reshaping how intelligence is embedded into everyday devices. This talk will deepen the transformative potential of this shift in driving the widespread adoption of Tiny Machine Learning (TinyML) in daily life. Going beyond traditional inference-only approaches, this talk will explore the challenges and opportunities of on-device learning in TinyML, addressing the constraints of memory, computation, and energy efficiency. The discussion will highlight adaptive mechanisms to optimize processing pipelines dynamically during inference, as well as lightweight and efficient on-device training strategies that enable continuous adaptation, personalization, and responsiveness to evolving environmental conditions.