Over the past thirty years, online video has evolved from simple file downloads on personal computers to large-scale streaming enabled by widespread broadband internet. Building on the authors’ experiences at YouTube, which launched in 2005, this article describes the key technologies behind large online video services like YouTube. Any such survey cannot be comprehensive. We seek to provide the reader an overview of the key building blocks and industry innovations along with observations of emerging technology trends and challenges that will shape the years to come.
Every major era of Internet Computing has been shaped by a technological shift that made complex distributed systems more manageable. This article argues that recent advances in agentic AI are driving the next such transformation: AI as the systems engineer. Internet infrastructures will increasingly rely on intelligent agents that act as AI systems engineers, translating human intent into coordinated engineering actions across cloud, edge, IoT and emerging computing platforms. As these agents assume operational roles in deployment, monitoring, optimization, resilience and sustainability, human engineers will focus more on supervision, governance and strategic design. This perspective outlines the opportunities and open research challenges that will shape the next generation of Internet Computing.
Industrial research is like trekking across shifting sand dunes — technologies rise, evolve, and disappear faster than any single skill set can keep pace with. Unlike academic research, which allows sustained focus in an area, industrial research is governed by practical forces: the cost of adoption, commercial strategy, and the pressure to deliver real-world impact. Over three decades at the IBM Watson Research Center, I watched wave after wave of innovation reshape the landscape beneath my feet. Each breakthrough opened new possibilities; each disruption forced a reassessment of skills, assumptions, and priorities. That volatility is simultaneously exhilarating and unforgiving. This article traces my career through those shifting conditions — what they taught me about technical innovation, organizational dynamics, and personal reinvention. From that journey, I have distilled eight lessons about adapting and growing in an environment where change is not the exception. It is the only reliable constant.
Low Earth orbit (LEO) satellite constellations are becoming orbital edge computing platforms that can sense, process, and learn near data sources. Federated Learning (FL) is promising in this setting because it supports repeated model adaptation without forcing every raw observation to be downlinked, centralized, or redistributed. Yet orbital FL is not terrestrial FL placed in space: it must operate over shell- and plane-structured constellations, sparse ground-station access, time-varying inter-satellite links, heterogeneous onboard hardware, limited energy and thermal budgets, and delayed or noisy labels. This article connects representative architectures, mission use cases, and open challenges for GS-centric, GS-assisted, and fully on-orbit FL.
