Next Generation (NextG) cellular networks will be natively cloud-based and\nbuilt upon programmable, virtualized, and disaggregated architectures. The\nseparation of control functions from the hardware fabric and the introduction\nof standardized control interfaces will enable the definition of custom\nclosed-control loops, which will ultimately enable embedded intelligence and\nreal-time analytics, thus effectively realizing the vision of autonomous and\nself-optimizing networks. This article explores the disaggregated network\narchitecture proposed by the O-RAN Alliance as a key enabler of NextG networks.\nWithin this architectural context, we discuss the potential, the challenges,\nand the limitations of data-driven optimization approaches to network control\nover different timescales. We also present the first large-scale integration of\nO-RAN-compliant software components with an open-source full-stack softwarized\ncellular network. Experiments conducted on Colosseum, the world's largest\nwireless network emulator, demonstrate closed-loop integration of real-time\nanalytics and control through deep reinforcement learning agents. We also show\nthe feasibility of Radio Access Network (RAN) control through xApps running on\nthe near real-time RAN Intelligent Controller, to optimize the scheduling\npolicies of co-existing network slices, leveraging the O-RAN open interfaces to\ncollect data at the edge of the network.\n