Solid-state quantum devices are inherently beset by uncertainty: Hamiltonian parameters are known only to within fabrication tolerances, and devices couple to spurious degrees of freedom such as two-level-system (TLS) defects. Operating them therefore requires fast, reliable ways to learn their properties. I will present two efforts toward this goal. First, I will describe an autonomous, high-throughput pipeline for detecting TLS defects in Josephson junctions. This has enabled correlating defect occurrence with fabrication parameters and atomic-resolution microscopy, revealing that TLS density tracks electrode microstructure, and that modified electrode fabrication can reduce it. Second, I will show how the initial tune-up of a superconducting qubit can be cast as a single optimization problem rather than a sequence of conditional calibration steps. Using one time-domain Rabi measurement and a suitably designed cost function, we jointly find drive and readout parameters in a way that scales favorably with parameter uncertainty. These results illustrate how autonomous, data-driven characterization can both accelerate and enable discovery in quantum hardware.