Computational antibody design advanced on two fronts: an adaptive diffusion strategy for humanizing and designing therapeutic antibodies and a separate study highlighting the essential role of negative training data in predicting antibody binding. The diffusion approach offers a probabilistic generative path to propose human-like sequences for nanobodies and monoclonals, easing downstream humanization. The negative‑data study demonstrated that including non‑binding examples materially improves model precision and reduces false positives in silico screening. Together, the papers argue for paired innovation in generative models and dataset curation to accelerate reliable antibody discovery. Biotech teams that combine advanced generative models with curated negative datasets may shorten lead selection timelines and reduce wet‑lab wasted effort, but they will still require rigorous biophysical and functional validation.