Two independent studies advanced computational antibody engineering: one presents an adaptive diffusion strategy to design humanized antibodies and nanobodies, and another demonstrates that including negative training examples materially improves antibody‑binding predictions. Both works were published in leading computational biology venues and aim to reduce experimental burden in therapeutic discovery. The diffusion approach maps sequence‑to‑structure constraints to generate candidate humanized leads, improving developability metrics in silico. The negative‑data study shows that models trained with confirmed non‑binders reduce false positives and improve downstream screening efficiency. Together, these methods suggest a practical pipeline: generate candidates using diffusion models, then triage with classifiers robust to negative data. Biotech teams and CDMOs will evaluate integration into discovery workflows to accelerate hit selection and reduce wet‑lab costs.