Researchers employed AI‑driven protein design to create custom proteases by grafting the catalytic cores of a zinc‑dependent crayfish digestive enzyme and a yeast cysteine protease onto entirely new scaffolds. Beginning with the known catalytic residues, they used RFdiffusion2 to generate protein backbones that positioned these residues correctly and accommodated a target cleavage sequence, then fed the backbones into ProteinMPNN to select amino acids that satisfied structural, catalytic‑site, and substrate‑binding constraints. For the cysteine protease they explored 12 backbone variants and 69 amino‑acid sets, yielding 13 functional designs (≈19 % success) that were less active than the native yeast enzyme. In the zinc protease project they first screened 135 designs; many lacked the essential water molecule, giving only 14 weakly active variants out of 89 tested. Adding a water‑position constraint raised success to 35 of 38 designs. Subsequent mutagenesis of residues near the active site boosted activity: the best cysteine variant accelerated peptide‑bond cleavage 109‑fold while retaining specificity, and an optimized zinc protease (v2) was 30 times more active than the original design, achieving a ~10¹⁰‑fold rate increase. Finally, they designed zinc proteases that successfully cleaved disease‑related proteins linked to Alzheimer’s and frontotemporal dementia in cells, demonstrating the approach’s therapeutic potential despite challenges in delivery and immune response.