Adeno-associated virus (AAV) has emerged as a leading platform for gene therapy. However, to unlock the full potential, their manufacturing yields, stability, and efficacy must all be improved. Rational design is limited by poor predictability and the potential impact of mutations on multiple important vector properties. Directed evolution requires little knowledge or predictability, but requires large libraries, and lacks appropriate selection methods to ensure that all properties needed for successful translation are achieved. Here we bring semi-rational design to AAV capsid engineering, combining co-evolutionary coupling and consensus-based design. We designed 110 mutants at sites with different degrees of coupling and mutated them either back to consensus or with various residue similarities according to BLOSUM62. Seven were selected based on their yields and transduction efficiencies before and after heat treatment. Our results demonstrate that mutations at sites within coupled networks exhibited the highest success rates. The BLOSUM62 selection strategy and consensus design approaches had similar success rates. K-means clustering of mutants further revealed that successful mutants clustered in regions with moderate coupling scores and positive BLOSUM62 values. This study highlights the potential of using evolutionary coupling analysis to guide AAV capsid engineering, suggesting that conservative mutations informed by evolutionary data can improve AAV stability and efficacy.