Fusidic acid (FA), a tetracyclic triterpenoid, has been approved to treat methicillin-resistant Staphylococcus aureus (MRSA) infections. However, there are few reports about FA derivatives with high efficacy superior to FA, manifesting the difficulty of discovering the derivatives based on experience-based drug design. In this study, we employed a stepwise method to discover novel FA derivatives. First, molecular dynamics (MD) simulations were performed to identify the molecular mechanism of FA against elongation factor G (EF-G) and drug resistance. Then, we utilized a scaffold decorator to design novel FA derivatives at the 3- and 21-positions of FA. The ligand-based and structure-based screening models, including Chemprop and RTMScore, were employed to identify promising hits from the generated set. Ten generated FA derivatives with high efficacy in the Chemprop and RTMScore models were synthesized for in vitro testing. Compounds 4 and 10 demonstrated a 2-fold increase in potency against MRSA strains compared to FA. This study highlights the significant impact of AI-based methods on the design of novel FA derivatives with drug efficacy, which provides a new approach for drug discovery.