The field of molecular synthesis and representation learning is rapidly advancing, with a focus on developing innovative methods for efficient and scalable synthesis planning. Recent developments have centered around improving the accuracy and speed of retrosynthetic methods, with a shift towards quadratic complexity and the integration of fragmentation algorithms. Additionally, there is a growing emphasis on generating synthesizable molecules, with novel frameworks being proposed to explore the neighborhood of given molecules in the synthesizable space. The use of deep reinforcement learning and large-scale datasets is also becoming increasingly prominent, enabling the rapid production of diverse and potent candidates for antibiotic discovery and other applications. Noteworthy papers in this area include: FragmentRetro, which introduces a novel retrosynthetic method leveraging fragmentation algorithms to achieve quadratic complexity. ReaSyn, which proposes a generative framework for synthesizable projection using a chain-of-reaction notation and achieves state-of-the-art performance in synthesizable molecule reconstruction and optimization. ApexAmphion, which presents a deep-learning framework for de novo design of antibiotics that couples a protein language model with reinforcement learning and exhibits a 100% hit rate in vitro. MolPILE, which provides a large-scale and diverse dataset for molecular representation learning, addressing the pressing need for a standardized resource in molecular chemistry. FragAtlas-62M, which introduces a specialized foundation model trained on the largest fragment dataset to date, achieving unprecedented coverage of fragment chemical space.