This paper explores the use of reinforcement learning (RL) to automatically discover fault-tolerant quantum circuits for logical state preparation. The authors propose a novel approach that integrates the tasks of logical state preparation and verification circuit synthesis, aiming to find circuits with fewer gates and ancillary qubits compared to existing methods. They demonstrate that their RL-based method can discover circuits with up to 15 physical qubits, outperforming other approaches in terms of circuit size and efficiency. The paper also discusses the robustness of the RL method under different qubit connectivities and gate sets, showing that restricting the actions of the RL agent based on hardware constraints during training is more effective than transpiling circuits for all-to-all qubit connectivity. Additionally, the authors highlight the potential of transfer learning to accelerate the discovery process for similar but different quantum circuit problems. The work opens the door to using RL for various fault-tolerant quantum circuit tasks, including magic state preparation, logical gate synthesis, and syndrome measurement.This paper explores the use of reinforcement learning (RL) to automatically discover fault-tolerant quantum circuits for logical state preparation. The authors propose a novel approach that integrates the tasks of logical state preparation and verification circuit synthesis, aiming to find circuits with fewer gates and ancillary qubits compared to existing methods. They demonstrate that their RL-based method can discover circuits with up to 15 physical qubits, outperforming other approaches in terms of circuit size and efficiency. The paper also discusses the robustness of the RL method under different qubit connectivities and gate sets, showing that restricting the actions of the RL agent based on hardware constraints during training is more effective than transpiling circuits for all-to-all qubit connectivity. Additionally, the authors highlight the potential of transfer learning to accelerate the discovery process for similar but different quantum circuit problems. The work opens the door to using RL for various fault-tolerant quantum circuit tasks, including magic state preparation, logical gate synthesis, and syndrome measurement.