The field of Large Language Model (LLM) agents is moving towards a more integrated approach, combining the strengths of symbolic and neural reasoning to create more reliable, explainable, and governable AI agents. This shift is driven by the need to address fundamental architectural problems in current LLM agents, such as entangled reasoning and execution, memory volatility, and uncontrolled action sequences. Researchers are exploring new architectures and mechanisms that explicitly separate agent cognition into distinct phases, applying symbolic constraints to probabilistic inference to preserve neural flexibility while restoring the explainability and controllability of classical symbolic systems. Another area of focus is the development of episodic memory architectures that store and retrieve past workflows to guide agents in suggesting plausible next tasks, enabling more effective human-AI co-creation in scientific workflows. Additionally, there is a growing interest in automating agentic workflow generation via self-adaptive abstraction operators, which can improve generalization and scalability. Noteworthy papers include: The Structured Cognitive Loop (SCL) architecture, which introduces a modular approach to agent cognition and achieves zero policy violations and complete decision traceability. The $A^2Flow$ framework, which proposes a fully automated approach to agentic workflow generation based on self-adaptive abstraction operators and achieves significant performance improvements over state-of-the-art baselines. The Ivy AI coaching system, which combines symbolic and LLM models to deliver structured explanations and improves the pedagogical value of AI-generated explanations in intelligent coaching systems.