AI researchers are exploring more effective methods for reasoning in artificial intelligence beyond the traditional large language models (LLMs), which often focus on predicting the next token in a sequence. The emerging approach, called Large Reasoning Models (LRMs), incorporates intermediate "thinking" tokens and uses reinforcement learning to optimize for correct outcomes, but faces challenges such as inefficiency and lack of robust generalization. Current LRMs struggle with transparency, control, and deterministic outputs, which are critical for enterprise applications. An alternative method involves planning in the space of actions, allowing systematic exploration of action sequences rather than just token sequences, employing decision-theoretic planning, and integrating human users throughout the process. AI21, for instance, is developing a planning and orchestration system called Maestro to enhance the performance of LLMs by combining them with verifiers and employing parallelism, ultimately aiming for more reliable and adaptable AI systems.