Agentic Reasoning for Large Language Models

Solid survey. The breakdown of in-context vs post-training reasoning is a helpful lens. Curious to see how the "long-horizon interaction" challenges they flag get addressed in practice.