01·Problem: reasoning without acting vs acting without reasoning

Introduction

Before ReAct there were two separate worlds: models generating chains of thought (Chain-of-Thought, Least-to-Most) but incapable of invoking external tools — and action-only systems that clicked, searched, and executed commands without internal justification. This lesson analyses the limitations of both approaches, shows the specific error types they generate, and explains why their integration is architecturally inevitable for agents performing multi-step tasks in open environments.