Inside the reason phase. How LLMs turn a goal into the next action, why chain-of-thought is just more tokens, and why your prompt is the agent's architecture.
