ReAct (Reasoning + Acting):
Interleave thinking and action. Thought → Action → Observation → Thought. Simple and interpretable.
Plan-and-Execute:
First create a plan, then execute steps. Good for complex tasks. Revise plan if needed.
Reflection:
Agent critiques its own outputs. Improves accuracy at cost of latency.
LATS (Language Agent Tree Search):
Explore multiple reasoning paths. Backtrack on failures.
Interview question: "When use Plan-and-Execute vs ReAct?"
Plan-and-Execute for complex multi-step tasks. ReAct for simpler, reactive tasks.