Instead of react-style one-step-at-a-time, planning agents form a plan first then execute. Here's when that wins.
ReAct (Reasoning + Acting) is the foundational agent pattern. Here's how it works and why most modern agents are a variation.
Reflection is having the agent critique its own work before finalizing. Improves quality at the cost of more tokens.
Self-correction lets agents recover from errors by detecting and fixing their own mistakes. Here's how it actually works in practice.
Tool use is the agent's ability to take actions in the world. Here's how modern tool-calling APIs work and what to watch out for.