AI Planning Systems

Introduction

AI planning = breaking a high-level goal into ordered steps before execution. The plan is a tree or DAG the agent follows, replanning when reality differs from expectation.

Beginner analogy: Like writing a project plan in Notion before starting, then ticking off tasks.

Understanding the topic

Core concepts to understand:

• ReAct — interleave Reasoning and Acting.
• Plan-and-Execute — plan first, execute second.
• Tree of Thoughts — explore multiple plans, pick the best.
• Replan when an observation breaks assumptions.

Syntax reference

Visual workflow / architecture:

Goal
  │
  ▼
[Plan] step1 → step2 → step3
  │
  ▼ execute step1
  │
  ▼ observe (unexpected result?)
  │
  ▼ REPLAN if needed
  │
  ▼ continue

Real-world use

Devin uses planning + execution. OpenAI o1/o3 models internally do tree-of-thought reasoning.

Best practices

• Always store the plan separately from execution logs.
• Allow easy human edits to the plan.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. What is ReAct?
• Q2. Difference between Plan-and-Execute and ReAct?
• Q3. What is Tree of Thoughts?