Agents105 · Module I · Lesson 02 of 13
Article · 12 min

Planning

How agents decompose tasks.

Summary

Planning is the agent's ability to decompose a goal into steps. Naive planning generates the full plan up front; adaptive planning revises as new information arrives. The choice defines the failure modes.

Objectives
  • 01Distinguish explicit from implicit planning.
  • 02State the tradeoff between plan depth and adaptability.
  • 03Recognize when planning is over-engineered for the task.
The Lesson

Approaches

Explicit planner-executor: separate LLM call for the plan, then execute each step. ReAct-style: interleave reasoning and action step-by-step. Tree-of-thoughts: explore multiple plans in parallel, pick the best. Increasing complexity, increasing cost, decreasing marginal gain.

When planning is over-engineered

For simple two-step tasks (search then summarize), explicit planning wastes tokens. Reach for planning when the number of steps is unknown, when steps depend on earlier results, and when execution failures require replanning.

Key Ideas
  • Adaptive planning beats one-shot plans on long-horizon tasks.
  • Simple tasks do not need planning.