Skip to content
Road to Intelligence

Concept · Chapter 13: Agents

Agent Planning and ReAct

Must knowUnderstand7 minDifficulty

Agent planning organizes actions toward a goal and revises them as new observations arrive.

The problem

A plan made before inspecting the environment can become wrong after the first action.

The solution

Use short plans with observable checkpoints and update them from feedback.

The consequence

The useful plan is the one that directs the next evidence-gathering action.

You should understand first

  1. Text as Data
  2. Probability and Distributions
  3. Conditional Probability and Bayes' Theorem
  4. Probability of Sequences
  5. Language Modeling
  6. Vectors
  7. Dot Product
  8. Embeddings
  9. Attention
  10. Softmax
  11. Self-Attention
  12. Causal Masking
  13. Entropy
  14. Loss Functions
  15. Cross-Entropy Loss
  16. Autoregressive Next-Token Prediction
  17. Pretraining at Scale
  18. GPT-1 → GPT-2 → GPT-3
  19. In-Context Learning
  20. The Turing Test
  21. Symbolic AI
  22. Logic and Rules
  23. Expert Systems
  24. Knowledge Representation
  25. The Knowledge-Acquisition Bottleneck
  26. From Rules to Learning
  27. Supervised, Unsupervised and Self-Supervised Learning
  28. Expected Value and Variance
  29. Reinforcement Learning
  30. MDPs, Policies and Value
  31. Decoding: Greedy, Temperature, Top-k, Top-p
  32. One-Hot Encoding
  33. Tokenization
  34. Chat Templates
  35. System Prompts and Instructions
  36. Structured Outputs and Constrained Decoding
  37. Tool Calling
  38. LLM Agents
  39. Search
  40. Planning
  41. Agent Planning and ReAct

Open loop and closed loop

“Read, edit, test” is an initial plan. If reading reveals that the function is imported from another package, the plan must change. Closed-loop control uses the new observation; an open-loop sequence blindly continues.

ReAct interleaves generated reasoning, actions and observations. This does not require every modern agent to display private reasoning or produce a thought paragraph before each tool call.

Plan around uncertainty

For the cart bug, first establish whether the failure reproduces. After patching, establish whether normal inputs still work. Each check answers a question that affects the next action.

A checklist is not a proof. “Tests passed” needs a result tied to the current code, not a tick added because the agent intended to run them. Compare a planning system against a simpler loop at equal budget before attributing better results to its plan format.

What to remember

  • Agent planning organizes actions toward a goal and revises them as new observations arrive.
  • Use short plans with observable checkpoints and update them from feedback.
  • The useful plan is the one that directs the next evidence-gathering action.

Key papers

Essential

ReAct: Synergizing Reasoning and Acting in Language Models

Shunyu Yao et al. · 2022

Interleaves reasoning, actions and observations in language-model task solving.

How to read it: Compare the action-only and reasoning-only examples with the interleaved trajectory.

~35 min readarXiv:2210.03629✓ verified 2026-10-05

Watch