Manual ReAct Trace: Walking Through One Query Step by Step

~20 min read

Running the ReAct loop manually — calling the agent, reading its Thought/Action output, executing the tool yourself, and feeding back the Observation — builds full visibility into how the loop actually behaves before automating any of it.

Manual ReAct Trace: Walking Through One Query Step by Step is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • Running the loop manually first gives full visibility into the agent's reasoning before trusting an automated controller
  • A blank ('') message tells the agent to continue reasoning from wherever it paused, without adding a new user question
  • The human plays the role of the controller in this manual version — actually calling the tool function and formatting the result as 'Observation: ...'
  • The Observation message must match the exact format the system prompt led the model to expect
  • This step-by-step trace is what the automated agent_loop() (next subtopic) replaces with code — same mechanics, no human in the loop