ReAct, think/act/observe cycles, agent steps, and trajectories — the core of how agents operate.
Key vocabulary
ReAct = Reason + Act — the agent thinks before each action.
Agent step — one full think/act/observe iteration.
Trajectory — the complete sequence of steps an agent took to complete (or fail) a task.
Stuck loop — when an agent repeats the same action without progress.
Max steps — a hard limit on how many steps the agent may take.
0 / 5 completed
1 / 5
A colleague says: “Our agent follows the ReAct pattern.” What does this mean?
ReAct = Reasoning + Acting. In each iteration the agent reasons, acts, then observes. This “think before you act” structure improves reliability over purely reactive agents.
2 / 5
In the think/act/observe loop, what happens during the observe phase?
Observe = receiving feedback from the environment, typically the output of a tool call. The agent uses the observation to inform its next reasoning step.
3 / 5
An “agent step” refers to:
Agent step = one full cycle: reason → act → observe. Counting steps helps with cost estimation and loop detection.
4 / 5
A team says: “The agent’s trajectory showed it tried three search queries before finding the answer.” What is a trajectory?
Trajectory = the full record of what an agent did from start to finish. Used for debugging, evaluation, and training on successful examples.
5 / 5
Your agent keeps calling the same search tool with slightly different queries and never produces a final answer. This is called:
Stuck in a loop is when an agent repeats actions without converging. Safeguards: max_steps limit, loop detection, and a fallback response. One of the most common production failure modes.