Framework

The reason–act–observe loop

The question it answers

When a team puts an AI agent to work on a real task, what does it do at each reason, act and observe step to get from receiving a goal to finishing?

Origin
Yao et al., 2023; Anthropic, 2024
Diagram type
Cycle
Level of analysis
Team
Topics
AI for leaders
Glass sphere with a glowing neural-network core, icon for AI for leaders

Diagram

The reason–act–observe loop

  1. ReasonPlan the next step.
  2. ActCall a tool.
  3. ObserveRead the result from the environment, then repeat.

Pitfalls

Pitfall

Needs an explicit stop condition and pauses for human feedback at checkpoints.

Sources

References

Listed A–Z by author surname, Harvard style. A source with a DOI links to https://doi.org/…; one without links to its page or says plainly that it has no DOI.

  1. Anthropic (2024) Building effective AI agents. Anthropic Research.No DOIanthropic.comSource page
  2. Yao, S. et al. (2023) ReAct: synergizing reasoning and acting in language models. (ICLR 2023 (camera-ready version); arXiv:2210.03629).doi:10.48550/arXiv.2210.03629Source page