Question

What is the primary operational difference between a standard LLM call and an Agentic workflow that utilizes a &#x27;ReAct&#x27; loop?

Accepted Answer

A standard Large Language Model call is a linear, one-way process where the model receives an input prompt and generates a single completion based entirely on its internal training data. In contrast, an Agentic workflow using a ReAct loop transforms the model into an autonomous decision-maker that functions through a repeating cycle of reasoning and action. ReAct, which stands for Reason and Act, forces the model to interleave its thinking process with external interactions. In this loop, the model first generates a thought to analyze the task, then performs an action such as querying a search engine, database, or calculator to retrieve missing information, and finally observes the result of that action. The model then feeds this new observation back into its own context window to refine its reasoning before deciding if more actions are needed or if the final answer is ready. While a standard LLM call is constrained by the knowledge it already possesses at the time of inference, the ReAct loop allows the model to dynamically augment its knowledge by interacting with the real world, enabling it to solve complex, multi-step problems that require factual updates or specific tool usage.

Home → All Courses → Engineering and Technology Courses → Generative AI Application Development → Flashcard

What is the primary operational difference between a standard LLM call and an Agentic workflow that utilizes a 'ReAct' loop?