Kanwal Mehreen Highlights 5 Papers Explaining Agentic AI From ReAct to AutoGen

1 articles · Updated · KDnuggets · Aug 14

Five papers anchor Mehreen’s guide to agentic AI, aiming to simplify a fast-moving field by mapping each paper to one core idea rather than starting with broad surveys.
ReAct and Toolformer explain the basic mechanics: ReAct pairs reasoning with actions in a think-act-observe loop, while Toolformer shows how models can learn when to call external tools.
Generative Agents and Voyager extend that foundation into longer-running behavior, emphasizing memory, reflection, planning, environment feedback, and reusable skills built through exploration.
AutoGen rounds out the list by shifting from a single assistant to multi-agent collaboration, where specialized agents can converse, use tools, execute code, and include humans in the loop.
Together, the 5 papers frame most modern agent systems as combinations of reasoning, action, tools, memory, feedback, planning, and collaboration.