I know there’s a lot happening in the agentic AI space. You’ll hear about agents that use tools, agents with memory, agents that plan, agents that collaborate with other agents, and agents that explore environments on their own. It can get confusing, and if you start with long survey papers, you’ll probably end up even more confused. In my opinion, a much better way to learn is to read a few important papers that each explain one key idea behind modern AI agents.
This article is part of our 5 Fun series. In a previous article, we looked at 5 Fun Papers That Explain LLMs Clearly. This time, we are moving one step further, from models that generate text to AI agents that can reason, use tools, remember, and collaborate. So, let’s get started.
# 1. ReAct: Synergizing Reasoning and Acting in Language Models
Authors: Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, Yuan Cao
This is one of the best papers to start with if you want to understand agentic AI. The main idea is that an agent should not only think, and it should not only act — it should do both together. ReAct introduces a prompting framework where the model alternates between reasoning steps and actions. The reasoning helps the model plan, track progress, and recover from mistakes, while the actions allow it to interact with external environments such as search APIs, knowledge bases, or decision-making tasks. This paper is important because many modern AI agents follow this same basic loop: think, act, observe, update, and continue. If you want to understand the foundation of large language model (LLM) agents, this is the paper to read first.
# 2. Toolformer: Language Models Can Teach Themselves to Use Tools
Authors: Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Luke Zettlemoyer, Nicola Cancedda, Thomas Scialom
Tool use is one of the most important parts of agentic AI. A language model may be good at writing and reasoning, but it can still struggle with arithmetic, factual lookup, translation, or current information. Toolformer explores how a language model can learn to use external APIs in a self-supervised way. The model learns when to call a tool, which tool to call, what arguments to pass, and how to use the returned result in its final answer. The paper includes tools such as a calculator, search engine, translation system, calendar, and question-answering system. This paper is important because it moves us from “LLMs as text generators” toward “LLMs as systems that can decide when outside help is useful.”
# 3. Generative Agents: Interactive Simulacra of Human Behavior
Authors: Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein
This is one of the most fun agent papers to read because it feels like watching a small AI society come alive. The paper introduces generative agents that simulate believable human behavior in an interactive environment inspired by The Sims. These agents wake up, make plans, remember past experiences, reflect on them, talk to other agents, and coordinate future actions. The key architecture combines memory, reflection, and planning. This paper is important because it shows that agentic behavior is not only about solving one task. It is also about continuity: what the agent remembers, how it updates its beliefs, and how past events influence future decisions. If you want to understand why memory and reflection matter in agent design, this paper is a great place to start.
# 4. Voyager: An Open-Ended Embodied Agent with Large Language Models
Authors: Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Jim Fan, Anima Anandkumar
Voyager is an interesting paper because it takes agentic AI into an embodied environment — in this case, Minecraft. Instead of solving a fixed task and stopping, it continues to explore the world around it, discover new things, and grow its library of reusable skills. The architecture has three important components: (1) an automatic curriculum for exploration, (2) a skill library for storing executable behaviors, and (3) an iterative prompting mechanism that uses environment feedback and execution errors to improve. This paper shows what a long-running agent needs and how it can continuously improve through feedback by interacting with its environment.
# 5. AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
Authors: Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Awadallah, Ryen W. White, Doug Burger, Chi Wang
Many real-world tasks are too large for a single agent to handle cleanly. AutoGen introduces a framework where multiple agents can converse with each other to solve tasks. These agents can represent different roles, use tools, include humans in the loop, execute code, and coordinate through conversation. The paper shows applications across coding, mathematics, question answering, operations research, decision-making, and more. This paper is important because it explains one of the biggest shifts in agentic AI: moving from a single assistant to a system of specialized agents that collaborate. If ReAct explains the basic agent loop, AutoGen explains how that loop can become a team.
# Wrapping Up
Together, these five papers give you a strong foundation for understanding agentic AI:
- ReAct explains the reason-and-act loop.
- Toolformer explains how models can learn to use tools.
- Generative Agents explains memory, reflection, and believable behavior.
- Voyager explains lifelong learning and reusable skills in an environment.
- AutoGen explains collaboration between multiple agents.
Don’t try to remember the implementation details on your first read. Just focus on the main idea — because once you understand these five papers, most agentic AI systems will become much easier to understand. They are usually built by combining the same pieces: reasoning, action, tools, memory, feedback, planning, and collaboration.
Kanwal Mehreen is a machine learning engineer and a technical writer with a profound passion for data science and the intersection of AI with medicine. She co-authored the ebook “Maximizing Productivity with ChatGPT”. As a Google Generation Scholar 2022 for APAC, she champions diversity and academic excellence. She’s also recognized as a Teradata Diversity in Tech Scholar, Mitacs Globalink Research Scholar, and Harvard WeCode Scholar. Kanwal is an ardent advocate for change, having founded FEMCodes to empower women in STEM fields.
