Autonomous Task-Planning Using Agentic AI: A Comparative Experimental Study

Abstract

Recent advancements in Artificial Intelligence have enabled the development of intelligent systems capable of performing tasks autonomously. Agentic AI represents a modern approach in which systems act as independent agents that can analyze problems, plan actions, and execute tasks without continuous human intervention. This research focuses on evaluating and comparing different Agentic AI frameworks used for autonomous task planning. The study considers frameworks such as AutoGPT, ReAct agents, and LangChain agents. These frameworks are analyzed based on their ability to perform task decomposition, reasoning, execution, and interaction with external tools. An experimental methodology is used where predefined tasks are assigned to each framework, and their performance is measured using metrics such as accuracy, execution time, and resource utilization. The findings indicate that Agentic AI significantly improves automation and decision-making capabilities. Each framework demonstrates unique strengths, with ReAct excelling in reasoning, LangChain in tool integration, and AutoGPT in autonomous execution. This research provides valuable insights into the effectiveness of agent-based AI systems and their future potential in real-world applications.

Introduction

The text presents a study on Agentic AI, an emerging approach in artificial intelligence where systems act as autonomous agents capable of planning, reasoning, and executing complex tasks without continuous human supervision. Unlike traditional AI systems that rely on fixed rules and workflows, Agentic AI integrates large language models, memory, and planning mechanisms to enable adaptive and goal-driven behavior.

The study highlights that Agentic AI is increasingly used in domains such as robotics, healthcare, finance, and workflow automation. However, most existing research focuses on individual frameworks rather than comparing them systematically, creating a gap in standardized evaluation methods.

To address this, the paper conducts a comparative experimental analysis of three major frameworks: ReAct, AutoGPT, and LangChain agents. The research uses predefined tasks such as data retrieval, code generation, and multi-step reasoning, and evaluates performance based on accuracy, execution time, resource usage, and reasoning efficiency.

The methodology involves controlled experiments implemented in Python using tools like OpenAI APIs and Jupyter Notebook. Each framework is tested under identical conditions to ensure fair comparison. The proposed system decomposes tasks into subtasks, executes them using different agent frameworks, and records performance metrics for analysis.

Results show that each framework has distinct strengths. ReAct performs well in reasoning-heavy tasks due to its step-by-step thinking process. LangChain achieves the highest accuracy and excels in tool integration and structured workflows. AutoGPT demonstrates strong autonomy by independently completing tasks but requires more time and computational resources.

Conclusion

This research highlights the growing importance of Agentic AI in modern intelligent systems. The comparative study demonstrates that autonomous agents significantly improve task planning and execution efficiency. Each framework offers unique strengths, and their performance varies depending on the nature of the task. ReAct is suitable for reasoning tasks, LangChain excels in structured environments, and AutoGPT is ideal for autonomous execution. Future research can focus on developing hybrid systems that combine the strengths of multiple frameworks. Improvements in memory management, reasoning capabilities, and resource optimization can further enhance the performance of Agentic AI systems.

References

[1] S. Russell and P. Norvig, Artificial Intelligence: A Modern Approach, 4th ed., Pearson, 2024. [2] J. Yao et al., “ReAct: Synergizing Reasoning and Acting in Language Models,” arXiv preprint arXiv:2210.03629, 2023. [3] OpenAI, “GPT-4 Technical Report,” OpenAI Research, 2024. [4] H. Touvron et al., “Large Language Models for Artificial Intelligence Systems,” Journal of Machine Learning Research, vol. 25, pp. 1-30, 2024. [5] T. Brown et al., “Language Models are Few-Shot Learners,” Advances in Neural Information Processing Systems, vol. 33, pp. 1877-1901, 2020. [6] LangChain Documentation, “Building LLM-Powered Applications,” 2024. [7] Y. Shavit and M. Barak, “AutoGPT: Autonomous AI Agents for Task Execution,” AI Research Journal, vol. 12, pp. 55-68, 2024. [8] S. Park et al., “Generative Agents: Interactive Simulacra of Human Behavior,” ACM Conference on Human Factors in Computing Systems, 2023. [9] D. Silver et al., “Mastering the Game of Go with Deep Neural Networks and Tree Search,” Nature, vol. 529, pp. 484-489, 2016. [10] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning, MIT Press, 2016. [11] A. Vaswani et al., “Attention Is All You Need,” Advances in Neural Information Processing Systems, 2017. [12] R. Sutton and A. Barto, Reinforcement Learning: An Introduction, MIT Press, 2018. [13] Y. Bengio, “Deep Learning for Artificial Intelligence,” Nature Machine Intelligence, vol. 3, pp. 1-3, 2021. [14] J. Dean and S. Ghemawat, “MapReduce: Simplified Data Processing on Large Clusters,” Communications of the ACM, vol. 51, no. 1, pp. 107-113, 2008. [15] T. Mitchell, Machine Learning, McGraw-Hill, 1997. [16] P. Domingos, The Master Algorithm, Basic Books, 2015. [17] M. Wooldridge, An Introduction to Multi-Agent Systems, Wiley, 2020. [18] N. Jennings, “Agent-Based Computing: Promise and Challenges,” Science, vol. 284, pp. 747-750, 1999. [19] D. Hendrycks et al., “Measuring Massive Multitask Language Understanding,” International Conference on Learning Representations, 2021. [20] IEEE Artificial Intelligence Society, “Recent Advances in Autonomous AI Systems,” IEEE Transactions on Artificial Intelligence, 2024.

Copyright

Copyright © 2026 Ashwin Unnikrishnan. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.