Position: Agentic AI System Is a Foreseeable Pathway to AGI

Junwei Liao, Shuai Li, Muning Wen, Jun Wang, Weinan Zhang

Abstract: Is monolithic scaling the only path to AGI? This paper challenges the dogma that purely scaling a single model is sufficient to achieve Artificial General Intelligence. Instead, we identify Agentic AI as a necessary paradigm for mastering the complex, heterogeneous distribution of real-world tasks. Through rigorous theoretical derivations, we contrast the optimization constraints of monolithic learners against the efficiency of Agentic systems, progressing from simple routing mechanisms to general Directed Acyclic Graph (DAG) topologies. We demonstrate that Agentic AI achieves exponentially superior generalization and sample efficiency. Finally, we discuss the connection to Mixture-of-Experts, reinterpret the instability of current multi-agent frameworks, and call for greater research focus on Agentic AI.

Lay Summary:

Today, many believe the only way to achieve true Artificial General Intelligence (AGI) is by building increasingly massive, single AI models. However, our physical world is incredibly diverse, and we show that forcing one giant "monolithic" model to learn every possible task leads to an "Average Trap"—the model becomes a compromised generalist that loses expert-level sharpness.

Instead, we propose that the inevitable path to AGI is Agentic AI. Rather than relying on one generalist brain, Agentic AI functions like a highly coordinated team of specialists. When faced with a complex problem, it breaks the task down and routes the pieces to dedicated, expert agents. Through rigorous mathematical proofs, we demonstrate that this teamwork approach bypasses the limitations of a single model because each agent only needs to master its specific, smaller niche.

This collaborative system achieves significantly better performance and requires a fraction of the data and computing power. Ultimately, this means the future of AI isn't just about spending massive resources to train bigger models, but about designing smarter, interconnected networks of specialized agents, making advanced AGI research far more efficient and accessible.

Primary Area: Research Priorities, Methodology, and Evaluation

Keywords: Agentic AI; AI Agents; LLM-based Multi-Agent Systems

Submission Number: 157