Beyond Pipelines: A Survey of the Paradigm Shift toward Model-Native Agentic AI

2025-10-19

AI Summary

This work addresses the limitations of traditional pipeline-based AI systems—namely, passive responsiveness and fragmented capabilities—by proposing a paradigm shift toward *model-native agents*, wherein large language models (LLMs) intrinsically internalize planning, tool use, memory, and reflection to transition from reactive to proactive behavior. Methodologically, we introduce a unified LLM+RL+Task framework, leveraging reinforcement learning as the core engine to enable end-to-end joint training and continual internalization across language, vision, and embodied tasks. Experiments demonstrate substantial improvements in agent autonomy and generalization on complex, long-horizon reasoning and GUI-interaction benchmarks. Key contributions include: (1) the first systematic architectural principles for model-native agents; (2) a scalable, RL-driven internalization mechanism; and (3) parameterized pathways for multi-agent collaboration and self-reflection—establishing foundational theory and practice for next-generation general-purpose agents.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

The rapid evolution of agentic AI marks a new phase in artificial intelligence, where Large Language Models (LLMs) no longer merely respond but act, reason, and adapt. This survey traces the paradigm shift in building agentic AI: from Pipeline-based systems, where planning, tool use, and memory are orchestrated by external logic, to the emerging Model-native paradigm, where these capabilities are internalized within the model's parameters. We first position Reinforcement Learning (RL) as the algorithmic engine enabling this paradigm shift. By reframing learning from imitating static data to outcome-driven exploration, RL underpins a unified solution of LLM + RL + Task across language, vision and embodied domains. Building on this, the survey systematically reviews how each capability -- Planning, Tool use, and Memory -- has evolved from externally scripted modules to end-to-end learned behaviors. Furthermore, it examines how this paradigm shift has reshaped major agent applications, specifically the Deep Research agent emphasizing long-horizon reasoning and the GUI agent emphasizing embodied interaction. We conclude by discussing the continued internalization of agentic capabilities like Multi-agent collaboration and Reflection, alongside the evolving roles of the system and model layers in future agentic AI. Together, these developments outline a coherent trajectory toward model-native agentic AI as an integrated learning and interaction framework, marking the transition from constructing systems that apply intelligence to developing models that grow intelligence through experience.

Problem

• Surveying the paradigm shift from pipeline-based to model-native agentic AI
• Reviewing how planning, tool use, and memory evolve into learned behaviors
• Examining how this shift reshapes applications like Deep Research and GUI agents

Innovation

• Model-native paradigm internalizes planning, tool use, memory
• Reinforcement Learning enables outcome-driven exploration across domains
• Capabilities evolve from scripted modules to learned behaviors