Baidu Search has released a 63-page paper titled “Towards an AI Search Paradigm”, introducing a modular, multi-agent (Multi-Agent) framework that reimagines how we search for information in the era of large language models (LLMs).
🚀 Unlike traditional retrieval-based search or linear RAG pipelines, this new paradigm coordinates four specialized LLM agents—Supervisor, Planner, Executor, and Writer—to handle everything from query analysis to tool invocation and final synthesis.
Here’s how the system works: 🔹 Supervisor Agent: Assesses query complexity and dynamically allocates the appropriate team. If the results are incomplete, it reflects and replans. 🔹 Planner Agent: Breaks down complex problems into a directed acyclic graph (DAG), selects appropriate tools through the Model Context Protocol (MCP), and supports replanning.
🔹 Executor Agent: Executes tool-related subtasks while ensuring fallback resilience and output quality.
🔹 Writer Agent: Synthesizes results into coherent, multi-faceted answers while filtering inconsistencies.
✨ What’s new?
- The Planner adapts to dynamic capability boundaries—selecting relevant tools in real-time.
- The DRAFT system iteratively improves tool documentation through LLM-tool interaction feedback.
- The customized COLT retriever ensures that tool selection is not only semantically relevant but also functionally complete.
- Utilizes reinforcement learning for joint optimization of agent coordination, employing multi-part rewards.
📌 Why is this important: This agent architecture reflects human-like reasoning in search, providing a scalable and adaptable foundation for building the next generation of AI search systems. 
https://arxiv.org/pdf/2506.17188Towards AI Search ParadigmBaidu Search
Recommended Reading
-
• Hands-on Design of AI Agents: (Orchestration, Memory, Plugins, Workflow, Collaboration)
-
• The Second Half of DeepSeek R1 + Agent
-
• Single Agent: AI Assistant for Enterprises
-
• The Evolution and Case Studies from Single Agent to Multi-Modal Agents to Multi-Modal Multi-Agent Systems (12,000 words, 20+ references, 27 figures)
Feel free to follow my public account “PaperAgent“, where I share a large model (LLM) article every day to stimulate our thinking, with simple examples and sophisticated methods to enhance ourselves.