Routa is an engineering‑focused multi‑agent framework that separates tasks, state, events, and execution into controllable modules, enabling open‑ecosystem AI coding agents to collaborate through structured specifications, event‑driven coordination, and verifiable tool interfaces rather than fragile prompt stitching.
In this final lesson of a nine‑day Agent engineering series, the author integrates the full Agent Loop, tools, MCP, skills, RAG, context handling, multi‑turn dialogue, and multi‑agent coordination to create a runnable RSS news‑briefing Agent that fetches feeds in parallel, filters content with LLMs, summarizes articles, and outputs a markdown report.
This lesson explains how to extend a single‑agent system with multi‑agent collaboration, covering context isolation, Handoff and Router patterns, flat coordinator architecture, code examples, task decomposition, and practical run‑time demos for building complex AI workflows.
An in‑depth analysis shows how GLM‑5 surpasses traditional code‑generation LLMs by autonomously designing, implementing, and debugging complex multi‑agent systems, from a fireworks HTML demo to a 35,000‑line TrustGraph refactor, highlighting its architecture, tool integration, and cost‑effective advantages.
This article synthesizes Anthropic’s recent Claude Agent blogs, presenting a layered architecture and practical steps to transform chat‑centric agents into reliable, production‑ready systems, covering when to adopt multi‑agent setups, the role of Skills and MCP, and a ready‑to‑use implementation checklist.
This article explains LangGraph's graph‑based design, runtime behavior, state management, checkpoint persistence, and flexible workflow modifications, providing concrete code examples and patterns that illustrate why the framework is well‑suited for complex multi‑agent AI systems.
GitHub is transforming from a code‑hosting platform into an AI‑driven development hub by integrating Claude, Codex and Copilot into its Agent HQ, enabling developers to issue commands to multiple AI agents, streamline workflows, and boost productivity across the entire software creation process.
CuaBot v1.0 introduces a new open‑source approach that lets AI agents interact with a desktop via independent cursors and sandboxed windows, avoiding full‑screen screenshots and mouse hijacking while supporting multi‑agent parallelism, H.265 video, audio, clipboard sharing, and a CLI built on Xpra and Docker.
This article explains what AI agents are, outlines single‑agent and multi‑agent use cases in AIOps such as knowledge retrieval, tool guidance, fault diagnosis, and process automation, and lists the key technical skills needed to build and manage these intelligent operational assistants.
Google’s recent study of over 8,000 reasoning tasks shows that advanced LLMs like DeepSeek‑R1 spontaneously develop multiple internal “expert” personas that debate, and that activating a discovered “social switch” dramatically raises accuracy, revealing that engineered conflict can enhance AI reasoning.
Oh My Claude Code transforms Claude Code into a multi‑agent orchestration platform, offering five execution modes, 32 specialized agents, automatic model routing, and simple installation, enabling developers to automate complex coding tasks from planning to testing with natural‑language commands.
As 82% of organizations plan to adopt AI agents within three years, this article outlines a full‑chain methodology—7‑dimensional classification, multi‑layer evaluation metrics, three‑stage validation, five‑step risk lifecycle, and progressive governance—to safely scale autonomous agents from prototype to enterprise deployment while addressing emerging multi‑agent challenges.
This article explains how the easy‑agent project evolved from a single monolithic AI into a multi‑agent architecture with specialized Foreman, Coder, and Researcher agents, covering design principles, communication mechanisms, task decomposition, fault tolerance, parallel execution, observability, and future extensions, complete with code examples and open‑source links.
This article provides an in‑depth comparison of nine mainstream AI agent development frameworks—Pydantic AI, SmolAgents, DeepAgents, LlamaIndex, CAMEL, AutoGen, CrewAI, LangGraph, and OpenAI Agents SDK—detailing their design principles, strengths, weaknesses, typical scenarios, and guidance for selecting or mixing them in production.
This article analyzes recent Claude Skills research, revealing security flaws in over a quarter of skills, a systemic performance collapse when single-agent skill sets exceed 50‑100 items, and how hierarchical routing and cognitive‑capacity limits can restore accuracy while mitigating security risks.
This article introduces AgentScope Java 1.0.7, showcases its new features such as Ollama integration, Agent Skill support, and Nacos‑based A2A architecture, and walks through a complete boba‑tea‑shop example that demonstrates a Supervisor‑Worker multi‑agent system, ReActAgent configuration, dynamic MCP registration, MySQL session persistence, Mem0 long‑term memory, and AutoContextMemory compression, plus quick deployment options for local, Kubernetes, and Docker environments.
AgentScope integrates Google’s A2A protocol with Nacos as a default registry to enable unified discovery, governance, and cross‑framework collaboration of agents across diverse language stacks, addressing challenges of multi‑team coordination, protocol heterogeneity, and production‑grade deployment.
The article explains how Tencent Cloud's open‑source TCAR router, a 4‑billion‑parameter model, tackles the limitations of traditional single‑label routers by first reasoning and then selecting agents, enabling cross‑domain, conflict‑aware, and adaptable task coordination in enterprise AI systems.
This article analyzes why single‑agent systems hit limits in context management and distributed development, compares four multi‑agent patterns (Subagents, Skills, Handoffs, Router) with concrete performance data across three scenarios, and offers a decision framework for selecting the most suitable architecture.
FinRpt introduces a novel multi‑agent pipeline that builds a high‑quality stock research report (ERR) dataset from six financial data sources, defines a comprehensive 11‑metric evaluation suite, and demonstrates that supervised‑fine‑tuned and reinforcement‑learned LLM agents significantly outperform single LLM baselines in both accuracy and efficiency.
This article explains LangGraph's network (decentralized) architecture for multi‑agent systems, compares it with supervisor and hierarchical designs, and provides a step‑by‑step Python example using the langgraph‑swarm library to build agents that can dynamically hand off control and preserve conversation continuity.
JoyCode Agent leverages a patch‑test co‑generation and iterative validation framework to achieve a 74.6% Pass@1 score on the SWE‑bench Verified benchmark, reducing resource consumption by 30‑50% and introducing a closed‑loop multi‑agent pipeline that integrates testing, patch generation, trajectory compression, similarity retrieval, and decision arbitration.
The article details how an AI‑driven test platform combines large language models with a corporate‑level knowledge graph to automatically generate end‑to‑end API test scripts for complex business flows, overcoming context loss, dependency gaps, and scalability limits of single‑interface generation.
The article examines how AI engineering builds on, rather than discards, traditional software engineering principles, using the evolution of AI‑driven search at Alibaba to illustrate architectural upgrades that manage uncertainty, integrate context engineering, and combine classic design patterns with new AI‑specific tools.
This guide explains how to use Alibaba Cloud's AgentScope framework and Container Service to build, orchestrate, and deploy enterprise‑grade AI agents, covering background, core features, step‑by‑step deployment, sandbox integration, and best‑practice recommendations for cloud‑native AI workloads.
The article reviews the 3S‑Trader framework, a training‑free multi‑LLM system that uses scoring, strategy, and selection modules to construct weekly stock portfolios, and shows that it outperforms rule‑based and deep‑learning baselines on DJIA and sector datasets with strong risk‑adjusted returns.
The paper introduces GLM, a multi‑agent Graph‑CoT framework with an optimized LLM serving architecture that dramatically improves accuracy, reduces token consumption, lowers latency, and increases throughput across diverse domains, as demonstrated by extensive GRBench evaluations.
This article analyzes recent research by East China Normal University and Fudan University on whether eight major LLM families exhibit a systematic “Chain-of-Affective,” revealing how internal emotional structures influence model outputs, multi‑agent interactions, and user experience, and offering practical guidelines for mitigating emotional loops in AI systems.
Druva’s multi‑agent AI assistant, built on Amazon Bedrock AgentCore, lets users resolve up to 90% of routine data‑protection tasks via natural‑language chat, shrinking backup‑failure troubleshooting from hours to minutes and delivering a 70% overall efficiency gain, backed by detailed performance evaluations.
Sina Weibo’s engineering team tackled the high technical barriers, low reuse, and long cycles of large‑model AI deployment by building a unified AI application platform that combines a layered architecture, low‑code workflow, multi‑agent orchestration, and knowledge‑base integration, enabling rapid, reliable AI solutions across the company.
A round‑table of AI experts and industry leaders examines the current capabilities, limitations, and future trajectories of code agents, covering topics from capability boundaries and autonomous evolution to large‑scale codebase challenges, multi‑agent collaboration, hallucination mitigation, and security safeguards.
BettaFish is a Go‑based, AI‑driven multi‑agent opinion analysis platform that tackles information silos, overload, and bias by aggregating data from diverse sources, iteratively refining results through reflection loops, and delivering visualized, actionable reports for scientific decision‑making.
ValueCell is an open‑source Python‑based multi‑agent AI platform that simulates a professional investment team, integrates various LLM providers, covers multiple markets (A‑shares, US stocks, crypto, Hong Kong), and offers one‑click deployment with detailed orchestration, enabling users to run AI‑driven portfolio management and replicate the Alpha Arena trading competition.
This article dissects LangGraph, the graph‑based workflow engine for LLM agents, explaining its stateful design, node‑edge architecture, compilation and execution process, and why it outperforms traditional linear LangChain chains for complex multi‑step AI applications.
This article walks through the open‑source DeerFlow framework—its multi‑agent architecture, core features, and a step‑by‑step implementation using the Go‑based LangGraphGo library, covering planner, researcher, reporter and podcast nodes, state‑graph design, CLI/web modes, and deployment instructions.
This article examines the chronic pain points of traditional marketing, explains how AI‑driven multi‑agent collaboration can create a data‑rich, automated, and continuously optimized marketing loop, and presents a real‑world case study with measurable performance gains and practical implementation guidelines.
This tutorial walks you through reproducing Claude Code’s core functionality using the LangGraph Python framework, covering ReAct agents, human‑in‑the‑loop control, sub‑agent orchestration, Todo task management, context compression, streaming output, and provides complete notebooks, installation steps, and example code for hands‑on learning.
This article explains the communication challenges of modern multi‑agent AI applications and demonstrates how RocketMQ for AI’s event‑driven, asynchronous messaging architecture can improve scalability, reliability, and cost efficiency through a step‑by‑step weather‑and‑travel planning example.
The 2025 Open Atom Developer Conference highlighted JD's JoyAgent project, an open‑source, 100% enterprise‑grade multi‑agent platform that excels in AI, data governance, and diagnostic analysis, with detailed features, performance metrics, and deployment experiences shared.
This article details the design and evolution of a multi‑agent AI system for automated data generation in integration testing, covering challenges, single‑ versus multi‑agent approaches, prompt engineering, tool governance, intent recognition, tool filtering, reasoning execution, performance gains, and practical recommendations.
This article explores how Wukong's AI‑driven multi‑agent architecture dramatically improves code security auditing by addressing context loss, scheduling imbalances, and integrating a data‑flywheel that turns bad cases into continuous model improvements, illustrated by a real NVIDIA Megatron‑LM vulnerability fix.
Since ChatGPT's debut, the LLM landscape has progressed through four stages—prompt engineering, chain orchestration, autonomous agents, and multi‑agent systems—each enhancing intelligence and automation, with this article detailing their evolution, advantages, drawbacks, and practical implementation examples in Go.
This newsletter curates five insightful articles covering voice agent applications in games and insurance, ten practical tips for building multi‑agent systems with context engineering, a deep dive into AI‑assisted coding from theory to practice, high‑accuracy AI coding workflow design, and the release of Rspack 1.6 for smaller, cleaner build artifacts.
This article provides a detailed analysis of LangChain's OpenDeepResearch project, comparing its Graph workflow and Multi‑Agent architectures, explaining core node implementations, execution flows, and offering step‑by‑step instructions for configuring and deploying the system locally.
This article walks through constructing a LangGraph graph that encapsulates three agents—task planning, web search, and report generation—into a DeepResearch application, then shows how to package and deploy the backend and frontend so users can interact with the system via a web UI.
This article explains how a multi‑agent system—comprising intent‑recognition, tool‑engine, and inference agents—solves the challenges of AI‑driven data generation (AI‑造数) by improving accuracy, speed, and scalability through modular design, prompt engineering, and sophisticated tool governance.
This article provides a comprehensive overview of the Coze low‑code AI agent platform, detailing its free, multi‑model capabilities and six core functions—plugins, knowledge base, database, image flow, workflow, and multi‑agent collaboration—while illustrating how each feature lowers development barriers and enables sophisticated agent applications.
This article categorizes and compares twelve popular large‑model AI Agent development frameworks—low‑code platforms, basic programming paradigms, advanced code libraries, and multi‑agent systems—detailing their core features, typical use cases, and trade‑offs to help developers choose the most suitable solution.
The article introduces JoyAgent‑JDGenie, an open‑source, product‑grade multi‑agent system from JD Cloud, explains its mission to eliminate the last‑mile barrier for rapid multi‑agent app creation, details its layered architecture, recent DataAgent addition, and discusses deployment options and challenges.
AlphaAgents introduces a role‑based multi‑agent framework—Fundamental, Sentiment, and Valuation agents—leveraging LLMs to analyze 10‑K reports, news, and price data, with a debate mechanism via Microsoft AutoGen; experiments on 15 tech stocks show superior cumulative returns and Sharpe ratios under risk‑neutral and risk‑averse settings compared to single‑agent baselines.
Alipay’s AI research team, together with Ant Group and East China Normal University, leveraged a self‑developed large‑model‑plus‑optimization framework to create a constraint‑aware multi‑agent system that won both the Original OS Track and DSL Track at the IJCAI‑2025 Autonomous Travel Itinerary Planning Competition.
JD Cloud has released JoyAgent‑JDGenie, the first fully product‑grade open‑source Java multi‑agent system that bundles front‑end, back‑end, framework, engine and core agents, supports major LLMs, offers layered architecture, Docker or manual deployment, and showcases demos such as PPT generation and sales analysis.
This article reviews the MARS framework, a risk‑aware multi‑agent reinforcement‑learning system for automated portfolio management that tackles market non‑stationarity and proactive risk control, detailing its hierarchical architecture, formal MDP formulation, training process, and superior experimental results on DJIA and HSI benchmarks.
This article explains how Ctrip’s AI‑driven Transpiler, built on a ReAct‑inspired multi‑agent framework, automates code transpilation from CRN to xTaro and streamlines large‑scale Java 21 upgrades, dramatically reducing manual effort and improving accuracy across the software development lifecycle.
Agentic AI transforms static large language models into autonomous agents by adding perception, goal orientation, planning, action, interaction, and iterative loops, tracing its evolution from early chatbots through Prompt Engineering, ReAct, AutoGPT, OpenAI Function Calling, to modern multi‑agent frameworks, while addressing challenges like memory, hallucinations, and scalability.
This report details a university project that built a flexible, high‑performance multi‑agent large‑model framework for ecological environment big‑data analysis, covering system architecture, individual agents, memory mechanisms, report generation, a FastAPI‑LangGraph backend, a React frontend, testing methodology, and future directions.
This article explains how a multi‑agent AI system shifts asset‑loss control from reactive to proactive by building a full‑link protection framework that extracts knowledge, identifies risks, automatically deploys safeguards, and continuously learns from incidents, delivering faster, more accurate financial security for e‑commerce platforms.
The article explores Ant Group’s agentUniverse team’s experience applying multi‑agent technology in finance, covering background on large language models, the agentUniverse framework, real‑world implementations, and the advantages of coordinated multi‑agent collaboration for complex analytical and decision‑making tasks.
This article explains how to re‑implement ByteDance's DeerFlow deep‑research framework in Go (Deer‑Go), covering the multi‑agent architecture, control‑hand‑off, interrupt & checkpoint mechanisms, integration with the Hertz SSE server, and step‑by‑step deployment instructions.
AniME introduces a director‑driven multi‑agent system that combines a custom model‑selection protocol (MCP) with the open‑source AniSora V3 model to automatically generate consistent, high‑quality long‑form animation from story scripts, handling everything from storyboard creation to video editing and quality evaluation.
The article walks through installing the open‑source Camel AI‑OWL agent, explains its multi‑agent architecture, runs two practical test cases—retrieving GitHub statistics and generating a physics animation—and highlights strengths, UI shortcomings, tool‑registration issues, and the project's recent inactivity.
This guide shows how to quickly build, configure, and deploy modular AI agents using the OxyGent framework—covering environment setup, minimal code initialization, tool integration, multi‑agent orchestration, and advanced deployment techniques—all illustrated with concise examples and screenshots.
This article walks through installing JDGenie's Docker image, fixing configuration and Dockerfile issues, and evaluating its code‑analysis and animation‑generation capabilities against other open‑source agents, highlighting both its strengths and current limitations.
JoyAgent‑JDGenie is an open‑source, end‑to‑end multi‑agent framework that lets users generate reports, PPTs, and data analyses directly from queries, supports easy integration of custom tools, and achieves a 75.15% accuracy on the GAIA benchmark, surpassing major competitors.
This guide introduces the open‑source OxyGent multi‑agent framework, walks through a quick 20‑line installation, demonstrates environment configuration, tool integration, visualization, and advanced features such as RAG, Reflexion, and distributed deployment for AI applications.
This guide introduces the open‑source OxyGent multi‑agent framework, provides step‑by‑step installation, a 20‑line hello‑world example, tool integration via SSE, MCP and FunctionHub, deployment features like data persistence and distributed setup, and outlines advanced use cases such as multimodal agents and plan‑and‑solve paradigms.
The article explains how the master‑slave (or manager‑worker) architecture, inspired by both software micro‑services and biological systems, solves context fragmentation and coordination challenges in large‑model multi‑agent applications, detailing design principles, technical implementations, advantages, limitations, and suitable use cases.
JD’s internal technical salon reveals the rapid evolution of large‑model AI tools, outlining design principles, current capabilities like JoyAgent, OxyGent and JoyCode, real‑world applications across office, code review, logistics, and future policy and multi‑agent visions, highlighting their impact on modern workplace productivity.
This article details the technical practice and core thinking behind Ant Financial’s front‑end team Multi‑Agent platform “Tiangong Wànxiàng”, covering its evolution from AutoGPT to Manus, the LangGraph foundation, ReAct agents, context engineering, architecture design, and real‑world front‑end code generation capabilities.
Learn how to quickly set up OxyGent, a flexible AI agent framework, by installing Python, Node.js, and the MCP tools, configuring environment variables, and using just 20 lines of code to build, debug, and deploy multi‑agent applications with features like RAG, tool integration, and distributed execution.
This article explains how Alibaba Cloud OpenSearch LLM version evolved from RAG 1.0 to RAG 2.0, introducing the DeepSearch multi‑agent architecture that combines offline data processing, online query handling, planning, clarification, search, and summarization agents to deliver more accurate and complex AI‑driven answers.
This article walks through constructing an agentic RAG pipeline that combines LangChain, LangGraph, Google Gemini embeddings, and Qdrant vector storage to enable automatic query rewriting, relevance grading, and concise answers to blog‑post questions via a Streamlit UI.
This article systematically compares major open‑source deep‑research agent frameworks—including DeerFlow, SmolAgents, LangChainAI, SkyworkAI, and Researcher—detailing their architectures, best practices, and commercial alternatives, to help developers and users choose the most suitable tool for automated research workflows.
JD Retail's Oxygen team open‑sources OxyGent, a pure‑Python multi‑agent framework that lets developers assemble AI teams like Lego bricks, offering elastic scaling, full‑traceable decision pipelines, GAIA benchmark leadership, and a suite of features for rapid prototyping and continuous evolution.
This article introduces multi‑agent architectures and the Agents Handoff pattern, then demonstrates a complete implementation using Spring Boot 3.4.2, Langgraph4j, and Spring AI, including code for a marketplace agent, a payment agent, configuration, testing, and suggestions for further improvement.
This article walks you through what Spring AI Alibaba 1.0 offers, highlights its major updates such as the Graph multi‑agent framework and ecosystem integrations, and provides a step‑by‑step upgrade path with Maven dependency changes, code fixes, and configuration adjustments for Java developers.
This comprehensive article explores the evolution of AI agents, explains key protocols like MCP and A2A, compares reasoning frameworks such as CoT, ReAct, and Plan‑and‑Execute, and demonstrates how Golang frameworks Eino and tRPC‑A2A‑Go enable elegant development, orchestration, and observability of complex multi‑agent systems with practical code examples and visual diagrams.
This article explores the evolving concept of AI agents, debates their definitions, outlines four major deployment challenges—including prompt instability, planning balance, domain knowledge integration, and response speed—and presents practical strategies such as prompt engineering, workflow design, multi‑agent architectures, and model optimization to build reliable, high‑availability agents.
This article introduces the OpenAI Agents SDK, explains its key capabilities such as Agent Loop, Handoffs, Guardrails, and Tracing, provides practical Python code examples, compares it with other multi‑agent frameworks, and discusses best practices for building reliable AI applications.
This article examines Copilot 3.0’s planning module, explains how DeepSeek R1’s GRPO reinforcement‑learning pipeline enables flexible multi‑agent orchestration, addresses the limitations of Copilot 2.0, and presents experimental results that show a 61% reduction in reasoning length and a 9% relative gain in accuracy.
An in‑depth overview traces the development of AI agents from early LLM milestones to modern “class‑Agent” models, examines core components such as memory, tool use, planning and reflection, analyzes current limitations, and outlines emerging solutions like workflows, multi‑agent systems, and model‑as‑product paradigms.
DeerFlow, an open‑source deep‑research platform built on LangStack, showcases a novel multi‑agent architecture, Meta‑Prompt engineering, MCP integration, and AI‑generated podcasts, providing developers with a powerful, extensible framework for advanced LLM‑driven workflows.
This guide explains how MetaGPT’s multi‑agent architecture eliminates the logical gaps of single‑agent systems, improves task stability, and can be rapidly deployed on Alibaba Cloud’s Serverless Function AI platform with step‑by‑step instructions, configuration details, and example applications.
This article explains how the LoopAgent component of the Agent Development Kit enables coordinated, iterative execution of multiple sub‑agents, detailing its key parameters, workflow, multi‑agent collaboration patterns, and how iterative refinement improves complex AI task outcomes.
Google’s open‑source A2A (Agent‑to‑Agent) protocol, unveiled at Google Cloud Next 2025, defines a standardized, secure, Web‑based framework that lets heterogeneous AI agents discover each other, negotiate interactions, and manage cross‑agent tasks, supporting multimodal communication, long‑running jobs, and push notifications.
The article provides an in‑depth comparison of Google’s Agent2Agent (A2A) protocol and Anthropic’s Model Context Protocol (MCP), examining their design philosophies, technical architectures, feature sets, ecosystem support, suitable scenarios, challenges, and potential future convergence for AI agent interoperability.
This talk presents a generative UI solution that leverages AI to automatically create design‑compliant interfaces, enabling data‑driven, personalized experiences while freeing frontend developers from repetitive design work and reducing manual intervention across the production pipeline.
This article walks through the major changes in Microsoft AutoGen 0.4, explains its layered modular architecture and event‑driven multi‑agent design, details the built‑in Tools types, and provides step‑by‑step Python code for creating a Tools Agent and integrating it with an MCP server.
The article introduces Spring AI Alibaba OpenManus, a Java‑based multi‑agent framework that showcases complex task execution through browser automation, travel itinerary generation, and document translation, while detailing its architecture, current limitations, and future development roadmap.
This article presents MaRCA, a multi‑agent reinforcement‑learning framework that allocates computation resources across the full ad‑serving chain, modeling user value, compute cost, and action rewards to maximize ad revenue while keeping system load stable under fluctuating traffic.
Google unveiled an AI co‑scientist built on Gemini 2.0 that uses multiple specialized agents to generate, evaluate, and refine research hypotheses, demonstrating promising results in drug repurposing, liver fibrosis target discovery, and antibiotic resistance, while also highlighting current limitations and community reactions.
The article provides a detailed analysis of Manus, the first global general‑purpose AI agent, covering its end‑to‑end task loop, cost efficiency, multi‑agent architecture, benchmark results, real‑world application cases, trial observations, limitations, and future outlook.
The article introduces the OWL (Optimized Workforce Learning) general‑purpose AI agent, explains its six‑step architecture, benchmark performance surpassing OpenManus, and argues that its innovations represent genuine application‑level advancement rather than mere “shell‑wrapping,” while highlighting its multi‑agent collaboration framework.
Manus, billed as the world’s first true general AI agent, combines multi‑agent architecture, tool integration, and superior GAIA benchmark performance to automate complex tasks, while its invitation‑only rollout and ethical concerns illustrate the tension between hype‑driven marketing and sustainable AI adoption.
The article explains what AI Agents are, how they differ from single large‑model systems, outlines four agent architectures—Reflection, Tool Use, Planning, and Multi‑Agent—and cites concrete examples from Grammarly, VS Code plugins, Image Describer X, ChatDev, as well as initiatives by Tencent and Google, highlighting the 2025 AI Agent boom.
The document describes the evolution, design principles, key technologies, online inference workflow, evaluation methods, and sample‑generation techniques of a large‑language‑model‑based multi‑agent system that powers a 24/7 e‑commerce merchant assistant, highlighting its benefits, challenges, and future work.
This article explains the evolution of multi‑agent systems for e‑commerce assistants, detailing the ReAct‑based planning framework, hierarchical master‑sub agent collaboration, evaluation methods, and sample‑generation techniques that together improve accuracy, efficiency, and scalability of AI‑driven merchant services.
JD’s Merchant Intelligent Assistant leverages a large‑language‑model‑based multi‑agent architecture to provide 24/7 e‑commerce support, detailing its evolution, planning techniques, online inference, evaluation methods, sample generation, and practical insights for scalable AI‑driven operations.
The JD Merchant Intelligent Assistant employs a large‑language‑model‑driven multi‑agent architecture with dynamic ReAct planning, enabling merchants to query and execute store operations in under a second with over 90 % decision accuracy, while reducing inference cost, hallucinations, and engineering effort across diverse e‑commerce tasks.
The article surveys the rise of multi‑agent LLM systems, detailing how specialized agents collaborate on tasks such as travel planning, outlining their workflow, comparing them with single‑agent models, listing prominent frameworks, and discussing current challenges and research citations.
