How JD’s Large‑Model Tools Are Transforming AI Workflows and Future Enterprises

With the rapid development of large‑model technology, AI tools have become essential productivity assistants, dramatically improving efficiency, solving business pain points, and extending knowledge through intelligent collaboration. Mastering these tools is now a key competitive advantage for modern professionals.

Industry Large Model Tool Development Status

Since the emergence of ChatGPT, large models have made breakthrough progress and are widely used in text generation, intelligent customer service, and other fields. Leading companies follow five design principles:

Low‑code UI : lower the usage threshold so non‑technical users can quickly adopt.

Security (local deployment) : private deployment prevents data leakage.

Usability (multi‑device compatibility) : one code works on many devices, reducing adaptation costs.

Scalability : align with open‑source trends for continuous iteration.

Cost‑effectiveness : ensure commercial sustainability.

JD’s Large Model Tool Capabilities and Applications

JD has upgraded its Yanxi large‑model brand to JoyAI , offering models from 3B to 750B covering language, speech, and image modalities. By improving inference efficiency and reducing training costs, JoyAI serves over one million merchants across retail, logistics, health, and industry, and powers platforms such as JoyAgent, JoyCode, and the multi‑agent framework OxyGent.

JD Large Model Tool Introductions

JoyAgent : the industry’s first 100% open‑source enterprise‑level agent platform, decoupled from the core platform, supporting independent local deployment and multi‑agent collaboration to accelerate AI adoption in production scenarios. Open‑source address

OxyGent : a multi‑agent collaboration framework that abstracts tools, models, and agents as plug‑in operators, enabling developers to assemble systems like building blocks with high extensibility and full‑traceability. Open‑source address

JoyCode : an intelligent coding assistant that evolved from AI‑assisted code completion and review to a multi‑agent autonomous programming stage. It showcases two success cases: H5 page development using Figma + JoyCode, and custom agent generation of SDK integration code. JoyCode 3.0 adopts a “plan‑first, execute‑later” strategy and custom agents to keep the AI on target, representing a new programming paradigm.

Experience and Business Practices with Large Model Tools

The three key factors for successful large‑model deployment are selecting the right platform, defining quantifiable goals, and iterating in phases. JD’s practice starts with MVP validation (e.g., low‑code workflow), then closes the data loop for continuous optimization (accuracy, fluency metrics). Effective tools reduce effort on model stability, security, and system integration.

Office Scenario

AI‑enabled office tools at JD focus on intelligent assistants such as Max, which aggregates tasks and information streams, enabling “create a task with one sentence.” Future work aims at “digital employees” that think, act, and collaborate, forming a one‑stop collaborative ecosystem.

Code Review

JD processes nearly ten thousand code reviews daily. Leveraging Yanxi models, DeepSeek, and AutoBots, AI‑driven review automates line‑level checks for standards, security, and other rules, integrating 17 rule libraries into the Coding platform. This has reduced Java defect rates by 32% for the logistics team and is now available across the group.

Local Life Applications

Three typical directions include cross‑language code translation (bridging Python and Java/Go), universal knowledge labeling (resolving regional knowledge conflicts), and multi‑agent simulation systems that model rider‑merchant‑user dynamics for dispatch optimization.

Logistics Applications

Key AI‑driven solutions include digital‑warehouse AR glasses (15% efficiency gain), intelligent outbound call systems (enhanced customer experience), address parsing engines (building‑level precision, reducing mis‑delivery to <0.0015%), and elevator‑recognition models (saving rider time).

Future Outlook of Large Model Tool Construction

Policy alignment is essential. Recent regulations such as China’s “Interim Measures for Generative AI Services” and the EU AI Act emphasize data quality, content safety, and copyright protection, urging enterprises to establish strict data cleaning and risk‑management mechanisms. The industry will shift from wild growth to standardized development, with government‑enterprise collaboration becoming a cornerstone.

Future Form of Large Model Agents

Agents will evolve toward multi‑agent collaboration, embodiment, and collective intelligence. They will transition from single‑task execution to coordinated complex tasks, become digital humans or robots for physical interaction, and continuously self‑evolve through environment feedback, supporting dynamic addition or removal of agents.

End‑to‑End Embodied Intelligence Path

By integrating perception, data closed‑loop, and lightweight deployment, embodied AI can move from rule‑based pipelines to end‑to‑end large‑model driving. Multi‑modal perception (camera, LiDAR, radar) enables robust scene understanding; data loops combine simulation and real‑world collection; quantization and specialized hardware allow 7B models to run on vehicle chips, achieving human‑like intuition in autonomous driving.