From Assisted to Autonomous: How DataWorks Data Agent Revolutionizes Data Intelligence

From Assisted to Autonomous: DataWorks Data Agent’s Paradigm Shift

When an operations colleague asks for weekend metrics on short notice, the traditional workflow forces you to align definitions, query tables, patch data, and build reports—often consuming the entire weekend. DataWorks Data Agent is designed to eliminate this pain by automating the entire pipeline from requirement understanding to task execution.

Five‑Stage Evolution

Stage 1 – Code Completion : After typing a line of code, the system suggests the next line automatically.

Stage 2 – Q&A & Code Assistance : Natural‑language prompts generate explanations, suggestions, and copy‑paste code snippets.

Stage 3 – IDE Copilot : The agent understands comments, translates code, and improves developer productivity by 30‑40%.

Stage 4 – Chat BI : Users can ask operational questions in chat; the agent analyses dependencies, proposes solutions, and awaits confirmation before acting.

Stage 5 – Autonomous Mode (Current Release) : Given a goal, the agent performs end‑to‑end work—including demand analysis, data exploration, code generation, testing, deployment, and post‑deployment attribution—without manual intervention.

Dual‑Engine Architecture

The system offers two interchangeable modes that share a single unified context:

CLI Mode : Reads project files and change logs, performs deep data insight, generates code, runs unit tests, sets quality rules, and hands over a reviewed package for release.

Claw Mode ("lobster" mode): Integrates with DingTalk, WeChat Work, Feishu, etc., handling point‑style incidents via natural‑language chat and executing confirmed actions automatically.

Both modes rely on a common semantic kernel that understands data, code, and security permissions, enabling seamless hand‑off between CLI‑driven batch jobs and real‑time chat‑driven incident handling.

Unified Technical Core

DataWorks Data Agent builds on the existing DataWorks infrastructure—resource groups, cloud‑native runtimes, and permission systems—so the agent inherits compute resources, networking, and workspace bindings with zero cold‑start cost. The unified runtime hosts scheduling, execution, and load balancing for all agents.

Open Ecosystem & Skill Extensions

Through the MCP Skill protocol, partners and customers can register custom skills that become instantly available across CLI, IM, IDE, and API entry points. The platform also supports major LLMs (Alibaba Tongyi Qianwen, GLM, DeepSeek) with Text‑to‑SQL fine‑tuning for Alibaba’s big‑data engines, and allows private model deployment when needed.

CPU‑GPU Joint Optimization

Performance analysis shows that CPU consumption dominates latency for many agent workloads. By co‑optimizing CPU core frequency and thread throughput alongside GPU acceleration, the agent achieves significant end‑to‑end efficiency gains, a result of close collaboration with AMD and Intel.

Conclusion

The release marks a paradigm shift from “enhanced” to “autonomous” data agents. With its dual‑engine design, unified runtime, open Skill ecosystem, and hardware‑aware optimizations, DataWorks Data Agent delivers a single‑sentence goal that drives an entire data workflow automatically—positioning it as a true digital employee for enterprise big‑data environments.