From “Lobster” to Ontology: DACon Reveals the Next Trend in Self‑Evolving AI Agents

On February 24‑25, 2026, DataFun hosted the DACon conference in Shanghai, attracting more than 8,000 technology developers, managers, and industry experts. The two‑day event featured 50 talks across 15 sub‑sessions, covering topics from enterprise ontology engineering to Agentic RAG, AI‑driven data governance, and AI+AR ecosystem building.

Keynote 1 – Self‑Evolving Agents : Professor Xiao Yanghua (Fudan University, Director of Shanghai Data Science Key Lab) presented “Technical Breakthroughs and Evolution Trends of Self‑Evolving Agents”. He introduced the open‑source “Lobster”‑class agent framework GenericAgent, scheduled for public release on 2026‑01‑11. Compared with OpenClaw, GenericAgent launches earlier, is fully self‑developed, and offers superior environment interaction and task execution capabilities. Its minimalist architecture (≈3,000 lines) and “seed philosophy” dramatically lower the entry barrier for agents and overcome traditional capability limits.

The talk dissected three core self‑evolution capabilities: (1) only nine built‑in atomic abilities enable autonomous skill composition for complex real‑world tasks; (2) automatic experience consolidation creates cross‑device, reusable long‑term memory; (3) a Fork mode allows agents to adapt to new scenarios and expand capabilities without human intervention, unlocking a new direction for domestically controllable agents.

Keynote 2 – Agent‑Ready Big‑Data Infrastructure : Wei Bowen (Alibaba Cloud Intelligent Group, Computing Platform) explained how agents are reshaping software development, content creation, and customer service. Alibaba Cloud will expose the full stack of multimodal data, vector retrieval, data freshness, and security governance, building an “Agent‑Ready” big‑data AI foundation that accelerates enterprise intelligence transformation.

Keynote 3 – Ontology in Banking : Yu Haohan (Shanghai Bank, Head of Data Management & Application) discussed “Ontology: The Bedrock of Data Governance and Knowledge Engineering in the Large‑Model Era”. He identified two pain points—high modeling thresholds and synchronization difficulty—and argued that embedding ontology into daily workflows, supported by organizational and incentive redesign (e.g., Palantir‑style FDE roles), is essential to reduce hallucinations and improve intent‑recognition accuracy.

Keynote 4 – Data Consumers Becoming Agents : Du Junping (Datastrato CEO, Apache Gravitino founder) traced a decade of solving “how humans can use data more efficiently”. He highlighted the shift from human‑centric data consumption to AI agents, requiring a new “Agentic Data Protocol (ADP)” for discovery, authorization, traceability, and compliance across multi‑cloud, multi‑engine, and multimodal data assets.

Keynote 5 – AI+AR Ecosystem : Yang Zhou (Rokid Global R&D Lead) presented “From AR to Agents: Evolution and Platform Layout of the Intelligent‑Agent Ecosystem”. He described Rokid’s dual‑core AI+AR products—Rokid AR Lite, AR Studio Pro, and Rokid AI Smart Glasses—and the “Lingzhu” no‑code agent development platform that has launched over 300 agents covering lifestyle, productivity, and interactive games, forming the leading AI ecosystem for smart glasses.

The agenda also included specialized sessions on enterprise ontology engineering, Agentic RAG, AI‑driven data development, LLM‑powered R&D efficiency, OpenClaw commercialization, contextual engineering, and cost optimization for large‑model applications. Notable case studies such as “From ChatBI to DataAgent: Scaling AI Data Retrieval at Xiaohongshu” and “DataAgent in Enterprise Credit Optimization” were highlighted.

In the exhibition hall, partners and sponsors—including Alibaba Cloud, Nvidia, Tencent Cloud, Ant Group, and many others—demonstrated cutting‑edge products and offered hands‑on experiences.

The conference concluded with an invitation to the upcoming Agentic AI Summit in Shenzhen (July 24‑25, 2026), emphasizing concrete, reproducible, and quantifiable solutions for intelligent‑agent deployment.