PandaAI's Li Yuqi: AI Trading Competition is Shifting from Factor Competition to Agent Competition

The Paradigm Shift in AI Trading

For the past two decades, quantitative investment has largely revolved around the discovery and application of Alpha factors. Researchers meticulously mined data, built models, and validated strategies to unearth excess returns, making factors the cornerstone of quantitative research. However, with the rapid advancements in large language models and multi-agent collaboration (A2A) technologies, the landscape is undergoing a profound transformation.

PandaAI's founder and CEO, Li Yuqi, emphasizes that the competitive edge derived from traditional factor-based approaches is diminishing. Standardized research processes, including data processing, strategy generation, backtesting, and validation, are increasingly being automated and efficiently executed by AI. This automation means that the efficacy of individual factors alone is no longer the sole determinant of success; instead, the ability to construct continuously learning, autonomously planning, and collaboratively working AI agent systems will define future leadership.

PandaAI's L0-L4 Evolution Framework

To provide a unified framework for understanding this evolution, Li Yuqi introduced the AI trading L0-L4 five-level evolution system. L0 represents a complete reliance on manual research and trading. L1 sees AI primarily assisting with tasks like code generation and prompt optimization. In the L2 stage, AI begins to participate in complete investment research processes through standardized agents, forming a single-agent closed loop.

Moving to L3, multiple intelligent agents collaborate (A2A), enabling autonomous planning, mutual checks and balances, and continuous iteration. Finally, L4 signifies a state where AI can perform standardized research and trading in public markets, with human involvement shifting to setting investment goals, acquiring non-public information, and controlling risk boundaries. This framework not only illustrates the increasing capabilities of AI but also reflects a fundamental change in financial production methods, positioning agents as true participants in financial research.

New Infrastructure for Agent-Driven Finance

Beyond theoretical frameworks, PandaAI is actively building the infrastructure to support this agent-driven future. The company unveiled a new product ecosystem designed to empower individuals with an 'AI quantitative trading team.' QUBE, a conversational strategy assistant, allows users to describe trading ideas in natural language, generating strategies, backtesting, and outputting signals, significantly shortening the path from concept to validation.

Trading OS offers a visual AI workflow canvas for arranging data, skills, and agents, enabling users to build and automate complete quantitative research tasks through natural language. Complementing these, EVO serves as an integrated AI trading agent research platform, facilitating agent creation, task management, collaborative research, and strategy execution. During the summit, PandaAI demonstrated A2A multi-agent collaboration within Trading OS, showcasing how research, analysis, risk control, and trading agents can autonomously divide tasks, provide feedback, and continuously optimize strategies, marking a shift from individual AI assistants to collaborative AI teams.