The pace of AI's evolution may soon be determined by AI itself.

On March 25, during the 2026 Zhongguancun Forum Annual Meeting, Yue Zhi An founder Yang Zhilin delivered a major speech. He predicted that within the next year or so, the research and development methods of artificial intelligence will undergo a fundamental transformation, and the role of human researchers will shift from "hands-on work" to "resource scheduling."

New R&D Paradigm: From "Human Brain Driven" to "Token Driven"

Yang Zhilin pointed out that AI development is entering a new stage led by AI. In this paradigm, the working style of researchers will change significantly:

Tokens as Core Tools: In the future, each researcher will be equipped with massive AI Tokens (tokens). These tokens will no longer just be for chatting, but will become productivity resources.

AI Self-Exploration: With massive tokens, AI can help researchers synthesize new tasks, create new environments, and even define the most suitable "reward functions," autonomously exploring new network architectures.

Efficiency Singularity: "Self-Driving" in AI R&D

This transformation means that AI iteration will no longer be completely limited by the energy limits of human experts.

Accelerated Evolution: When AI begins to lead the research process, the entire technology development speed will grow exponentially.

Ecosystem Building: Yue Zhi An stated that it hopes to closely collaborate with the open-source community to jointly push the boundaries of intelligent technology and build a more vibrant ecosystem.

Industry Context: Large Models Enter the "Execution Agent" Deep Water Zone

At the same time Yang Zhilin spoke, the entire industry also showed a surge in "Agent (Intelligent Entity)" trends.

Tencent: Its Crab (WorkBuddy) has been updated, further enhancing its ability to assist in office scenarios.

Step Star: The cloud-based AI assistant StepClaw based on OpenClaw has also been officially launched.

Conclusion: When AI Begins to Research AI

From "teaching AI to speak" to "letting AI do R&D," this is not only an upgrade of tools, but also a reshaping of production relations. In Yang Zhilin's vision, in the top laboratories of the future, the standard for measuring a researcher's level may become the efficiency of driving AI Tokens to innovate.

When AI R&D enters the "driverless" stage, the path to general AI (AGI) may be shorter than we imagine.