Qwen's Former Lead on What Hybrid Thinking Got Wrong — and Why He Now Backs Agents

Junyang Lin, former technical lead of Alibaba's Qwen project, delivered a talk titled 'Qwen: Towards a Generalist Model / Agent' and later expanded it into a detailed post. In the talk, Lin emphasized a shift in focus from 'training models' to 'training agents.' He argued that the previous approach of 'hybrid thinking' was incorrect and that the field should now prioritize agent-based systems.

Lin's remarks highlight a change in direction for AI development within the Qwen project. He did not provide specific technical details on what 'hybrid thinking' entailed or why it was wrong, but he positioned the move toward agents as a necessary correction.

This announcement reflects ongoing debates in the AI research community about the most effective path toward generalist AI systems. Lin's shift is one of several high-profile pivots in the field as researchers reassess the value of different architectural and training approaches. No further details about specific capabilities or timelines for agent-based Qwen models were provided.