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

Editor's take

Junyang Lin, formerly of Alibaba's Qwen, has critiqued his past work on "hybrid thinking" within large language models, now championing agentic architectures. This shift is significant as it represents a potential pivot away from monolithic, generalist models towards more modular, task-specific AI systems that can leverage external tools and execute complex workflows. The implications are far-reaching for developers and businesses seeking to integrate AI more effectively into existing processes, moving beyond simple text generation to actionable intelligence.

The core of Lin's argument appears to be that the dynamic interplay between different model "modes" in hybrid thinking, as explored with Qwen3, failed to deliver on the promise of true generality. His current focus on agents suggests a belief that explicitly designed reasoning and action capabilities, rather than emergent properties of model architecture, are key to building more robust and capable AI. Future developments will reveal whether this agent-centric approach can overcome the combinatorial complexity and potential brittleness of earlier hybrid designs, and if it can scale to match the performance of the largest foundational models.