Executive Summary
Sourcegraph CTO Beyang Liu reveals a critical strategic vulnerability: American AI application builders are increasingly dependent on Chinese open source models because US policy has created a competitive disadvantage in open weight model development. Liu's company, which built the top-performing coding agent on merge rate benchmarks, uses Chinese models (Kimi K2, Qwen, GLM) for production workloads because they outperform US alternatives in agentic tool use. This isn't ideological—it's practical. The models work better. Liu argues that Terminator-style AI safety narratives have migrated into policy circles, creating regulatory uncertainty that makes US companies gun-shy about releasing competitive open source models. Meanwhile, Chinese models dominate the open weight landscape that application builders prefer for post-training and cost optimization. The irony is stark: the US invented the AI revolution but is ceding the open source model layer to China through policy-induced self-handicapping. Liu's experience suggests this dependency will deepen as model capabilities flatten and applications optimize for specialized, smaller models rather than frontier intelligence. The window to reverse this trend may be closing as regulatory complexity increases and state-by-state AI laws create compliance nightmares for smaller model developers.
Key Insights
what Beyang Liu said“When you plop those [US models] into like an agentic application, you know, the tool use isn't quite robust enough. It's not quite there yet... the ones that we find most effective at agent workloads, they're almost all, I would say they are all of Chinese origin right now”
This is a preview. Log in to see the full analysis including investment opportunities, risks, catalysts, and detailed insights.