Sarah Guo on Open Models and the Model-Labs vs. Agent-Labs Divide

Summary

Latent.Space examines investor Sarah Guo's framework distinguishing "model labs" from "agent labs" and what is "trainable" versus "untrainable" in AI systems. The thesis: foundational model capability is increasingly commoditized and trainable, while the durable, harder-to-replicate value is migrating to the agent layer — orchestration, context engineering, tool integration, and the surrounding product/workflow scaffolding that turns a capable model into reliable autonomous work. Open-model adop