Tim O'Reilly argues big AI labs miss everyday user needs, pushes open-source freedom

Tim O'Reilly, a longtime technology commentator, discussed with Steven Levy how today’s biggest AI labs are betting on ever-larger models while overlooking the practical needs of ordinary users. He argues that open-source AI must go beyond releasing model weights to sharing the full stack, enabling developers to separate the model, harness, and application layers. According to O'Reilly, this architecture of freedom would counter the control-centric designs that let companies track users.

He points to security incidents tied to frontier models as a reason to slow them rather than restrict open-weight versions. O'Reilly highlighted initiatives such as the open-source agentic harness Pi and an open-memory consortium being explored by his nonprofit, the AI Disclosures Project. He also criticized the venture-capital-driven funding model that dominates AI, likening it to the early 1990s PC battles before the web emerged. The interview concludes with O'Reilly’s belief that AI will become a creative medium comparable to writing or painting, with open-source tools empowering a diverse range of innovators.