Tim O'Reilly: Big AI Labs Are Misreading What Users Actually Want
Original: Tech Visionary Says the Big AI Labs Don’t Get What People Want
Why This Matters
The open vs. closed AI debate has major implications for who controls the next layer of digital infrastructure.
Tech publisher and internet pioneer Tim O'Reilly argues that major AI labs like Anthropic and OpenAI are building architectures of control rather than participation, and that open-source AI — not just open-weight models, but the full stack — is essential for broad societal benefit.
Tim O'Reilly, publisher, VC, and longtime tech commentator, is advocating for fully open-source AI systems that separate the model, the harness, and the application layer. In a Wired interview, he argues that frontier AI labs are optimizing for use cases that don't match what ordinary users actually need. He warns that the biggest models — such as Anthropic's Claude and OpenAI's latest releases — are getting worse at certain tasks like writing, even as they advance on benchmarks. O'Reilly's core concern is that hyperscalers are repeating Microsoft's 1990s playbook, locking users into proprietary ecosystems with tracking built in. He also pushes back on the security argument against open-source AI, contending that cybersecurity incidents and pathogen-related risks have originated from frontier models, not open-source ones. He cautions that if the US focuses only on frontier AI dominance, China could gain the upper hand by diffusing capable smaller models widely across society. His guiding principle — 'create more value than you capture' — underpins his call for open architectures that enable broad participation and innovation.