AI
O'Reilly: Big AI Labs Are Building the Wrong Future
Tech entrepreneur Tim O'Reilly argues that major AI companies are pursuing a strategy of control that misses what ordinary users actually need.
WIRED — AI
Tim O'Reilly, a longtime publisher and tech strategist, has long operated on a simple principle: the most valuable enterprises create more value than they capture. Now he's applying that lens to artificial intelligence—and concluding that the industry's largest players are getting it wrong.
In a conversation with Steven Levy, O'Reilly made the case that open-source AI represents the future, while the dominant corporations are building systems designed to lock users in rather than liberate them. He sees echoes of Microsoft's 1990s playbook in how today's AI labs operate, and he worries that approach will ultimately backfire.
The Architecture Question#
O'Reilly distinguishes between simply releasing model weights and true openness. The real issue, he argues, is architectural—whether AI systems enable genuine participation or enforce control. Current frontier models from companies like Anthropic and OpenAI, he contends, bundle together the model, the user interface layer, and the application in ways that prevent customization.
"They built an architecture of control rather than an architecture of freedom and participation," O'Reilly said. This matters because it allows companies to track users and dictate which use cases matter, rather than letting developers and users adapt AI to their own needs.
When Bigger Isn't Better#
O'Reilly challenges the assumption that the largest, most capable models necessarily win. While frontier models excel at certain tasks, smaller, open-weight models often outperform them in specific areas—and they can be deployed anywhere without vendor lock-in. He points to a strategic vulnerability: if the US focuses only on frontier AI while China distributes capable smaller models widely, China may gain a broader advantage despite lacking the single most advanced system.
"The goal is to give people the ability to innovate freely, to paint outside the lines," he said.
Security and Strategy#
On the common concern that open-source AI poses security risks by removing safeguards, O'Reilly inverts the argument: all major cybersecurity incidents involving AI have involved frontier models, not open-weight alternatives. He suggests that restricting open-source development while accelerating frontier models actually increases risk rather than mitigating it.
