Mira Murati left OpenAI in 2024 with a different vision for superintelligence. While his former colleagues pursued increasingly large-scale models of automating human work, Murati co-founded Thinking Machines Lab on the contrarian bet that the future belonged to AI that constantly monitored humans.
The highlight is an “interaction model” in which AI is trained to communicate naturally through a camera and microphone. Unlike existing voice interfaces that transcribe speech and process it like a chatbot, Thinking Machines’ model natively understands continuous, messy human communication.
When someone interrupts or changes the subject, they adapt. Natively understanding the rhythms of real human conversation represents a fundamental departure from how today’s AI interfaces work.
OpenAI, Anthropic, and Google are developing large-scale models to tackle increasingly complex tasks, creating entire software applications from text prompts with minimal human involvement. Murati’s approach reverses the equation. She’s building AI that amplifies human intent, rather than replacing it.
“At some point, we will have super-intelligent machines,” Murati told WIRED. “But we think the best way to actually achieve a good future with many possibilities is to keep humans informed.”
Launched in October 2025, Tinker enables researchers and engineers to use custom data to improve frontier AI models. The new interaction model represents the next step: AI that actually understands what a person is trying to accomplish, rather than just executing commands.
Alexander Kirillov, a founding team member specializing in multimodal AI, explains this change harshly, saying, “The model is always aware of what you are doing and is always there to respond and provide information, search for information, or use other tools.” this is something that no one can do [today’s other] Models can actually do that. ”
She envisions people building, customizing, and collaborating with their own frontier models, rather than relying on closed AI systems built by large corporations.
“When AI actually understands and predicts intent, it actually amplifies people’s own preferences and values,” Murati explains.
