Imagine you are trying to work with an AI system that waits silently until every sentence is finished, then answers after the moment has passed. Mira Murati, the former chief technology officer of OpenAI and founder of Thinking Machines Lab, is promoting a different approach: AI systems designed to collaborate with people in real time. The issue matters because major AI companies are racing toward more capable systems, while public concern remains high about automation, jobs, and control. Murati told WIRED that she still expects superintelligent machines, but argues that better outcomes require keeping humans involved.
Thinking Machines Lab is an artificial intelligence research and product company founded by Murati after she left OpenAI in 2024. Its stated audience includes researchers, builders, engineers, and people who want AI systems that are more understandable, customizable, and useful for their own needs. The company says it is emphasizing human-AI collaboration rather than focusing only on fully autonomous AI systems. Its first released product, Tinker, is aimed at researchers and developers who want to fine-tune open-source models through an API while Thinking Machines manages the underlying infrastructure.
The company’s new work fits where people need AI to respond during live activity, not only after a completed written prompt. Thinking Machines describes “interaction models” as systems that continuously take in audio, video, and text while thinking, responding, and acting in real time. The Verge reported that the company has shown examples involving animal mentions in a story, real-time translation, and posture feedback. Public reporting says the models are not generally available yet; Thinking Machines plans a limited research preview in the coming months and a wider release later in 2026.
In practice, Thinking Machines says the goal is to make interactivity part of the model itself rather than adding separate components around a conventional model. The company says such models can manage dialogue, respond to pauses or interruptions, handle simultaneous speech, and use tools while still listening. That is different from a turn-based exchange in which the user speaks, the model waits, and then the model responds. The analogy is simple: the company wants AI interaction to feel less like sending letters back and forth and more like sharing a workbench with a responsive colleague.
What comes next is still partly untested in public. WIRED reported that the interaction models have been demonstrated in videos but have not been publicly released. TechCrunch also noted that the idea remains a research preview, not a finished product available for broad use. For readers, the practical next step today is to review Thinking Machines’ public materials and compare the company’s claims with independent reporting before treating the technology as proven in everyday workflows.