Mira Murati wants her AI to ‘keep humans in the loop’

At a glance:

• Thinking Machines Lab, led by Mira Murati, introduces AI interaction models that natively grasp human nuances like pauses and tone.
• The approach emphasizes human collaboration over replacement, differing from trends at OpenAI, Anthropic, and Google.
• Tinker, launched in October 2025, allows researchers to refine frontier AI models with custom data.

Interaction models redefine human-AI communication

Thinking Machines Lab has developed what it calls "interaction models," trained to communicate through camera and microphone inputs. Unlike conventional voice interfaces that capture speech, transcribe it, and feed it into a language model, these systems natively understand continuous, messy human communication. This enables them to grasp the meaning of pauses, interruptions, and changes in tone, allowing real-time adaptation when a person clarifies a point or shifts topics. The company demonstrated these capabilities in several videos, though the models have not been released publicly. This approach marks a departure from typical AI interfaces, which often treat conversation as a series of discrete turns processed by less sophisticated subsystems. By handling fluid interaction natively, Thinking Machines aims to create AI that feels more intuitive and responsive to human intent.

A vision of human-centric AI development

At a time when many AI firms focus on automating complex tasks—such as writing entire software applications from text prompts—Murati argues for a more optimistic path. "At some point we will have super-intelligent machines," she tells WIRED. "But we think that the best way to actually have many possible futures—good futures—is to keep humans in the loop." This vision prioritizes human empowerment over replacement, suggesting AI should augment rather than automate away human roles. Thinking Machines is not alone in this perspective. Other labs, including Humans&, also aim to build AI systems that emphasize collaboration. Prominent economists have echoed this call, urging researchers and companies to design AI that enhances human capabilities. Murati’s approach directly contrasts with strategies at OpenAI, Anthropic, and Google, where large models increasingly perform intricate work with minimal human intervention.

Thinking Machines Lab: From OpenAI to independence

Mira Murati left her role as chief technology officer at OpenAI in 2024, co-founding Thinking Machines Lab with several prominent engineers. The startup has raised billions of dollars to build frontier AI, positioning itself as a major player in the race toward artificial general intelligence. Despite its rapid growth, the company has released only one product to date: Tinker. Murati’s departure from OpenAI was part of a broader wave of executive exits, but she has since doubled down on her vision for human-inclusive AI. The founding team includes experts like Alexander Kirillov, who specializes in multimodal AI—models that handle audio, video, and text. This expertise underpins the development of interaction models that can perceive and respond to human behavior in real time.

Tinker: Customizing AI with user data

Launched in October 2025, Tinker is a product that makes it possible to refine a frontier AI model using custom data. It is available as an API that researchers and engineers can use to fine-tune open-source models. This tool aligns with Murati’s belief that people should be able to build and customize their own AI systems, then work with them to achieve personal or organizational goals. Tinker democratizes access to advanced AI customization, allowing users to adapt models to specific domains or datasets. While still in its early stages, it represents a practical step toward the company’s larger vision of putting AI tools directly in the hands of users, rather than concentrating power in a few large tech firms.

Expert insights on collaborative AI

Alexander Kirillov, a founding team member at Thinking Machines, highlights the transformative potential of interaction models. "The model constantly perceives what you're doing and is constantly there to be able to reply and give you information or search for information or use other tools," he explains. "This is something that none of [today’s other] models can actually do. The turns [in a conversation] are determined by a much less intelligent system." Murati frames this as the first bet on human collaboration in AI. "This is showing the first bet on human collaboration," she says. "Where this is going is really amplifying people's own preferences and values, with AI actually understanding intent and predicting intent." This philosophy suggests a future where AI acts as a true partner, enhancing human decision-making rather than supplanting it.

Implications for the future of AI

The shift toward human-inclusive AI could have profound implications for employment, creativity, and technological equity. By enabling customization and collaboration, Thinking Machines aims to distribute AI’s benefits more broadly, countering concerns that automation will concentrate power and eliminate jobs. However, challenges remain, including technical hurdles in achieving seamless interaction and ethical questions about AI’s role in society. As the AI industry evolves, Murati’s approach offers a compelling alternative to fully autonomous systems. Investors and developers will watch closely to see if human-in-the-loop models gain traction, especially as competition intensifies among AI labs. The success of Tinker and future interaction models could redefine how humans and machines coexist in an increasingly automated world.