Mira Murati reemerges with Thinking Machines Lab's interface bet
The former OpenAI CTO previewed models for continuous audio, text and video, but left release timing and customer traction unstated.
By Ryan Merket ยท Published
Why it matters
Murati has capital, credibility and a rare founder brand in AI. Her next test is whether Thinking Machines Lab can move from mystique to measurable product traction.
TechCrunch reports that Mira Murati (@miramurati) used a Bloomberg interview in San Francisco to preview Thinking Machines Lab's next product direction: AI "interaction models" built for continuous streams of audio, text and video rather than the familiar prompt-and-response loop.
The appearance matters because Murati has been unusually selective about public exposure since leaving OpenAI, where she was CTO from May 2022 to September 2024, according to her public biography. TechCrunch described Thursday's interview as her first major public appearance in roughly 18 months. For a founder with her profile, that silence has been both a recruiting asset and a constraint: mystery helps when hiring elite researchers, but it does less when customers, partners and investors are trying to understand what Thinking Machines Lab is actually building.
Murati's own path explains why the market is watching. Before founding Thinking Machines Lab in February 2025, according to the company's public profile, she helped lead OpenAI through the period when ChatGPT turned the lab into the center of the generative AI boom. She also became interim CEO during OpenAI's November 2023 leadership crisis after Sam Altman was briefly removed. In the Bloomberg interview, according to TechCrunch, Murati called that period "the blip" and said OpenAI would have "imploded" without her involvement during the five-day stretch and immediate aftermath.
What Murati is now willing to say
The product tease was specific enough to signal a direction, but not enough to define a launch. TechCrunch says Murati described Thinking Machines Lab's interaction models as systems that process audio, text and video in 200-millisecond intervals, with the aim of capturing real conversation patterns such as interruptions, pauses and mid-thought corrections.
That is a notable shift in emphasis from Thinking Machines Lab's only shipped product cited by TechCrunch: Tinker, an API for fine-tuning open-source AI models. Tinker speaks to developers and researchers who want more control over model adaptation. The interaction-model framing points at a broader interface layer, where the contest is not just which model answers best, but which system can follow a human conversation as it unfolds.
Murati did not give a release date, according to TechCrunch. That omission is not a small detail. In AI, demo language can outrun deployable product quickly, and the difference between a research preview and a customer-ready interface is where reliability, latency, cost and safety all become operational problems.
The attention problem
Thinking Machines Lab has raised attention without saying much. Its public profile lists a $2 billion early-stage round by July 2025, a $12 billion valuation, Andreessen Horowitz as lead investor, and Nvidia, AMD, Cisco and Jane Street among investors. Those figures should be read as publicly listed background, not current operating metrics. The available material does not show revenue, customer count, usage for Tinker or current headcount.
That gap is the tension around Murati's reemergence. Thinking Machines Lab has a founder investors know, a technical narrative the market wants to believe in, and enough capital to compete for scarce AI talent. But Thinking Machines Lab is still publicly defined by one developer product and a promised next interface, while OpenAI, Anthropic and xAI occupy more of the daily customer and talent conversation.
TechCrunch also reported that Murati downplayed recent high-profile researcher departures, saying that building a frontier AI lab from scratch compresses years of organizational volatility into months. That is a reasonable founder argument, especially in a market where senior AI researchers are being recruited like professional athletes. It also raises the practical question every well-funded lab faces: whether the founding story can survive the messy middle of turning research ambition into a stable organization.
The bet beneath the caution
Murati's careful return looks less like a product launch than a positioning exercise. She is reminding the market that Thinking Machines Lab is not merely another model lab chasing benchmarks. Her stated direction is interface-level AI: systems that listen while people talk, adapt in real time and behave less like chat windows.
That is a founder-friendly story, and a commercially useful one. If Murati can turn it into something customers can use, Thinking Machines Lab gets to compete on experience rather than raw model scale alone. If she cannot, the spotlight she avoided for 18 months will make the unanswered questions sharper: when the interaction models ship, who uses them, and whether Thinking Machines Lab can convert Murati's OpenAI-era credibility into a company with visible traction of its own.