Mira Murati sat down with Bloomberg in San Francisco on Thursday for her first major media appearance in roughly 18 months, using the interview to preview a new product direction at Thinking Machines Lab: 'interaction models' that process continuous streams of audio, text, and video in 200-millisecond intervals. The pitch is that the system can register interruptions, mid-thought corrections, and pauses in something close to real time, rather than the turn-based prompt-and-response loop that defines most AI products today. Murati framed it as a first step, declined to attach a release date, and otherwise gave away little.
Thinking Machines has been quiet for most of its 1.5-year existence. The company has raised capital, hired researchers, and shipped exactly one product — Tinker, an API for fine-tuning open-source AI models. In an environment where OpenAI is in the news cycle daily, Anthropic is dominating investor mindshare, and Elon Musk's xAI has been folded into SpaceX ahead of an expected public offering, prolonged silence carries a cost. The Bloomberg appearance reads as a deliberate, minimal correction.
Murati spent six years as CTO at OpenAI before leaving to start Thinking Machines. That tenure ended in the shadow of the November 2023 board crisis — internally called 'the blip' — during which the board fired Sam Altman and Murati served as interim CEO for the five-day stretch that followed. Asked about it on Thursday, she said the company would have 'imploded' without her involvement, while acknowledging she would have pushed for more information, a better transition plan, and more transparency in hindsight.
Key facts
- 01Murati's Bloomberg interview was her first major media appearance in roughly 18 months since leaving OpenAI.
- 02Thinking Machines' new 'interaction models' process continuous audio, text, and video in 200-millisecond intervals.
- 03The company has shipped one product to date: Tinker, an API for fine-tuning open-source AI models.
- 04Murati spent six years as CTO of OpenAI before founding Thinking Machines roughly 1.5 years ago.
- 05Nine-figure compensation packages have become standard in the AI talent war, Murati acknowledged.
Asked directly whether she still trusts Altman, Murati sidestepped and redirected toward a structural argument: that too many consequential AI decisions are concentrated in too few hands, across the industry rather than at any single company. Her concern, she said, is less about the character of individual leaders than about the absence of governance checks. Good people make bad calls; well-intentioned organizations drift. The industry, in her telling, has spent too much energy on virtue and too little on structure.
The talent question came up next. Thinking Machines has lost several high-profile researchers in recent months, a topic Murati has largely avoided in public. She offered two framings: that building a frontier AI lab from scratch compresses years of organizational volatility into months, and that nine-figure compensation packages — now standard currency in the AI talent war — capture imaginations but rarely tell the whole story.
“When I wake up in the morning, I am not thinking about how to kill the competitor.”— Mira Murati, CEO of Thinking Machines Lab
On her own competitive posture, Murati drew laughs from the audience.
The broader-future questions drew a more measured response. Murati, born in Albania, pushed back on framings of inevitable dystopia or inevitable utopia, arguing the current period is the one that decides the trajectory. She returned more than once to a warning about human oversight.
“if humans take their hands off the wheel too soon, the future will look very different, and not better.”— Mira Murati, CEO of Thinking Machines Lab
What was conspicuously absent from the appearance was anything resembling a product roadmap, a revenue figure, a headcount, or a benchmark. Tinker remains the only shipped product. The interaction-models preview is a research direction, not a release. For a company competing for talent against OpenAI, Anthropic, and xAI — and against the nine-figure offers those companies are writing — silence on commercial traction is a choice that will need to be defended at some point.
Thinking Machines' positioning bet is that real-time, continuous-stream interaction is the next interface shift, distinct enough from chat-based assistants to justify a new lab and a new architecture. That's a defensible thesis on paper. The execution risk is that voice-and-video latency is already a battleground OpenAI, Google, and a half-dozen startups are contesting with shipping products, and a 200-millisecond processing claim from a lab with one API in the wild is a thesis, not a moat.
Murati's reemergence matters more for what it signals about the state of Thinking Machines than for any specific news she broke. A lab that has spent 18 months heads-down doesn't put its CEO on Bloomberg to discuss governance philosophy unless the cost of further silence has started to exceed the cost of speaking. The interaction-models preview gives the market something concrete to anchor on, the OpenAI questions get answered enough to retire them, and the talent attrition gets a public framing that isn't 'we're losing the war.' Whether any of that translates into a product the market notices depends on what ships next, and how soon.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
