The Verge reported Monday that Thinking Machines, the artificial intelligence company founded by former OpenAI chief technology officer Mira Murati, has announced it is developing what it calls "interaction models," a concept the company describes as enabling people to collaborate with AI in ways that mirror natural human collaboration.
To understand why this matters, it helps to know who Mira Murati is and what her departure from OpenAI represented. As CTO of what became the most prominent AI laboratory in the world, Murati was among the small group of people most directly responsible for shaping the systems — including successive versions of GPT and the original release of ChatGPT — that defined the current era of generative AI. Her resignation from OpenAI in late 2024 was one of the more consequential executive departures in recent technology industry memory, arriving during a period of significant internal turbulence at the company and prompting immediate speculation about what she would build next. The founding of Thinking Machines answered part of that question. The nature of what the company is actually building has, until now, remained largely opaque.
The phrase "interaction models" is doing a lot of work in this announcement, and it is worth pausing on it. The dominant paradigm for AI products today is essentially transactional: a user issues a prompt, the model returns a response, and the exchange ends. Even the most sophisticated AI assistants largely operate within this loop. What Thinking Machines appears to be gesturing toward is something structurally different — a mode of engagement that is less like querying a database and more like working alongside a colleague. The distinction, if the company can actually realize it, is not trivial. Most of the frustration that sophisticated users have with current AI tools involves precisely this gap: the models are capable but passive, responsive but not genuinely participatory. A system that could initiate, remember context across extended collaborative sessions, and adapt its behavior to the working style of a specific person or team would represent a meaningful departure from what exists today.
The likely reading is that Thinking Machines is positioning itself not in direct competition with the mass-market chatbot products that OpenAI, Google, and Anthropic have built, but in a different layer of the market — one oriented toward professional and creative collaboration rather than information retrieval or task automation. This would be a strategically sensible choice. The chatbot market is crowded, heavily capitalized, and increasingly commoditized. The collaboration layer, by contrast, remains genuinely unsolved. Enterprise software companies have spent decades trying to make digital tools feel more like genuine partners in work and have largely produced more sophisticated filing systems. If Murati's team has a credible technical approach to this problem, the addressable market is enormous.
The consequences of this announcement fall on several groups simultaneously. For investors, the signal is clarifying. Thinking Machines has attracted significant attention on the basis of Murati's reputation alone, but reputation is not a product roadmap. Naming a specific conceptual direction — even at this early and abstract stage — gives potential backers something to evaluate. For the broader AI industry, the announcement is a form of competitive pressure. When someone with Murati's institutional knowledge and standing declares that the current model of human-AI interaction is insufficient, it implicitly challenges incumbents to defend or accelerate their own work in this space. OpenAI in particular will be watching closely: Murati knows its technical architecture, its research priorities, and its organizational culture in ways that no outside competitor does.
For ordinary users, the consequences are more distant but potentially more significant. If interaction models develop into a genuine product category, the experience of working with AI tools could shift from something that feels like using a very fast search engine to something that more closely resembles a working relationship. That would change the nature of the productivity questions companies are asking, the skills workers are being trained to develop, and the way teams are structured around AI capabilities. It would also raise new questions — about memory, data retention, and the appropriate boundaries of a system that is designed to learn how a specific person thinks and works.
What to watch for next is relatively straightforward. Thinking Machines will eventually need to show, not just describe. The concept of interaction models is intriguing but still entirely abstract, and the history of AI product development is dense with promising framings that dissolved on contact with the actual difficulty of building the thing. The next meaningful signal will be a demonstration of the technology in some form — a research publication, a limited product release, or a technical specification detailed enough to be evaluated on its merits. How long it takes Murati's company to move from announcement to artifact will say a great deal about how far along the work actually is.
