Interaction Models: Mira Murati's new AI

The problem that Thinking Machines wants to solve

The AI models available today operate on a sequential logic. The user writes or speaks. The model waits. Then it processes and responds. This pattern, while functional, introduces a profound discontinuity compared to natural human communication.

Thinking Machines clearly describes this limitation: «Today’s models experience reality in a single thread» . In other words, the model perceives nothing until the input is complete. It doesn't see the user's hesitation. It doesn't catch the tone of voice. It doesn't interpret the visual context.

Therefore, the announcement of the interaction models is born from a precise ambition. The goal is to bridge this gap between artificial intelligence and authentic human collaboration. As reported by The Verge , Thinking Machines aims for models that "think, respond, and act in real-time".

What are interaction models: architecture and functioning

Thinking Machines' interaction models are designed to simultaneously process three input channels: audio, video, and text. Furthermore, they do so continuously, without waiting for the user's interaction to conclude.

This real-time multimodal approach is an architectural leap from traditional large language models. In fact, current models—even the most advanced—treat each conversational turn as a discrete event. Conversely, an interaction model maintains an active and persistent perception of the context.

In practical terms, this means the system can notice if the user is hesitating before completing a sentence. It can interpret a facial expression during a video call. It can adapt the response based on the emotional tone detected in real-time. Therefore, it's a model that doesn't just react, but Join .

To delve deeper into the topic of multimodal AI and its technical foundations, the MIT Technology Review offers precise analyses on the evolution of these systems.

Who is Mira Murati and why the market follows her

Mira Murati served as CTO of OpenAI until 2024. In that position, she oversaw the development of GPT-4 and ChatGPT. Her departure from OpenAI generated significant attention in the industry.

Thinking Machines' foundation has confirmed that Murati intends to build something structurally different. Not an alternative to ChatGPT. Rather, a new paradigm of human-machine interaction. Therefore, the market follows every announcement from the company with interest.

Similarly to what happened with other startups founded by former executives of large AI labs, Thinking Machines benefits from immediate technical credibility. However, the distance between an announcement and a commercially mature product remains significant. This is particularly true for such ambitious technologies.

According to the analyses of Gartner , multimodal AI technologies are in a phase of rapid maturation. Consequently, business adoption times are shortening compared to previous cycles.

Immediate impact for Italian SMEs: what really changes

For an Italian SME, the concrete question is: does this development change anything today? The answer is complex.

In the short term, interaction models aren't yet available as a commercial product. Thinking Machines announced the direction, not the launch. However, the indirect impact is already measurable. In fact, announcements like this accelerate the evolution of the entire AI ecosystem, including platforms already in use.

Specifically, companies operating in highly relational fields — customer service, consultative sales, internal training — should monitor this trajectory. Furthermore, those considering investments in AI solutions for their business must take into account this paradigm shift in planning.

We at SHM Studio work daily with SMEs that are wondering how to integrate artificial intelligence into their operational flows. Therefore, understanding where the technological frontier is moving is an integral part of our consulting approach.

The still open construction site: limits and unknowns

It's important to keep a realistic perspective. Interaction models present non-trivial technical challenges. Processing audio, video, and text simultaneously in real-time requires considerable computational power. Furthermore, latency must be low enough to make the interaction smooth.

Beyond this, privacy-related issues emerge. A system that continuously perceives the user's visual and auditory context raises relevant regulatory questions. In Europe, the GDPR and AI Act framework impose precise constraints. Therefore, the enterprise adoption of these technologies will necessarily have to contend with the regulatory perimeter.

Finally, the question of the interface remains open. How do you design a user experience for a system that doesn't wait? How do you handle interruptions or corrections in real-time? These are interaction design problems that are still largely unresolved. For those working in web design and digital interfaces, this is already a relevant area for reflection.

What to do now: orient yourself without chasing

Faced with a technological announcement of this magnitude, the most effective response is neither passive waiting nor uncritical immediate adoption. Instead, it's useful to structure a conscious strategic posture.

First, it's worth mapping out business processes where real-time interaction could add value. For example, tech support sessions, customer onboarding, staff training. Then, you can assess which existing tools align with this paradigm and test them in controlled environments.

For SMEs wanting to structure a coherent digital strategy, the services of Digital marketing and of SEO remain fundamental pillars. However, AI is increasingly entering these areas transversally. Therefore, ignoring it means giving up a growing competitive advantage.

Those who manage campaigns on professional channels like LinkedIn can already leverage AI today to optimize targeting and messages. Our services of LinkedIn campaigns and of Google Ads already integrate automated optimization logic. Likewise, the SEO copywriting benefits from AI tools for semantic search and content structuring.

Outlook: where this trajectory leads

In the 2027-2028 biennium, it's reasonable to expect that real-time multimodal models will become a standard component of enterprise AI platforms. Thinking Machines will not be the only player in this space. In fact, OpenAI, Google DeepMind, and Anthropic are all working on advanced multimodal capabilities.

According to projections by McKinsey , the adoption of generative AI in companies is destined to accelerate significantly in the next two years. Consequently, SMEs that begin today to build internal skills and AI-ready processes will find themselves in a structurally advantageous position.

For this reason, SHM Studio accompanies its client companies not only in managing current digital activities but also in understanding the ongoing transformations. Those who wish to delve deeper into these topics can visit our Blog or contact us directly for a consultation.

In summary, Thinking Machines' interaction models represent a clear signal about the direction of AI. It is not yet time for operational adoption. However, it is the right time to understand, plan, and position yourself.