Former OpenAI CTO pushes AI beyond chatbots with human-like conversations

Mira Murati’s startup is developing next-generation AI systems designed for real-time, natural conversations, signalling a major shift from traditional prompt-based chatbots towards more responsive and human-like digital interactions.

The artificial intelligence industry may be preparing for its next major transformation as startups and technology firms explore systems that can interact with users more naturally and continuously, moving beyond the traditional chatbot format that has dominated the AI boom so far.

For the last two years, most mainstream AI platforms have relied on a turn-based interaction model. Users type or speak a prompt, the system processes the request, and then delivers a response after a short pause. Platforms such as OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, and xAI’s Grok continue to follow this structure.

However, Mira Murati, the former Chief Technology Officer of OpenAI, is now working on a different approach through her startup Thinking Machines Lab. The company recently shared details of what it calls “interaction models” — AI systems capable of processing conversations in real time while users are still speaking.

AI conversations could become more natural

The new interaction model aims to make AI assistants behave less like automated tools and more like active participants in conversations. Instead of waiting for users to finish speaking, the system can reportedly analyse speech, context, and visual inputs simultaneously in short processing intervals.

Researchers say this could allow AI systems to respond more fluidly, recognise conversational cues, and even react during ongoing discussions. The approach is expected to improve the natural rhythm of human-machine interaction by supporting interruptions, contextual replies, and live conversational feedback.

AI educator Ansh Mehra believes the technology represents a deeper structural shift rather than merely faster response times. According to him, humans naturally process conversations continuously instead of waiting for complete sentences before forming reactions or opinions.

The startup’s preview also demonstrated early signs of “backchanneling” — subtle conversational behaviours such as brief acknowledgements, nods, or verbal cues that signal active listening. Experts believe these features could significantly improve how users emotionally connect with AI systems.

Dual-layer architecture powers continuous interaction

To support the real-time conversational experience, Thinking Machines Lab is reportedly developing a dual-layer architecture. One layer handles immediate interaction and maintains conversational flow, while a secondary reasoning model performs more complex analytical processing in the background.

Industry analysts say this setup mirrors the way humans balance instinctive responses with deeper thinking during conversations.

The development could have significant implications for markets like India, where voice-based internet usage is rapidly expanding. Experts believe AI systems capable of naturally understanding and responding in regional languages may help improve accessibility for first-time internet users, students, and small businesses.

Technical challenges continue to persist

Despite growing interest around real-time conversational AI, researchers caution that the technology remains at an early stage. Long conversations still create context-management challenges, which can affect the consistency and accuracy of AI responses over time.

Several experts also note that many of the company’s claims are yet to be independently verified, as the technology remains in research preview mode rather than public release.

Still, industry observers believe the direction is clear. As AI companies race to build systems that feel more responsive and intuitive, the next generation of digital assistants may increasingly resemble collaborative conversational partners instead of traditional chatbots waiting for prompts.