AI that doesn’t just talk like you — it talks with you

Try interrupting your AI mid‑sentence. You won’t get far. The moment it starts responding, the conversation locks into a rigid turn‑taking mode. You speak, it listens. It speaks, you wait. For years, that’s been the default structure of human–AI interaction — polite, orderly, and completely unlike the way real people actually communicate.

Because real conversations are messy. We overlap, cut in, change direction, react to tone, and adjust mid‑thought. We don’t wait for a “beep” to speak. And until now, AI simply hasn’t been built to handle that level of fluidity.

Thinking Machines, the new AI startup led by former OpenAI CTO Mira Murati, wants to break that pattern. They’ve introduced a new class of “interaction models” designed to operate in real time. These models process audio, video, and text in rapid 200‑millisecond slices — fast enough to track what’s happening as it happens, not after the fact.

This shift isn’t just a technical upgrade. It’s a philosophical one. Instead of treating conversation as a sequence of turns, these models treat it as a continuous stream. That means an AI that can listen while it talks, adjust while you’re still forming your sentence, or jump in when something changes in the environment. It’s closer to how we collaborate with colleagues, friends, or teammates — dynamic, responsive, and context‑aware.

The architecture behind this is still experimental, but the possibilities are already clear. Imagine an AI that can translate live during a meeting without waiting for pauses. Or one that can follow your train of thought even when you pivot halfway through a sentence. Or one that can watch a video with you and react in real time. This isn’t turn‑taking with a shorter delay. It’s a fundamentally different interaction model.

Thinking Machines is betting that productivity won’t come from AI that simply answers questions better, but from AI that behaves more like a human partner — listening, talking, seeing, showing, and yes, interrupting when it needs to. The goal is not just intelligence, but presence.

The models are still in research preview, but the direction is unmistakable. The next leap in AI isn’t just about smarter outputs. It’s about keeping pace with the way we actually communicate. AI that doesn’t just respond — it keeps up.