Your Weekly AI news roundup @18.05.2026

That is why Thinking Machines introduced interaction models – AI models built to handle audio, video and text continuously, instead of waiting for a user to finish speaking or typing. Their model works in 200 millisecond micro-turns, meaning it can listen, watch, speak and react almost at the same time. It can notice visual cues, jump in when needed, stay silent when the user is still thinking, or even speak while the user is talking, for example in live translation or real-time coaching.

The interesting part is not only the speed. Thinking Machines separates the system into two layers: one model stays present in the conversation, while another background model handles slower work like reasoning, searching, tool use or longer tasks. This is close to how a good colleague works: one part of the collaboration stays responsive, while deeper analysis continues in the background. In the uploaded weekly summary, this same idea was described as one model keeping the conversation alive while another handles slower reasoning and tools.

For everyday AI users, this could be a much bigger shift than just “a faster chatbot”. It points toward AI that can work beside us in real time – during meetings, coding, learning, design, customer support, language practice or hands-on tasks. The risk is also clear: real-time AI means new safety, privacy and attention challenges, because the model is no longer just reading a message after the fact. It is present while things happen. But if Thinking Machines is right, the next big AI interface may not be a chat window at all. It may feel much more like working with another person.