From OpenAI Powerhouse to Thinking Machines AI Visionary
Mira Murati, the former CTO who helped steer ChatGPT’s rise at OpenAI and briefly served as interim CEO during its leadership turmoil, has re-emerged with a new ambition: make AI feel like a real collaborator. In February 2025, she founded Thinking Machines Lab, a startup already attracting intense investor interest and even acquisition attempts from major tech players. Rather than building yet another large language model, the Mira Murati startup is focused on what it calls “interaction models” – systems designed from the ground up to converse, watch, and respond continuously. This marks a clear pivot from static, prompt–response tools toward AI that behaves more like a colleague sitting next to you. For enterprises, Murati’s move signals a coming shift in expectations: AI will not just answer questions, but participate dynamically in how work gets done.
Inside the 0.4-Second Interaction Model
Thinking Machines’ flagship model, TML-Interaction-Small, responds in about 0.40 seconds, putting it ahead of Google’s Gemini-3.1-flash-live at 0.57 seconds and OpenAI’s GPT-realtime-2.0 at 1.18 seconds. But speed is only part of the story. Unlike traditional systems that wait for you to finish speaking or typing, this model processes audio, video, and text at the same time, updating its understanding every 200 milliseconds. That means it can listen while it talks, watch a video feed while replying, and adjust mid-sentence as new information arrives. In demos, the system counted exercise reps from video, translated speech in real time, and even noticed posture changes – all while maintaining a fluid conversation. This technical design moves human-AI interaction closer to the intuitive, overlapping exchanges people are used to with each other, rather than the rigid turns of today’s chatbots.
Collaborative AI Models That Mirror Human Teamwork
At the heart of Thinking Machines AI is an attempt to remove what the company calls the “bandwidth bottleneck” between people and algorithms. Current large language models treat interaction as a sequence of discrete requests: you send a prompt, wait, then receive a block of text. Murati’s team instead breaks collaboration into continuous micro-steps, where one part of the model manages conversation flow while another quietly tackles harder tasks in the background. This mirrors how humans think ahead while speaking, shifting focus as a meeting or discussion evolves. The goal is not just natural conversation, but an AI partner that can jump into a dialogue, share observations from video or audio, and adapt context as quickly as teammates do. For businesses, that could mean assistants that contribute proactively in meetings, operations, and training – not merely tools sitting behind a chat window.
A New Playbook for Enterprise AI Adoption
Thinking Machines’ approach hints at a different trajectory for enterprise AI adoption. Instead of deploying static chatbots to handle tickets or generate documents, organizations could integrate collaborative AI models directly into workflows where timing and context matter – live customer calls, safety monitoring, coaching, or control rooms. Because the system is built to watch, listen, and respond concurrently, it may fit more naturally into environments where humans already coordinate in real time. That could also reshape return-on-investment calculations: the value shifts from one-off task automation to continuous, context-aware assistance. For now, the technology remains in research preview, with limited access for partners and a broader release planned later. But Murati’s framing – that interactivity should scale alongside intelligence – is likely to resonate with enterprises that have tested standard large language models and now want AI that behaves less like a search engine and more like a colleague.