Thinking Machines

Overview

Thinking Machines is an AI company led by Mira Murati and positioned around human–AI collaboration, model interactivity, and frontier model development. In recent coverage, the company has been associated with two core themes: building highly interactive AI systems for end users, and advancing a modular technical approach to agent design through its published work on "Interaction Models." For AI Product Managers, that combination makes Thinking Machines notable not just as another model lab, but as a company explicitly framing interaction as a first-class product and model capability.

The company also appears to be investing across the full stack required to support that vision: leadership in research and engineering, large-scale compute infrastructure, and a product thesis centered on persistent memory, retrieval-augmented generation, and reactive planning. That makes Thinking Machines relevant to AI PMs tracking where the market may be headed beyond one-shot chat interfaces toward more stateful, collaborative, and task-oriented systems.

Key Developments

• 2026-01-15: Mira Murati announced a CTO transition at Thinking Machines. Barret Zoph departed, and Soumith Chintala was named the new CTO.
• 2026-03-11: NVIDIA partnered with Thinking Machines to deploy at least 1 gigawatt of Vera Rubin systems for frontier AI model training, signaling major investment in compute capacity.
• 2026-05-12: Mira Murati launched Thinking Machines' first interactive AI platform, describing human–AI collaboration and built-in model interactivity as central to the product vision.
• 2026-05-12: Thinking Machines published a technical report on "Interaction Models," outlining a modular agent framework that combines persistent memory, retrieval-augmented generation, and reactive planning, along with early evaluation results showing improved long-context performance.

Relevance to AI PMs

• Designing products beyond chat: Thinking Machines is pushing the idea that interactivity should be built into models themselves, not added as thin workflow layers. AI PMs can use this lens when evaluating whether their products need deeper state, turn-taking, memory, and collaborative controls.
• Operationalizing agent architectures: The company's Interaction Models framework highlights practical building blocks such as persistent memory, retrieval-augmented generation, and reactive planning. PMs can map these modules to product requirements like personalization, knowledge grounding, and multi-step task completion.
• Planning for infrastructure-product fit: The NVIDIA and Vera Rubin partnership suggests Thinking Machines is aligning ambitious product behavior with frontier-scale compute. For AI PMs, this is a reminder that differentiated UX often depends on model training, inference budgets, and systems architecture—not just prompt design.

Related

• Mira Murati: Founder and public face of Thinking Machines; central to the company's human–AI collaboration narrative.
• Barret Zoph: Former CTO whose departure marked an early leadership transition.
• Soumith Chintala: Named CTO, indicating the company's emphasis on technical and research leadership.
• NVIDIA: Infrastructure partner supporting large-scale frontier model training.
• Vera Rubin: NVIDIA system architecture referenced in the reported 1 GW deployment for Thinking Machines training workloads.
• Interaction Models: Thinking Machines' technical report and framework for modular, interactive agent systems.
• Persistent Memory: A key component in the company's modular architecture for more stateful AI behavior.
• Retrieval-Augmented Generation: Part of the framework used to ground outputs in relevant information.
• Reactive Planning: Another core module in the Interaction Models approach, aimed at dynamic multi-step behavior.
• @thinkymachines: Alias used in social mentions and announcements.