GenAI PMThinking Machines
Mira Murati’s AI company, noted here for launching an interactive AI platform and publishing Interaction Models. It is positioned around human-AI collaboration and model interactivity.
Key Highlights
- Thinking Machines is positioned around human–AI collaboration and model interactivity rather than purely autonomous AI.
- Its Interaction Models framework combines persistent memory, retrieval-augmented generation, and reactive planning.
- The company launched its first interactive AI platform in May 2026 with interactivity framed as a core model capability.
- A large NVIDIA partnership for Vera Rubin systems signals serious frontier-scale training ambitions.
- For AI PMs, the company is a useful reference point for designing assistants that are collaborative, stateful, and modular.
Thinking Machines
Overview
Thinking Machines is an AI company led by Mira Murati and positioned around human–AI collaboration, model interactivity, and frontier-scale model development. In newsletter coverage, the company stands out for two parallel bets: building a large-scale training and infrastructure footprint for advanced AI systems, and introducing an interactive AI platform that treats interactivity as a core property of the model rather than a thin interface layered on top.For AI Product Managers, Thinking Machines is notable because it represents a product philosophy that goes beyond static chat or fully autonomous agents. Its work on “Interaction Models” suggests a modular approach to agent design that combines persistent memory, retrieval-augmented generation, and reactive planning to improve performance on long-context and multi-step tasks. That makes the company relevant not just as a model builder, but as a signal of where product architecture for AI assistants and agents may be heading.
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, highlighting a major investment in compute capacity.
- 2026-05-12: Mira Murati launched Thinking Machines’ first interactive AI platform, framing human–AI collaboration as a core design principle and arguing that interactivity should scale with model intelligence.
- 2026-05-12: Thinking Machines published a technical report on “Interaction Models,” describing a modular agent framework built from persistent memory, retrieval-augmented generation, and reactive planning, along with early evaluation results showing strong gains on long-context tasks.
Relevance to AI PMs
- Designing interactive AI products: Thinking Machines reinforces the idea that interactivity should be part of the model and system design, not only an orchestration layer. PMs building copilots, assistants, or workflow agents can use this lens to evaluate whether their product supports iterative clarification, memory, and adaptive collaboration.
- Choosing agent architecture patterns: The company’s “Interaction Models” framework highlights practical building blocks—persistent memory, retrieval-augmented generation, and reactive planning—that PMs can map directly to product requirements such as personalization, knowledge grounding, and multi-step task completion.
- Planning for capability and infrastructure strategy: The NVIDIA partnership signals that frontier product experiences may depend on significant compute access. PMs should consider how infrastructure constraints, model latency, and training or inference scale affect roadmap choices, pricing, and product reliability.
Related
- Mira Murati: Founder and public face of Thinking Machines; central to its positioning around human–AI collaboration and interactive systems.
- Barret Zoph: Former CTO whose departure marked an early leadership transition at the company.
- Soumith Chintala: Named CTO after Barret Zoph; relevant as a technical leader shaping the company’s engineering direction.
- NVIDIA: Infrastructure partner working with Thinking Machines on large-scale deployment of Vera Rubin systems for frontier model training.
- Vera Rubin: The NVIDIA system referenced in the compute partnership, underscoring Thinking Machines’ frontier-scale ambitions.
- Interaction Models: Thinking Machines’ published framework for modular agent design and a key concept for understanding its product and research direction.
- Persistent memory: One of the core components in the company’s Interaction Models architecture, relevant for personalization and continuity across sessions.
- Retrieval-augmented generation: Another major architectural component, connecting model outputs to external knowledge sources.
- Reactive planning: A planning layer in the Interaction Models approach, relevant for dynamic multi-step task execution and agent behavior.
Newsletter Mentions (4)
“Thinking Machines published a technical report on “Interaction Models,” detailing their modular agent framework—combining persistent memory, retrieval-augmented generation, and reactive planning—and shared early evaluation results demonstrating marked improvements in long-con...”
#11 𝕏 Thinking Machines published a technical report on “Interaction Models,” detailing their modular agent framework—combining persistent memory, retrieval-augmented generation, and reactive planning—and shared early evaluation results demonstrating marked improvements in long-con... #12 📝 Simon Willison You Need AI That Reduces Maintenance Costs - James Shore argues that AI coding agents must substantially reduce maintenance costs proportional to the productivity gains they provide, otherwise increased output will multiply long-term maintenance burden.
“Mira Murati launched Thinking Machines’ first interactive AI platform to advance human–AI collaboration.”
#25 𝕏 Mira Murati launched Thinking Machines’ first interactive AI platform to advance human–AI collaboration. She argues interactivity must be built into models and scale with their intelligence, not just serve as scaffolding around autonomous cores.
“#5 𝕏 NVIDIA AI partners with @thinkymachines to deploy at least 1 gigawatt of Vera Rubin systems for frontier AI model training.”
The newsletter notes a large-scale deployment partnership between NVIDIA and Thinking Machines for frontier AI training. The emphasis is on compute capacity rather than end-user features.
“Thinking Machines CTO change: Mira Murati @miramurati announced Barret Zoph’s departure and named Soumith Chintala as the new CTO of Thinking Machines .”
AI Industry Developments & News Meta alum joins Airbnb: Sam Altman @sama congratulated Ahmad on joining Airbnb , highlighting the potential of AI in travel and experiences. Thinking Machines CTO change: Mira Murati @miramurati announced Barret Zoph’s departure and named Soumith Chintala as the new CTO of Thinking Machines . GPT 5.2 coding feat: Kevin Weil @kevinweil reported that GPT 5.2 ran for one week straight and generated 3 million lines of code , showcasing its endurance.
Related
A major AI infrastructure company building hardware and software for training and inference workloads. In this newsletter it is mentioned in connection with TokenSpeed and networking for large AI clusters.
A technique for grounding model outputs in retrieved information. It is cited here as a component of a modular agent framework.
Former OpenAI CTO and founder of Thinking Machines, quoted on the limitations of current AI interfaces and the importance of interactivity. She appears here as a product vision voice for human-AI collaboration.
Stay updated on Thinking Machines
Get curated AI PM insights delivered daily — covering this and 1,000+ other sources.
Subscribe Free