Mira Murati's 'Interaction Models' Challenge Core AI Paradigm

'''Former OpenAI CTO Mira Murati’s new venture, Thinking Machines, has revealed its focus on "interaction models," a direct challenge to the dominant request-response paradigm of current generative AI. This move reframes the race for AI agents away from mere task completion and toward creating a persistent, collaborative intelligence layer. It strategically positions the company to redefine human-computer interaction, representing a fundamental threat to incumbents whose value is tied to the current chatbot-style interface. This mirrors the industry-wide scramble to find a post-smartphone form factor, but approaches it from a software-first principle, unlike the hardware-led attempts of Humane or Rabbit. At a technical level, interaction models fundamentally alter the AI stack by shifting from discrete, stateless API calls to continuous, state-aware multimodal processing of audio and video. This creates an asymmetric advantage for any company that can solve the immense power consumption and on-device processing challenges. The primary losers in this scenario are the thousands of "wrapper" applications built on APIs from OpenAI and Google, which risk becoming obsolete. The competitive response from foundation model providers will be a forced recalculation, pressuring them to evolve beyond simple API endpoints and build their own integrated, persistent interaction frameworks, as previewed by Google’s Project Astra. The real test for Thinking Machines will be whether it can create a trusted, secure user experience for an "always-on" AI, a hurdle that has doomed previous ambient computing ventures. In the next 12-18 months, the key indicator will be the release of a developer SDK, which would signal a go-to-market strategy focused on building an ecosystem, not just a product. This trajectory suggests a future where the primary interface for computing is a cognitive partner, not a series of applications. The critical variable is whether the immense technical and privacy hurdles can be overcome before incumbents can pivot their own platforms.'''