Nadella's Architecture Doctrine on the July 29 Earnings Call: 'Keep Your Harness Separate From the Model'
On Microsoft's Q4 FY2026 call, Satya Nadella gave the clearest statement yet of Microsoft's agent strategy: 'We are very, very clear about the architectural sort of design of the platform, which is you got to keep your harness separate from the model, that means any model at any given time is swappable.' The argument is that enterprises should own the harness holding memory, context and action space, while the model underneath stays interchangeable across the 11,000+ models now in Microsoft's catalogue, including OpenAI, Anthropic, Mistral, xAI and Microsoft's own MAI family. For builders this is the same separation-of-concerns argument the harness-engineering crowd has been making, now stated as platform policy by the vendor holding stakes in two of the labs it is making swappable.
↳ Follow the thread