⚡️Satya Nadella: No Priors x Latent Space Crossover Special at Microsoft Build

What It Covers

Satya Nadella joins a crossover episode of No Priors and Latent Space at Microsoft Build 2025, outlining Microsoft's ecosystem strategy around AI harnesses, private evaluations, MAI model training, enterprise agent deployment, data center expansion, and how every company can operate at the intelligence frontier.

Key Questions Answered

• •Private Evals as Core IP: Every company should build private evaluation sets rather than relying on public benchmarks, which can be gamed. The true test of enterprise AI ownership is whether you can swap underlying models — from model A to model B — and still hill-climb on your private eval without leaking traces to vendors.
• •Hill-Climbing Scaffold Framework: Microsoft's MAI model strategy pairs a clean-lineage pretrained model with a hill-climbing scaffold that lets companies collect their own traces, build domain-specific reward functions, and create specialist agents from a generalist base — demonstrated by a 5B reasoning model outperforming larger models on private tasks.
• •Harness Architecture Over Model Selection: The enterprise AI stack should prioritize the harness — defining models, data, and tools in a closed loop — over chasing frontier model upgrades. Microsoft's multimodal harness, available via Azure Foundry, proved this when its security tool found vulnerabilities that dedicated security benchmarks missed.
• •Agent Traces as Balance Sheet Assets: Enterprise value will increasingly reside in accumulated agent traces — the recorded interactions between human workers and AI agents over time. These traces train company-specific veteran agents, functioning like tacit institutional knowledge that was previously impossible to capture or transfer systematically.
• •Pricing Model Evolution: Enterprise AI pricing will layer three tiers: per-user subscriptions bundling usage entitlements, consumption-based metering for agent workloads, and outcome-based pricing — though Nadella notes customers consistently revert from outcome pricing once they realize it means sharing upside, making flexible hybrid models the practical standard.