Keep pulling the thread on Satya Nadella.
Microsoft built more Azure capacity in the last 15 months than it did in the first 15 years of the platform's existence.
In a demo for Land O'Lakes, Microsoft used traces from a frontier model, referred to as "GPT-5.5", to fine-tune a 5-billion parameter reasoning model to achieve superior performance.
Satya Nadella predicts that within 6 months, it will be common for professionals to use long-running AI agents, or "autopilots," that work overnight on their behalf with delegated authority.
Microsoft has developed a platform called the "GitHub Harness" for managing AI models, data, and tools, which is used across its internal products and is available to customers via its Foundry platform.
Microsoft's WorkIQ product exposes data from Microsoft 365 applications like Teams and Word, creating what Satya Nadella considers one of a company's most important and previously untapped databases.
Satya Nadella predicts that AI agent-based usage of Microsoft 365 data will eventually exceed direct human usage and create a 10x larger value opportunity.
The Azure networking team at Microsoft has redefined its mission from directly managing the network to building an AI agent system that automates Azure networking operations.
The Azure networking team at Microsoft has shifted its resource requests from headcount to compute tokens to manage its operations.
Satya Nadella believes a significant startup opportunity exists in creating a new type of university or a new educational pedagogy.
Microsoft's MAI models are developed using a "clean lineage" approach, which emphasizes high-quality pre-training data and extensive ablation studies.
Satya Nadella states that Microsoft underestimated the complexity of deploying large language models to deliver measurable value in real-world applications.
Satya Nadella states that a critical lesson learned at Microsoft over the last two years is that preparing the context layer is the most important work for making AI agents execute plans efficiently.