AI: "Real", "Open" and at "Scale"

Callouts from this week's Google I/O and Microsoft Build 2025

• What is fascinating is some of the announcements made at Google I/O this week (From Research to Reality) like “Agent Mode” under project Mariner, comes as part of Gemini app later in the year – find property, schedule inspections and keep refining it based on various filter criteria… https://www.zillow.com/ …Project Starline (Google Beam) now opens up 3D teleconferences, and Google Meet soon have real time speech translation capabilities. In less than a year, Google’s AI token has grown by 50x, 7m+ developers on Vertex AI and advancements in Gemini model (2.5 pro) shows exponential growth in the AI landscape.

• Microsoft going “Open” at MS Build 2025:  The message is clear: Microsoft is leaning to be more “Open” in the AI World. Moving from discrete apps to a open, scalable platform where AI agents become core to how we work, build, and interact with technology. The term Open Agentic Web was coined to explain a Vision of Interoperability and Shared Intelligence. Five callouts

• Unprecedented Scale: With Visual Studio and its family having 50 million users and GitHub 150 million, the reach is immense. GitHub Copilot, already used by 15 million developers, is just "getting started."  - In a way a global talent pool being equipped with next-generation capabilities.

• Pair to Peer Programmer: Satya demonstrated assigning a GitHub issue directly to Copilot, which then created a plan, a branch, and began work. isn't just delegation; it's augmenting development teams with AI peers, capable of tackling complex tasks like framework upgrades (Java 8 to 21, .NET 6 to 9) and cloud migrations.

• Openness as a Catalyst: A pivotal announcement was the open-sourcing of Copilot in VS Code, integrating AI-powered capabilities directly into the core of dev tool. This move, alongside continued development of GitHub Copilot, fosters an open ecosystem where developers can extend, customize, and build upon these agentic capabilities.

• Copilot Tuning: Your AI, Your Company's DNA: One of the most impactful announcements for enterprises in my view is "Copilot Tuning." Allows to fine-tune models on own company's data, workflows, unique tone, language, and specific expertise. E.g. consultancy could tune models for specific vertical industry know-how.

• Model Choice and Flexibility: Azure Foundry now supports a vast array of models now (1,900+), including the latest from OpenAI (Sora is coming next week). A new "model router" will automatically choose the best OpenAI model for a given job. Crucially, Microsoft announced that Grok from XAI is coming to Azure. This commitment to model diversity, including Llama and Mistral, is significant. Provisioned throughput can now be used across multiple models, a game-changer for cost and flexibility.