On May 21, 2025, Microsoft Chairman and CEO Satya Nadella delivered the opening keynote. His keynote focused on ushering in a new era of agentic computing, where AI agents and copilots operate across the full technology stack — from developer tools to business productivity and cloud infrastructure.

Nadella outlined Microsoft's commitment to open ecosystems, model choice, and secure AI deployments, all working toward a unified vision: empowering developers across every layer of the stack to build an open, interoperable agentic web.

Microsoft laid out a bold vision for AI agents — not just as productivity tools, but as intelligent collaborators transforming development, operations, science, and the enterprise. At Janea Systems, we’ve long worked closely with Microsoft on complex system-level innovation.”

Mirza Baig, Business Development Director at Janea Systems

Here are the most important announcements made by Nadella:

1. Open-source Copilot extension is launched for Visual Studio Code
2. Autonomous Coding Agent is officially released on GitHub
3. Microsoft 365 Copilot reaches general availability with new Copilot Tuning capabilities
4. Azure AI Foundry expands to support over 1,900 models including Grok and Mistral
5. Copilot Studio is now integrated directly with Foundry for seamless agent deployment
6. Foundry Local announced for on-device agent development and local runtime
7. Windows AI Foundry launched with native MCP server and registry integration
8. Windows Subsystem for Linux (WSL) is now fully open source
9. NLWeb framework introduced to enable websites and APIs as MCP-compliant agent interfaces
10. Data announcements: SQL Server 2025 launched, Cosmos DB integrated into Fabric, Copilot added to Power BI
11. Azure becomes first cloud to bring NVIDIA GB200 GPU clusters online at production scale
12. Microsoft Discovery platform will accelerate scientific research using AI agents

These innovations kick off a new chapter in AI-assisted development — one where agents are embedded in every development platform, technology, or SDLC stage. At Janea Systems, we're excited about this shift, as it aligns with our mission to build practical AI solutions for enterprises. Now, let’s dig a little deeper into the details.

This week was a chance to look more deeply at the technology that drives it all…and is opening up new ways to approach & solve problems that will help people. This has always been my “why” and is what I love about my job.”

Bill Sanders, Business Development Director at Janea Systems

Copilots Everywhere

Microsoft is evolving Copilot from a productivity assistant into an intelligent collaborator, aiming to equip developers with more versatile tools for building enterprise agents. Together, these updates simplify the development process while supporting deployment across channels and compliance-heavy environments.

Copilot in Visual Studio

Microsoft brings together the Microsoft 365 Agents SDK, Azure AI Foundry, and built-in project scaffolding, along with testing and deployment support — all from within Visual Studio. It now also includes native support for TypeSpec for Copilot.

Azure SRE Agent

Designed to maintain and stabilize production cloud environments, the SRE agent dramatically improves incident response and system reliability. The Agent leverages telemetry data, logs, and LLM-powered observability metrics to perform root cause analysis and quickly recommend or automate mitigations.

Copilot Control System

With centralized orchestration of multi-agent workflows, Copilot Control System lets users configure agent roles, workflows, dependencies, and fallback strategies. It also handles real-time tuning via Copilot Studio and provides native support for A2A protocols.

AI Coding Agent

A multi-phase development partner that helps scope features, plan tasks, write code, generate unit tests, conduct static analysis, and even handle CI/CD handoffs. It uses structured context windows and integrates with GitHub repositories for full-lifecycle tracking.

Copilot Studio

Now supports Model Configuration Profiles (MCP) to define logic flows, multi-agent collaboration logic, access policies, and integration hooks. With this tool, teams can build complex agents with behavior orchestration tailored to different departments or user types.

Enterprise Integrations

Copilot’s reach extends to Workday, ServiceNow, and Microsoft 365. Users can execute HR, finance, and IT processes via natural language. Tasks like onboarding, ticket triage, or generating reports can now be delegated to secure, context-aware agents.

AI Travel Agent

A domain-specific Copilot handles everything from booking and itinerary changes to expense categorization and compliance checks. It interfaces with calendars, email, and ERP systems to streamline the full trip lifecycle while staying policy-aligned.

Infrastructure Behind It All

As AI workloads grow more dynamic and agents require context switching and multimodal inference, the underlying infrastructure must scale accordingly. That means optimizing for throughput, latency, and energy efficiency across silicon, systems, and software layers. Microsoft invests in systems that support multimodal inference, model diversity, and seamless deployment across cloud and edge.

Azure OpenAI Foundry

Azure OpenAI Foundry serves as the full-stack engine for building and operating AI applications and agents. It supports model fine-tuning, distillation, provisioning, and deployment at scale. The Foundry now supports over 1,900 models, allowing developers to select task-specific, reasoning, or multimodal models for diverse enterprise needs.

Grok Integration

Grok, developed by xAI, is now available on Azure. Developers can provision throughput once and flexibly use it across Grok and other models, streamlining compute resource management while enhancing the natural language responsiveness of Copilot-powered systems.

Cobalt ARM CPU

Microsoft’s custom Cobalt processors are optimized for cloud-native AI workloads. These ARM-based chips power low-latency, high-throughput compute in Azure, aligning with Microsoft’s push for performance-per-dollar and sustainable infrastructure scaling.

Edge-Optimized Models

Microsoft supports inference across a variety of hardware, from CPUs and NPUs to GPUs and edge devices, allowing for hybrid deployment strategies that reduce latency and preserve data privacy.

Windows AI Foundry

Built into the Windows platform, this toolset allows local model inference and fine-tuning on user endpoints. It includes built-in MCP servers for common system actions and a registry for securely discovering vetted agent interfaces on-device.

Supercomputing for Climate

Microsoft is scaling its investments in high-performance AI infrastructure to support weather forecasting and climate modeling. This includes AI-powered simulations aimed at disaster response, agriculture, and global sustainability goals.

What’s Next: Implications for Developers

On day 2 keynote, Jay from the Microsoft Core AI team introduced a transformative vision aimed at revolutionizing the developer experience in the age of AI. His primary objective is to reimagine the entire software development cycle with AI tools, with a special emphasis on full-stack agentic applications. This vision is anchored in three core principles:

• AI-driven tools rather than AI-assisted features
• overhauling the entire developer experience from end to end
• fostering an ecosystem rooted in openness and choice
• Demonstrations showcased GitHub Copilot's role as a proactive coding assistant:
• Created onboarding documentation and README with setup instructions
• Handled repetitive tasks like input validation, typants, and TODOs
• Used "NextEdit Suggestions" to proactively suggest code changes
• Auto-completed commit messages based on change context and style
• Enabled agentic workflows with GitHub Copilot using MCP (Model Context Protocol)

Our internal research about AI-Assisted Development Across the SDLC evaluated ChatGPT and GitHub Copilot across software engineering tasks. We explored whether these tools could independently handle development workflows without detailed, step-by-step instructions. The evaluation spanned simple and complex algorithms, code refactoring, bug fixing, boilerplate code generation, unit test creation, and documentation tasks.

Following this initial study, we launched a second phase focused on AI in Frontend and Backend Engineering. On average, engineers using AI completed tasks 30% faster than without it. AI tools accelerated project initiation, connected developers with relevant documentation and best practices, and served as valuable brainstorming companions.

Azure AI Agent Integration for DevOps & Incidents

Microsoft demonstrated how the Azure Site Reliability Engineering (SRE) agent can automatically detect incidents, initiate rollbacks, and generate detailed post-mortem summaries without human intervention. This drastically reduces downtime and the burden on on-call engineers.

Microsoft showcased GitHub Copilot's ability to support legacy system upgrades. Developers were able to migrate from Java 8 and RabbitMQ to Azure Service Bus using Copilot's guided process, including:

• Planning the migration, executing required code changes across multiple files, and summarizing modifications for review
• Extending beyond Java to .NET and COBOL systems

From our perspective, the vision laid out for intelligent DevOps agents aligns with the direction we're already pursuing with clients.

How Janea Systems can help:

• Enable secure data integration across enterprise platforms including SharePoint, Microsoft Fabric, and Databricks.
• Integrate AI evaluation tooling into DevOps workflows via GitHub Actions to ensure quality, compliance, and trust.
• Design and implement custom multi-agent workflows using Azure Foundry SDKs.

We believe that the combination of agentic automation and platform extensibility can dramatically accelerate modernization initiatives and reduce operational toil.

Foundry Agent Factory and Multi-Agent Systems

Foundry is Microsoft's platform for building, deploying, and managing AI agents at scale. Used by over 70,000 organizations, it combines model development, agent orchestration, and enterprise integration into a single system. Key features include a model leaderboard, integration with HuggingFace, real-time model routing, support for the Model Context Protocol (MCP), and many more.

Core components:

• Agent Service for scalable deployment and monitoring
• Agent Catalog with templates for fast agent development
• Agent Knowledge for connecting agents to enterprise data (e.g., Fabric, SharePoint, Databricks)
• Agent Tools with 14,000+ services via MCP and ADA
• Multi-Agent Workflows for complex decision-making

Additionally, Foundry incorporates Microsoft security tools:

• AI Red Teaming for adversarial testing
• Guardrails with content filters and prompt shields
• Microsoft Defender and Entra integration for monitoring and access control

Heineken, for example, uses Foundry's evaluation tools and personality checks to ensure their chatbot behaviors align with company values and ethical guidelines. Accenture has adopted Foundry to drive end-to-end business process transformations, citing a 30% increase in efficiency, a 20% cost reduction, and a 50% reduction in time to build AI applications.

Janea Systems engineers are equipped to secure agent access, integrate compliance tools, and design robust workflows. We provide the hands-on expertise needed to put these best practices into action quickly and effectively.

Local & Edge AI

Foundry now enables seamless deployment of AI models from the cloud to edge environments, supporting scenarios where low latency, data locality, or offline functionality is critical. Local model execution leverages GPU acceleration and is tightly integrated with the semantic registry, ensuring consistent versioning and discoverability.

Windows AI Foundry builds on this by offering a curated catalog of optimized models, allowing developers to run and manage models locally as a service. This setup is hardware-aware, with support across CPU, GPU, and NPU, and integrates into Windows' AI toolkit for uniform deployment.

The platform supports parameter-efficient tuning like LoRA (Low-Rank Adaptation) and retrieval-augmented generation (RAG). For example, the Filora video editing app demonstrated how a local RAG pipeline could dynamically select and apply visual effects based on semantic input, all while running directly on a device.

Throughout our work, we helped clients minimize sensor latency, optimize local computation under power constraints, and maintain data consistency across distributed systems. Our work involved:

• System-level engineering for OtoNexus diagnostic ultrasound device. We optimized ML models into C++ for real-time analysis, improved battery management, and secured diagnostic data transmission.
• Automating QA for Cognex's industrial vision systems with dynamic configuration workflows, reducing software engineering dependency, and increasing defect detection speed.
• Porting PyTorch to ARM64 for scalable deployment of AI models on low-power edge devices and resolving compatibility challenges.

Janea Systems supports organizations in operationalizing edge intelligence through a combination of software engineering, embedded development, and AI model optimization. Our team will assist you with:

• Optimizing AI workloads for edge use cases, implementing real-time inference pipelines, and minimizing resource usage without compromising accuracy
• Building edge-first agents tailored to regulated industries and embedding AI capabilities into devices
• Porting AI agents to Windows-native execution environments

Looking Forward

Microsoft Build 2025 confirmed what many of us already anticipated: AI agents are the next frontier. Not just for productivity, but for rethinking how systems are built, managed, and evolved.

At Janea Systems, we’re ready to help organizations get the most out of these innovations. Whether it’s optimizing AI workloads for edge deployment, integrating multi-agent pipelines, or ensuring secure data integration across platforms, our engineers bring hands-on experience and a problem-solving approach to it.