Weekly AI Roundup: Agent Frameworks, Foundry Models, VS Code

by TechHub

This week, AI updates include new tools, frameworks, and guidance for implementing agent-based systems, multimodal applications, and workflow integration across Microsoft Foundry, VS Code, and Azure. Progress continues on cross-language agent frameworks, enterprise modeling, and affordable deployment, all supporting faster, more flexible development spanning industries such as healthcare, telecom, and business software. Live events and guides make these advancements easier to adopt and use in practice.

Microsoft Agent Framework: Cross-Language Orchestration and Migration

Microsoft Agent Framework is now a Release Candidate, providing cross-language orchestration for AI agents in .NET and Python. The framework delivers a unified way to build agents, type-safe APIs, and graph-based workflow support, working with major providers like Microsoft Foundry, Azure OpenAI, OpenAI, GitHub Copilot, Anthropic Claude, AWS Bedrock, and Ollama. Features include orchestration packages, migration guides, and support for standards like MCP, A2A, and AG-UI. Upgrading from Semantic Kernel or AutoGen is described in dedicated resources, with community support for adoption. This milestone completes the journey from initial previews to a unified platform for managing .NET and Python agents, building on momentum toward more standardized agent management.

Microsoft Foundry: Model, SDK, and Platform Updates

Microsoft Foundry brings in updated AI models and SDKs, including GPT-5.2, GPT-5.1 Codex Max, Mistral Large 3, DeepSeek V3.2, Kimi-K2 Thinking, Cohere Rerank 4, and enhanced image and audio generators. Serverless fine-tuning, persistent agent memory, and improved agent-to-agent communication support stronger workflow integration and visualization. Managed Foundry MCP Server now offers secure endpoints for running models, with a unified portal, improved VS Code integration, and updated SDKs for easier project standardization. Developers are encouraged to migrate from the AzureML SDK v1 in advance of retirement, with resources available for migration and deprecation. These features improve workflow flexibility and multimodal project support. This release builds on earlier efforts to integrate advanced models and align SDK experiences across different programming languages, making it easier to develop and migrate projects.

Building Practical Solutions with Microsoft Foundry Local AI

Foundry Local AI documentation showcases privacy-focused workflows, such as a smart building HVAC digital twin that simulates real-world HVAC systems using modular design and KPIs with 3D visualization. Features include debugging, Copilot integration, and ready-to-run deployments with BACnet, Modbus, and MQTT connections. Another guide reviews privacy-aware medical transcription with OpenAI Whisper and ASP.NET Core 10, covering secure deployment, API design, compliance, electronic health record (EHR) integration, and production roll out for healthcare. These examples provide deeper, enterprise-ready blueprints for using privacy-first, local and modular AI in regulated environments such as healthcare and facilities management.

Microsoft Foundry: Frontier Model Integration and Reasoning Agents Challenges

Claude Sonnet 4.6 is now in Foundry, featuring Opus-class reasoning with a beta 1 million-token context window and adaptive controls. New browser automation, search, summary, and compliance tools are available for enterprise AI. The Agents League challenge invites teams to build reasoning agent solutions, with starter kits, online sessions, and prize competitions for orchestration and integration. MCP and SDK-driven environments enable both programmatic and visual control, providing flexibility to experiment with practical agent-based designs. Claude Sonnet’s introduction and the Agents League demonstrate continued growth in model diversity and hands-on adoption through challenges.

Agentic AI Workflows and Integration in Visual Studio Code

VS Code introduces more agent-oriented features, supporting orchestration, team collaboration, and background work assignment, as well as real-time security and workflow metrics for team environments. Events highlight agent coordination and monitoring with actionable telemetry. Tutorials cover everything from prompt-based prototyping and live workflow demos to building AI-powered projects with little to no code. These improvements follow recent enhancements to SDK support and workflow monitoring, providing secure and versatile agent-driven development tools for all experience levels.

Affordable AI Development and Practical Tutorials

The Budget Bytes series provides step-by-step guides to build working AI applications on Azure for less than $25, covering topics like inventory control and insurance. Each tutorial comes with sample code (GitHub repos), deployment steps, and Azure SQL trials so that teams can manage costs and debug issues easily when starting or prototyping with large AI projects. With a focus on hands-on learning and transparent costs, these resources help new teams build practical AI solutions.

Telecom Infrastructure, Agentic BSS, and Intelligent Edge

Telecom solution updates analyze AI’s role in enhancing mobile networks and business support systems. AI improves radio access networks with spectrum management and anomaly detection—powered by Azure and Foundry. Open RAN projects and the Janus initiative offer orchestration and support for global testing. Guides for agent-driven business support use Copilot Studio and MCP to automate telecom billing, quoting, and other tasks through APIs and rapid prototype workflows. Mobile World Congress sessions walk developers through live agent-building. Industry adoption and workflow integration remain in focus as telecom projects highlight AI and agent tools for practical transformation.

Developer Workflow and Software Practices in the Age of AI Coding

Martin Fowler’s session reinforces the importance of test-driven development (TDD) in a world where AI now writes both code and tests. TDD still provides discipline, reduces risk, and ensures code quality when agent outputs can be unpredictable. With AI’s help, security and review cycles increase in importance, and teams may need to shift from manual coding bottlenecks to stronger coordination. Fowler advises keeping Agile principles and adapting them to fit AI-driven and agent-rich workflows. This practical guidance underscores how long-standing engineering methods still matter as AI changes processes.

Other AI News

SQUAD, an open-source orchestration framework for .NET, provides design patterns and building blocks for .NET and Azure teams to create AI agent teams for cloud and hybrid projects. The trend toward agent frameworks for .NET continues, making orchestration and team-based build patterns more approachable.