Weekly AI Roundup: Agent Skills, Foundry Updates, and Secure Runs

by TechHub

The AI section outlines current innovations and tool adoption, especially among Microsoft and open-source developers. The focus is on new agent frameworks, skills SDKs, toolkits, and orchestration options for agentic applications, plus guidance on deploying secure and scalable AI solutions. This continues themes from last week, adding new practical toolkits for production-ready deployment.

Microsoft Foundry Ecosystem: Model, Agent, and SDK Advancements

Microsoft Foundry’s February 2026 release brings new models and agent features, picking up from recent updates on edge AI and hybrid privacy. Anthropic's Claude Opus/Sonnet 4.6 models enable deep reasoning, scalable deployment, and large context handling. Models like GPT-Realtime-1.5 and GPT-Audio-1.5 support internationalized and voice-driven use cases. The Grok 4.0 engine enhances agent workflows, and FLUX.2 Flex improves text-image generation for interface prototyping. Model updates complement new features for privacy and on-prem deployments. Foundry Local allows for disconnected, compliance-aware hardware. The Agent Framework (Python) reaches API stability, with better credential management, session orchestration, and migration instructions. Durable agent orchestration via Azure Functions and SignalR enables agents to work around long delays or restarts—useful in public sector and telecom scenarios. The new Foundry REST API v1 is stable, with SDKs for Python, .NET, JS/TS, and Java. It introduces consistent naming and credential handling. Migration is supported for previous versions. There are also improvements in the AI Toolkit for VS Code (v0.30.0), introducing debugging, a model inspector, and catalog tools for quick prototyping and release. Documentation now further supports onboarding and protocol usage.

Modular Skills and Agent Frameworks: Agent Skills SDK, Reusable Skills, and Dynamic Tool Discovery

The Agent Skills SDK, an open-source Python toolkit, helps developers package common agent knowledge as portable skills. These can be published or discovered across different storage providers (local, HTTP, Azure/S3, databases). Skills are loaded only when relevant to save resources. The SDK works with tools like LangChain and Microsoft Agent Framework, supporting skill reuse and modular agent composition. It is MIT-licensed and customizable for DevOps, incident response, and retrieval agents. Microsoft Agent Framework also now supports reusable skills for .NET/Python agents, packaged with scripts and configuration to support on-demand discovery. Skills are loaded only as needed, reducing context and token requirements. Guidance is included for safely sharing and maintaining skills, with forward-looking features for creating new ones dynamically. Developers can use mcp-cli, a Bun-powered CLI, for finding tools in a token-efficient way, letting agents only fetch what they need. This aligns with earlier updates on secure modular agent deployment.

Secure Agent Execution and Durable Tasks: Azure Container Apps Dynamic Sessions, MCP C# SDK

Updates in this area give developers ways to run untrusted or agent-supplied code in Azure Container Apps with dynamic sessions, offering sandboxed runtimes for various workloads. Integration with Agent Framework and MCP is supported, while Azure AD and OpenTelemetry are used for authentication and traceability. Templates and deployment instructions make rollout easier, focusing on safe and repeatable ephemeral compute. The MCP C# SDK v1.0 brings improved capabilities to .NET developers, providing durable AI operations, OAuth 2.0, client/server APIs, tool calling, SSE event streaming, and more. These changes enable secure, large-scale, and async tasks for agentic systems.

Architectures and Best Practices: High-Performance Agentic Systems, Open Standards

A Microsoft guide details best practices for enterprise-scale agentic AI engineering with Foundry and Copilot Studio, emphasizing differences from traditional chatbots, and focusing on autonomy, orchestration, and clear boundaries. The architecture makes use of tools like Microsoft Graph, Logic Apps, and Power Automate. Memory, access control, observability, and compliance requirements are addressed, with case studies for contract analysis and customer support. A separate panel review discusses how open standards (MCP, Agent2Agent, OpenTelemetry, OAuth) help with secure, interoperable agent deployments, as well as how to write requests-for-proposal and select technologies focused on compatibility.

Other AI News

VS Code has added five new agent features: skills on demand, message steering, integrated browser, conversation forking, and lifecycle hooks. These updates are aimed at improving productivity and automation for developers.