Weekly AI Roundup: Agents, Local Models, and Tool Integration
This week, the AI landscape included updated frameworks for agent-based development, additional options for local deployment, enhancements to developer tools, and resources for constructing agentic and generative AI systems. The focus continued to be on flexible integration—local or cloud—and supporting enterprise-scale agent architectures as well as individual productivity needs. Tutorials centered on adaptable model options, privacy, and streamlined AI workflow orchestration.
Agentic AI Development on Azure AI Foundry
Building on previous content about Agent Factory and orchestration, this week presents new guides and technical references for deploying agents within Azure AI Foundry. Resources support enterprise use, including documentation for MCP integration and Logic Apps connectivity. A multi-agent architecture reference, complete with industry case studies (such as cybersecurity and retail), extends earlier material about governance and versioning in production environments. Guidance on selecting models is now broader and more practical for first-time deployment. New resources walk through the use of the Foundry Model Catalog and Model Router, continuing to address compliance and matching use cases to specific business needs—reiterating points made in earlier roundups.
- Agent Factory: Building Your First AI Agent with Azure AI Foundry
- Designing Multi-Agent Intelligence: Microsoft Reference Architecture and Enterprise Case Studies
- How to Choose the Right Model for Your AI Agent: A Developer’s Guide
Advanced AI Features and Integration in Developer Tools
Extending earlier GPT-5 integration news, Visual Studio Code now allows developers to switch between GPT-5 and GPT-5 mini, giving them more direct control over price and speed. These adjustments are part of the Copilot/VS Code move toward greater convergence and personalization. Additional features such as ‘beast mode’ and task list automation give users new customized workflows. Azure AI Foundry’s new GPT-5 freeform tool calling allows for flexible Python/SQL execution, moving beyond previous, more restricted function-call API patterns. These capabilities reflect the increasing sophistication of agentic and orchestrated workflows. In addition, a new tutorial on Mistral Document AI provides hands-on steps for incorporating document parsing into developer environments, supporting conversion of unstructured PDFs and handwriting to structured, AI-ready data.
- Hello GPT-5 & GPT-5 mini: New AI Features in VS Code Agent Mode
- Unlocking GPT-5’s Freeform Tool Calling in Azure AI Foundry
- Mistral Document AI Integration with Azure AI Foundry
Local Model Hosting and Deployment with .NET and Foundry Local
Following prior coverage of local model inference, this week’s resources include detailed instructions for running open-source models such as GPT-OSS within C# projects using Ollama. These methods align with recent developments for Foundry Local and ONNX integration—letting developers deploy streaming chatbots and retrieval-augmented generation solutions on local machines. Forthcoming enhancements in Windows and hardware acceleration reinforce the trend toward hybrid AI workloads. “Beginner’s Guide” articles detail use of Foundry Local alongside Microsoft Olive, covering ONNX optimization, choosing formats, and troubleshooting—helping more developers move toward flexible AI deployments.
- Running GPT-OSS Locally in C# Using Ollama and Microsoft.Extensions.AI
- Beginner’s Guide: Using Custom AI Models with Foundry Local and Microsoft Olive
Workflow Automation and Copilot Studio Development
Continuing from last week’s focus on no-code tools, this week offers resources for deeper organizational integration in Copilot Studio. In-depth guides explain the creation of custom plugins and connectors, including advanced OpenAPI authentication, supporting more tailored organizational automation strategies. Step-by-step resources cover integrating Copilot Studio with Power Automate for process automation involving platforms like SharePoint and Dynamics 365. The system’s library of over a thousand prebuilt connectors further supports broader enterprise workflow automation.
- Creating Custom Plugins and Connectors in Copilot Studio
- Integrating Copilot Studio with Power Automate for End-to-End Workflows
Specialty Agents and Agent-Centered Design
This week’s updates highlight the transition from traditional user experience (UX) to agent experience (AX). The Lacuna agent, created using Copilot Studio and AI Foundry, is designed to identify hidden assumptions in product design, expanding on last week’s discussion of agent-based collaboration. Agents focused on domains such as code review and risk assessment are also featured. The discussion encourages a new focus on agent-centered design, emphasizing planning, orchestration, and domain expertise over simple chatbot-based systems. This builds on last week’s analysis of GPT-4-based planners and Semantic Kernel tools, demonstrating real-world adoption in risk and assumption management.