Weekly AI Roundup: Open agent standards, hybrid build flows

by TechHub

Recent updates in AI center around hardware research, adoption of open standards, expanded low/no-code automation, and new developer integrations. Microsoft’s latest work enables more adaptable agent systems, hybrid developer experiences, and improved governance for enterprise AI. New research tackles infrastructure efficiency and scalable model deployments.

AI Infrastructure: MOSAIC MicroLED Architecture

Microsoft Research introduced MOSAIC, an optical networking approach for data centers using bundles of slow microLED channels instead of a few fast ones. This increases throughput, reduces power use, and improves reliability—grouping thousands of channels can reach up to 3.2 Tbps while lowering costs and complexity for AI infrastructure. MicroLEDs also resist temperature problems and offer lower failure rates than lasers. MOSAIC is compatible with standard protocols, simplifying integration. Full details are available in technical papers.

Agent Interoperability and Azure AI Foundry

Azure AI Foundry added more open standards, enabling agents, apps, and enterprise data to interoperate via Model Context Protocol (MCP) and Agent2Agent (A2A). MCP lets agents share workflow context, while A2A models collaborative agent activity. Foundry supports thousands of connectors, unified monitoring, and comprehensive app integration—making multi-agent workflows possible across platforms and ensuring secure data access and compliance. Continuing last week’s movement toward open standards and orchestration, this update shows how Foundry leverages MCP and A2A for flexible agent systems.

Copilot Studio and Azure AI Foundry Integration

New resources explain how Copilot Studio’s low-code conversational agent builder connects with Azure AI Foundry, supporting custom model workflows and managed deployment. The workflow moves from quick agent creation to enterprise-grade deployment, bridging low-code tools and professional lifecycle management. Interviews and analysis highlight both current and planned integrations for a more unified Microsoft AI developer experience. These updates continue coverage of orchestration and automation, now connecting low-code agent design to pro deployment.

Copilot Studio: Document Automation and Hybrid Development

Copilot Studio demonstrated automation with no-code agents integrated into Power Platform and AI Builder, streamlining document-heavy tasks like vehicle permit processing. Resources cover setup, configuration, and monitoring, showing live metrics and error reduction. Additional analysis shows Copilot Studio connecting low-code and pro-code workflows—making it easier for both non-technical and experienced developers to collaborate and scale solutions. Teams can prototype quickly and migrate to professional-grade solutions as needed. Continuing last week’s coverage of hybrid workflow automation, this set of articles explores Copilot Studio's progress in team document workflows and solution scaling.

Language Model Updates and Migration

Microsoft has deprecated Phi-3 and Phi-3.5 models in GitHub Models, recommending that users migrate to Phi-4 and Phi-4-mini-instruct. Migration instructions and mapping are provided for updating workflows. Teams are encouraged to track changes for stability and current support. This builds on last week’s efforts to keep teams informed about model life cycles and minimize disruption in workflows.

Developer Tooling for Model Integration and Workflow

Visual Studio Code introduces the Language Model Chat Provider API (BYOK), letting extension developers integrate models from any provider for privacy and policy flexibility. Technical resources detail API usage and implementation. Warp has added embedded AI agents to its CLI, offering prompt-driven scripting, code review, and editing for terminal-focused workflows. These tools support easier daily AI integration in both IDE and CLI environments, following the trend of improving developer access to AI capabilities.

Azure AI Foundry Translation and No-Code AI Workflows

Azure AI Foundry's Translator API is now in public preview, enabling developers to add both neural and LLM translation features to apps with step-by-step multilingual integration. The Cozy AI Kitchen show explored CalcLM—a no-code option for bringing GPT-4.1 to spreadsheets via Azure OpenAI, allowing analysis and planning through natural language queries. Demos demonstrate agent customization and further integration possibilities. These resources extend AI accessibility across technical and business user scenarios, complementing recent efforts to make AI more usable for non-developers.

Avatar-Powered Education and AI Content Creation

This guide covers building avatar-powered educational solutions with Azure, incorporating neural text-to-speech, avatars, and secure CI/CD, storage, and identity management. Developers receive templates and instructions for automating onboarding and training resources with compliance controls. Supporting last week’s coverage on AI accessibility, Azure continues to offer tools for innovative and inclusive learning solutions.

AI Coding Platforms for Enterprise Apps

Empromptu launched “vibecoding” for explainable, compliant enterprise AI apps. The platform uses retrieval-augmented generation, LLMOps tooling, and output scoring—plus DevOps integration with SOC 2 governance and credit-based pricing. This matches last week's interest in enterprise-ready AI, highlighting the benefit of transparency and production monitoring.