Building Production-Ready, Secure, Observable, AI Agents with Real-Time Voice with Microsoft Foundry
Summary
Microsoft Foundry has announced the general availability (GA) of its next-generation Agent Service, a modern API and developer platform for building and operating AI agents. This release includes GA for evaluations, tracing, and monitoring within the Foundry Control Plane, offering out-of-the-box and custom evaluators for continuous quality, groundedness, safety, and risk measurement. The Microsoft Foundry portal at ai.azure.com is also generally available, providing a unified interface for AI app and agent development. Additionally, Voice Live integration with Agent Service is now in public preview, enabling real-time speech-to-speech interactions. Foundry Agent Service supports private networking, various authentication methods for Model Context Protocol (MCP), and integrates with partner security solutions like Palo Alto Networks' Prisma AIRS and Zenity for enhanced runtime protection.
Key takeaway
For AI Architects and CTOs evaluating platforms for production-ready AI agents, Microsoft Foundry's GA of its Agent Service, Control Plane observability, and unified portal provides a compelling, secure, and scalable option. You should explore its private networking, robust authentication, and partner security integrations to ensure your agent deployments meet enterprise security and governance requirements, while the Voice Live preview offers immediate opportunities for natural, real-time voice interactions.
Key insights
Microsoft Foundry offers an enterprise-grade platform for building, securing, and observing AI agents with real-time voice capabilities.
Principles
- AI agents require continuous quality and safety measurement.
- Enterprise AI demands robust security and private networking.
- Unified platforms simplify scaling AI across organizations.
Method
Build agents using Foundry Agent Service SDKs (Python, JS, Java, .NET), integrate with Voice Live for real-time speech, and use Foundry Control Plane for continuous evaluation, tracing, and monitoring.
In practice
- Deploy agents via REST endpoints or directly to Microsoft 365/Teams.
- Use private VNets for isolated agent environments.
- Integrate Palo Alto Networks or Zenity for runtime security.
Topics
- AI Agents
- Microsoft Foundry
- Real-Time Voice AI
- AI Observability
- Enterprise AI Security
Code references
Best for: CTO, AI Architect, VP of Engineering/Data, Machine Learning Engineer, AI Engineer, MLOps Engineer
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.