Ship agents faster with expanded model choice, voice agents, and continuous optimization

Microsoft Foundry introduces a model-agnostic platform allowing teams to adopt newer AI models as they emerge without overhauling existing enterprise systems, tools, or controls. The platform supports continuous improvement by evaluating and refining agents using production traces to balance quality, latency, and cost. Customers like Fashable report cutting product development time from months to weeks and reducing physical sample budgets by 60% by integrating Foundry’s capabilities with their workflows.

Frontier AI models from OpenAI’s GPT-6 family and Anthropic’s Claude Opus 5.5 are now available in Foundry, expanding the platform’s portfolio for developers to evaluate against their specific workloads. Voice agents, now in public preview, integrate natively into Foundry’s agent foundation, supporting spoken interactions across 80+ languages and 140+ locales with options for custom voices or avatars.

Long-running resilience features in Foundry Agent Service allow hosted agents to continue work after disconnections or unexpected stops, resuming from durable checkpoints. Microsoft Agent Framework enhances these capabilities with workflow state management, isolated code execution, and episodic memory to reuse successful task patterns, while the Foundry Toolkit for Visual Studio Code simplifies agent development and deployment.

Foundry’s continuous improvement loop turns production insights into measurable optimizations, reducing manual effort in addressing performance issues. Governance controls, including Microsoft Entra and Agent 365 integrations, enforce lifecycle policies such as blocking or disabling agents, while network egress rules manage outbound access for hosted agents.