Microsoft Foundry now supports continuous improvement of AI agents, allowing organizations to build, run, govern, and progressively enhance their agents over time. Leveraging a broad ecosystem of AI models and intelligent routing, Foundry reduces cloud costs while giving developers granular control over updates and deployments.
- Continuous agent lifecycle integrates coding, deploying, evaluating, and refining.
- Model router cuts cloud compute costs by intelligently balancing workloads.
- Security features maintain strict access and human verification during runtime.
Infrastructure signal
Microsoft Foundry introduces a comprehensive system that runs AI agents at global scale using a multi-model architecture encompassing over 11,000 models from providers like OpenAI, Anthropic, Meta, and Microsoft’s own. This model routing system intelligently assigns workloads based on task complexity, reducing unnecessary compute use and delivering cost savings up to 65%. By layering Entra Agent ID and runtime guardrails, Foundry enforces least-privilege access and ensures sensitive actions require human authorization, enhancing security and compliance throughout the cloud environment.
The platform supports flexible deployment options including a choice of orchestration layers, with Foundry Agent Service handling high concurrency and rapid session start. Additionally, Microsoft’s Copilot harness offers a managed runtime providing reusable skills and modular sub-agents to accelerate agent building and deployment. This infrastructure design empowers organizations to maintain control over their AI workloads while optimizing reliability and operational cost-effectiveness within Azure’s global cloud environment.
Developer impact
Developers benefit from seamless integration with GitHub Copilot, which generates production-ready code based on plain language descriptions, speeding up agent creation without deep expertise in AI model selection or orchestration. The underlying Copilot Fusion technology automates model routing decisions to balance performance, latency, and cost, relieving developers from manual tuning and enabling focus on business logic.
Once deployed, developers gain continuous feedback loops via automated agent evaluations defined through customizable rubrics. These evaluations score agent outputs against team-approved criteria, allowing iterative improvements driven by empirical data and expert feedback. This system fosters a DevOps-like workflow for AI where changes are validated before production release, enhancing reliability and accelerating innovation in AI-powered automation.
What teams should watch
Product and engineering teams should closely monitor agent performance metrics and evaluation results to identify areas for optimization. Foundry’s continuous improvement process—establish, identify, optimize, validate—relies on iterative testing and validation cycles which require active engagement by domain experts to define success criteria and approve changes. This ensures AI behavior stays aligned with evolving business objectives and quality standards.
Security and compliance teams must leverage Foundry’s runtime guardrails and agent identity controls to maintain governance over AI actions, particularly where human oversight is mandated for sensitive workflows. Understanding the interaction between agent policies and organizational standards will be critical in scaling AI deployments safely. Teams should also assess cost implications and benefits of model routing strategies, balancing the tradeoffs between model accuracy, latency, and cloud expenditure.