Azure AI Foundry receives new features for building, managing, and evaluating multi-agent AI systems. Additionally, a public preview of an AI Red Teaming Agent is introduced to automatically detect risks.
Microsoft announces new features within Azure AI Foundry. This platform was introduced by Redmond in 2024 as a solution for businesses to design, customize, and scale AI applications. Central to the update is the general availability of the agent framework: an extension of the open-source Semantic Kernel project. This extension aims to simplify the orchestration of multi-agent AI systems.
The new framework allows developers to have specialized AI agents collaborate within a single coordinated system. Microsoft emphasizes that this reduces the complexity of development processes, as less code is required.
Additionally, Microsoft introduces an observation and feedback system that enables developers to monitor, A/B test, and improve AI agents based on user feedback. This cycle of continuous improvement is intended to enhance the reliability and performance of AI systems.
Agent for Agents
To manage the growing risk of more advanced AI agents, Microsoft is also launching the AI Red Teaming Agent in public preview. This agent systematically tests AI systems for vulnerabilities and integrates with Microsoft’s own PyRIT framework. The Red Teaming Agent is designed to automate risk assessments and help development teams build safer AI solutions.
Again with an eye on safety and quality, Microsoft introduces new evaluation tools for agents. These allow developers to assess the behavior of AI agents for reliability and regulatory compliance.
Finally, Microsoft announces new integrations with development tools such as GitHub and Visual Studio Code. Developers can now work directly within their IDE on agent-based applications thanks to a new extension for Visual Studio Code. Agent mode in GitHub Copilot is also becoming widely available, helping developers perform complex modifications and test code.