At the core of Microsoft’s AI strategy is the Semantic Kernel, an open-source toolkit designed to simplify the creation and management of AI prompts. Initially introduced to streamline the development of retrieval-augmented generation (RAG) applications, it has since evolved into a comprehensive framework for creating and managing agent-based AI systems. As AI technology continues to advance, Semantic Kernel has emerged as a pivotal tool for developers looking to build sophisticated, autonomous AI applications.
During Microsoft’s Ignite conference in 2024, the company introduced several new features for Semantic Kernel, marking it as the preferred framework for large-scale agentic AI projects. These updates lay the foundation for the Semantic Kernel’s roadmap into 2025, with some of the first new features already available to developers. This shift signifies Microsoft’s commitment to positioning Semantic Kernel as an essential platform for developing AI applications that require robust orchestration and intelligent agent interactions.
One of the most significant additions to Semantic Kernel is the Agent Framework, which is moving out of preview and into general availability. This shift will ensure that the tools provided within the framework are stable and fully supported for enterprise-level applications. The Agent Framework will also serve as a cornerstone for Semantic Kernel’s integration with Microsoft Research’s AutoGen, as well as the release of a unified runtime for agents built from both platforms. This will enable developers to create more powerful and flexible AI systems capable of performing complex tasks autonomously.
The Agent Framework is designed to simplify the creation of “goal-oriented” applications by allowing developers to build agent-based workflows. These workflows let AI agents handle specific tasks, often involving collaboration between multiple agents to complete complex operations that span various APIs and data sources. As an extension to the Semantic Kernel, the Agent Framework is delivered as a set of .NET libraries that facilitate human-agent interactions and provide access to OpenAI’s Assistant API. It is primarily controlled via conversational interfaces, though it can also respond to system events and incorporate approval processes into dynamic workflows. This allows developers to leverage agents to manage intricate tasks and integrate them seamlessly into larger, more sophisticated AI applications.