Microsoft's Semantic Kernel - Finally, Intelligent Software Agents
Microsoft's Semantic Kernel tools and framework for building AI agents using large language models finally delivers on the decades-old vision of intelligent, autonomous software agents.
Summary
- Semantic Kernel provides tools and runtime for building contextual conversations and interactions using AI like OpenAI's models.
- It manages conversation state, history, inputs and outputs allowing developers to focus on business logic code.
- Plugins abstract complex AI service interactions into simple method calls that can be orchestrated via chat.
- Semantic Kernel can mix and match different AI models, ground them with real-world data and orchestrate workflows.
- Its planners allow creating autonomous workflows that can call plugins conditionally to complete tasks described in chat.
- Prompt Flow tool can evaluate accuracy of plugins and planners using benchmark data.
- Approach differs from original 1990s agent concept as its not running remote arbitrary code but orchestrating cloud services.
- Modern APIs and microservices provide secured way to query and extract data instead of running untrusted code.
- Decades of agent research has finally resulted in tools like Semantic Kernel realizing the vision using latest AI.