AI Plugins & Native Code

An LLM is a thinker, not a doer. Plugins (formerly Skills) are the "Hands" of the AI. They allow the model to reach out and touch your database, your file system, or your internal APIs.

1. Semantic vs Native Plugins

• Semantic Plugins: Prompts saved as code. Example: `RewriteToProfessional`.
• Native Plugins: Real C# methods. Example: `GetStockPrice(string symbol)`.

2. The [KernelFunction] Attribute

To turn a C# method into a plugin, you just add an attribute.

[KernelFunction, Description("Fetches the weather for a city")]
public string GetWeather(string location) { ... }

The AI reads the **Description** to understand when it should call your function. Metadata is code!

4. Interview Mastery

Q: "What is 'Tool-Calling' and why is it better than simple text parsing?"

Architect Answer: "Tool-calling (or Function Calling) is when the LLM returns a structured JSON payload like `{ "function": "GetWeather", "args": { "location": "Dubai" } }` instead of just text. This is safer because the model has been trained specifically to output these parameters accurately. It allows for reliable integration between non-deterministic AI and deterministic C# business logic."