-
|
Looking at the samples for a ChatCompletionAgent I don't see anything that cannot be done with just using a ChatCompletionService. What am I missing here. Are agents in de context of Semantic Kernel only useful when you need orchestration or .. ? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
Hi @Expecho, good question. Under the hood both the agent and the service call the same AI model. The difference is in what the abstraction gives you. A If your scenario is just "call an LLM with some tools," then |
Beta Was this translation helpful? Give feedback.
Hi @Expecho, good question. Under the hood both the agent and the service call the same AI model. The difference is in what the abstraction gives you. A
ChatCompletionServiceis a direct call to the model with optional tools or plugins. An Agent, on the other hand, is designed to participate in orchestration. This means you can compose agents into larger workflows, coordinate them with other agents or services, and take advantage of patterns like sequential, concurrent, or handoffs.If your scenario is just "call an LLM with some tools," then
ChatCompletionServiceis perfectly fine. If you want that call to live inside a broader orchestration story, that’s when the Agent abstraction matters.