Using intelligent planners in our AI Agents with the Semantic Kernel SDK and C# 🤖

Using the Semantic Kernel SDK, we can integrate NLP prompts and custom plugins with Large Language Models (LLMs). In previous posts, I talked about how we can manually invoke prompts and functions.

In this video, we're going to talk about Intelligent Planners and how they support automated task planning. Planners are dynamic tools that use AI to automatically coordinate which plugins and functions should be invoked based on our user's requests.

In this video, we'll learn how we can use the Semantic Kernel SDK to automatically invoke functions, and how planners can generate plans to accomplish tasks. We'll also talk about how we can fine-tune planners to optimize their performance!

Contents

0:00 What are planners?
1:11 Using a planner
1:43 Optimizing our planners
3:15 Setting up our application
3:50 Installing the Handlebars package
4:15 Creating our MusicConcertPlugin
5:13 Creating our Prompt
7:11 Creating and invoking our planner
10:35 Generating a plan template
13:35 Using a handlebars template
17:08 Wrap up

Learn more!

Connect with me!

[James_PET]

Hi Will, amazing video. You just got a new subscriber ➕. A few suggestions:
1. Do not always start the code from the beginning; build up on the existing videos
2. It would be more interactive if you do not have hardcoded user's input but actually start interacting with the chat by typing messages
3. I would love it if you could show more complex solutions with the automatic planner - e.g. RAG+SQL. Here it is an example:
We have an Azure AI search field with all articles from a website (e.g. CNN), how to handle queries:
- give me a summary of "title of the particular article" - RAG
- Who won Formula 1 2023 - again, RAG simple query against azure AI search
- give me all that we published in the last 10 days - this is maybe an SQL query instead Azure search

It would be amazing to see how to build an automated planner that can handle all these scenarios

[asdasdaa]

Will, this was such a great video! Please make more content on Semantic Kernel! Subbed to your channel thanks!

[TheLeads]

As per Microsoft : Handlebars planners will be deprecated in a future release of Semantic Kernel.

[yaojingquek]

Great video dude! Just a question, Is planner can be used along side with Semantic Kernel Agent? Say that if I have multiple Agent, and every agent has their own plan?

[edwardcarandang]

Great tutorial!!! Subscribed!!! Will definitely watch this space.

[CecilPhillip]

Such a great series. Any plans on covering Prompt flow?

[AliIssa]

Hello Will! Really great tutorials. I hope you continue producing more content on SK. I have a question, so we have to use a planner in order to execute a native function based on the users input?

[marcinterlecki]

Hi Will. I am new to Semantic Kernel and its planners and there is one thing here I don't understand fully. I though planner is all about model matching the goal (expressed during runtime by human in natural language) with means to achieve this goals, pre-prepared and available to model (functions in plugins). But here you seem to use intermediary step in the form of skprompt; the match is really between pre-prepared skprompt input variables descriptions (which are controlled by programmer, you in this case) and preprepared available functions descriptions, also controlled by programmer. ?