As AI applications evolve, their increasing complexity poses challenges related to token allowances and prompt construction. To address this, the use of AI agents has emerged as a powerful approach. These agents autonomously make decisions in dynamic environments, akin to human assistants. Imagine an agent that not only books your vacation but also tailors recommendations based on your preferences, plans your itinerary, and even invites friends. In the workplace, agents can execute tasks like sending emails or managing calendars. Google's Astra and OpenAI's GPT-4o exemplify this trend.

In this talk I will explain through demonstration the merits of this approach and how we can use frameworks like Semantic Kernel and AutoGen to leverage this approach to break down complex tasks and allow AI to dynamically build applications from smaller pieces.