The speech delves into the evolution of artificial intelligence, drawing a clear distinction between traditional chatbots, which are reactive and based on scripts or predefined models, and modern AI Agents. The latter are proactive systems, capable of acting, executing complex, goal-oriented tasks, coordinating multiple steps, and interacting autonomously with external systems and databases.

The core of the presentation explores in detail the intrinsic limitations of LLMs (Large Language Models), such as their nature as sequential and stateless "prediction engines," poor multitasking capabilities, "margin bias" (more weight on initial and final instructions), and "context rot," which compromises the quality of their responses as interactions increase.

To address these challenges, two key disciplines are introduced: Prompt Engineering, defined as the art of writing precise instructions to obtain correct answers, and the more advanced Context Engineering. The latter is the art of organizing and providing the AI with all relevant information—clearly structured using tags or titles—to enable it to make better decisions and act more consistently and accurately, especially in the operational realm of agents.

Approaches like Zero-shot, Few-shot, and Step-back Prompting are discussed. In closing, the ReAct (Reason & Act) reasoning structure was introduced, explaining how AI agents manage to plan, act, observe, and update their reasoning, demonstrating how agents can automate complex processes, while also emphasizing the importance of human supervision.

Finally, the speech highlights the importance of adopting AI agents to automate complex processes and improve efficiency, while also reaffirming the necessity for continuous human supervision, as agents still operate within defined boundaries.

As a bonus, a creation from scratch of a simple yet able to search and scrape the web, writing documents, reasoning and remembering agent is shown.

This speech can be translated in various languages and can be simplified or more technical. The bonus live code can be replaced by demonstration.