Section 2 of Building Agentic AI Applications with LLMs
Abstract
This section serves as the introductory preamble for the curriculum on Building Agentic AI Applications with LLMs, establishing the necessary conceptual groundwork before advancing to implementation details. Its central technical contribution is the formal delineation of Agentic AI within the context of Large Language Models (LLMs), distinguishing autonomous capability from standard pass-through generation. This framing matters profoundly within the deck’s progression because it sets the boundary conditions for system design, ensuring that foundational principles of agency and responsible AI governance are understood prior to architectural engineering.
Key Concepts
- Agentic AI Definition: The section explicitly interrogates the definition of Agentic AI, positing that an agent is distinct from a standard model by its capacity for autonomous action and goal-directed behavior. This concept is the primary focus, requiring the learner to distinguish between reactive models and systems that can perceive, plan, and execute tasks within an environment.
- Foundational Principles: The term “Foundations” indicates that the course will not solely rely on heuristics but will ground Agentic AI in underlying theoretical constructs. These foundations likely encompass the mathematical and logical structures necessary for agents to function reliably, serving as the bedrock for all subsequent technical modules.
- Responsible AI Framework: The inclusion of “Responsible AI” in the preamble suggests that ethical constraints and safety mechanisms are treated as integral to the system architecture, not as post-hoc considerations. This concept motivates the integration of guardrails and bias mitigation strategies at the earliest stages of agent design.
- Course Preamble Status: The designation of this content as a “Preamble” implies its role as a prerequisite knowledge state for all following sections. It establishes the vocabulary and context required to interpret the technical specifications that follow in the rest of the deck.
- LLM-Based Agency: Drawing from the deck title, this concept merges the probabilistic reasoning of Large Language Models with the deterministic requirements of agent loops. It highlights the specific challenge of maintaining coherence when using stochastic models for autonomous decision-making.
- System Boundaries: The section implicitly defines the boundary between the internal logic of the AI and its external environment. Understanding this boundary is crucial for determining where human oversight is required and where the agent assumes full control.
- Goal-Directed Behavior: A core component of the “What is Agentic AI” inquiry is the shift from pattern matching to intentional action. This concept necessitates a discussion on how objectives are represented and optimized within the context of an LLM-driven system.
- Contextual Awareness: Implicit in the definition of agency is the ability to retain and utilize context across multiple steps. This concept is foundational to the technical implementation of long-horizon tasks, where information must persist beyond the immediate inference window.
Key Equations and Algorithms
- None. This section focuses on conceptual framing and definitions rather than specific mathematical formulations or algorithmic procedures. No equations or pseudocode are provided in the text of the source slides.
Key Claims and Findings
- The section claims that Agentic AI requires a distinct definition separate from traditional generative models, centering on autonomy and task completion.
- It asserts that Responsible AI considerations must be addressed at the foundational level of the curriculum, preceding technical application.
- The material posits that understanding the fundamental nature of agency is a prerequisite for building successful LLM-based applications.
- It establishes that the course structure is built upon a preamble designed to align learner expectations with the safety and theoretical constraints of the domain.
- The text suggests that the core challenge in Agentic AI lies in bridging the gap between probabilistic language generation and deterministic action execution.
Terminology
- Agentic AI: A system architecture utilizing LLMs to perform autonomous actions, plan sequences, and interact with environments to achieve specific goals, distinct from passive query-response models.
- Foundations: The theoretical base upon which the practical application of Agentic AI is built, encompassing logic, control loops, and safety theory.
- Responsible AI: A design philosophy emphasizing safety, fairness, transparency, and accountability in AI systems, treated here as a mandatory constraint rather than an optional feature.
- Preamble: The introductory section of the course material intended to define scope, terms, and ethical boundaries before technical instruction begins.
- LLM (Large Language Model): The underlying probabilistic engine referenced in the deck title, serving as the reasoning component within the proposed Agentic AI applications.
- Agency: The capacity of a system to act independently and make decisions without direct human intervention for each step, a key property defined in the section.
- Application: The practical deployment context for Agentic AI, indicating that the theoretical concepts are intended for real-world utility and system building.
- Architecture: Refers to the structural design of the AI system, which the preamble prepares the learner to modify or construct by defining the role of the agent within it.