Module 2: Foundations of Agentic AI Systems

Lesson 1: What is an Agentic System?

In the current AI landscape, "Agent" is a term used loosely by marketers, developers, and CEOs. To a Claude Certified Architect, an "Agent" is a precise architectural construct with defined properties of Autonomy, Iteration, and Tool-Use.

In this lesson, we define the "Minimum Viable Agent" and learn to distinguish it from a "Pre-programmed Script" or a "Standard Chat Interaction."

---

1. The Death of the "Chatbox" Mental Model

Traditional AI interaction is Static.

1. Human sends a prompt.
2. Model sends a response.
3. Interaction ends.

In this model, the Human is the "Agent." The human remembers the goal, chooses the next step, and corrects the AI's mistakes.

An Agentic System flips this. We delegate the Executive Function to the model. The model is given a goal (e.g., "Fix the bug in this file") and it must decide how to achieve it without the human steering every turn.

---

2. The Three Pillars of Agency

To call a system "Agentic" in the CCA-F, it must possess three pillars:

Pillar 1: Iterative Reasoning (The "Loop")

The system can "Talk to itself." It takes a step, looks at the result, and then decides on a new step. If a system doesn't iterate, it's just a "Zero-shot" generation.

Pillar 2: Tool Access (The "Hands")

Agency requires the ability to change the state of the world. An agent must be able to call an API, read a file, or execute code. Without tools, the agent is trapped in its own weighted averages.

Pillar 3: Goal-Orientation (The "Motive")

Standard code is Imperative: "Do X, then Y." Agentic systems are Declarative: "Make the website faster." The agent chooses the 'X' and 'Y' based on the motive you provided.

---

3. Agency is a Spectrum

You will often hear the term "Agentic Workflow." This refers to systems that aren't fully autonomous but use agentic concepts to improve reliability.

• Low Agency: A "Chain" where Agent A summarizes a transcript and Agent B turns the summary into a blog post. (No loops, low autonomy).
• High Agency: A "Researcher" that searches for a topic, finds contradictions, decides it needs to verify a specific source, and continues until it has a cohesive report. (Multiple loops, high autonomy).

---

4. The Architect's Definition

"An Agentic System is a software architecture that uses an LLM to determine the control flow of a multi-step task, typically utilizing tools and a feedback loop to reach a specified non-trivial goal."

Why this matters for the Exam:

Many scenario questions will ask: "Is an agent appropriate here?"

• If the task is Deterministic (e.g., "Add two numbers"), the answer is NO.
• If the task is Non-Deterministic and requires Judgment (e.g., "Summarize these 50 inconsistent support tickets"), the answer is YES.

---

5. Summary

Agency is the delegation of Control Flow.

• Agents iterate.
• Agents use tools.
• Agents are driven by goals, not just instructions.

In the next lesson, we will look at the "Heartbeat" of every agent: The Agent Loop.

---

Interactive Quiz

1. Compare and contrast a "Script" and an "Agent."
2. What are the three pillars of agency?
3. What is "Executive Function" in the context of an AI system?
4. True or False: Every AI application that uses an LLM is an "Agentic System."

---

Reference Video: