Agentic AI vs. Generative AI

Few months back, I posted an article about AI Agent + LLM + RAG.

It seems too complicated. And today I want to simplify it further after I learn more about AI.

Let’s start with …

Generative AI

#

There are 2 generations of Generative AI (GenAI): LLM and RAG.

We can call the 1st generation of AI as: Rule-based LLM or simply just LLM.

Because it:

• understands NLP and provides answer just like human.
• produces output in response to static prompts.

Next, Retrieval Augmented Generation (RAG).

It is a technique that enhances LLMs by allowing them to access and incorporate external knowledge-bases.

---
title: Generative AI (GenAI)
config:
  look: handDrawn
  layout: elk
---
%%{init: {
    'theme': 'base',
    'themeVariables': {
      'clusterBkg': '#212529',
      'clusterBorder': '#fff',
      'titleColor': '#fff',
      'textColor': '#fff',
      'primaryColor': '#303030',
      'primaryTextColor': '#222',
      'secondaryColor': '#282828',
      'secondaryTextColor': '#fff',
      'lineColor': '#FFC107'
    }}}%%

flowchart TB

    subgraph RAG
        direction TB

        P2(**Prompt ?**)
        A2(**Answer =**)
        L2[***LLM*** 🦙]
        V2[(**VectorDB** 🛢️)]

        P2 -.-> A2
        P2 --> L2
        L2 --> A2
        V2 --> L2
        A2 -.-> P2

    end

    subgraph LLM
        direction TB

        P1(**Prompt ?**)
        A1(**Answer =**)
        L1[***LLM*** 🦙]

        P1 -.-> A1
        P1 --> L1
        L1 --> A1
        A1 -.-> P1

    end

    linkStyle 0,4,5,8  stroke:#8080ff,stroke-width:1px
    linkStyle 3 stroke:#ffc107,stroke-width:2px

In short, GenAI is nothing but an AI that can creates contents, such as text, image, or audio, based on pre-trained data.

AI Agent and Agentic AI

#

AI Agent is a program that takes input, thinks and acts to complete a task using tools, memory, and knowledge.

What is Agentic AI?

In fact, Agentic AI is referring to a software system that capable of autonomously making decisions and taking action to achieve a set of objectives.

This system has advanced memory functions that allow them to remember past actions and outcomes. It can learn from new experiences and feedback, continuously improving its performance over time.

Agentic AI has 3 core behaviors: Autonomy¹, Adaptability², and Proactivity³.

Automation vs. Autonomy

Automation involves performing perdefined, rule-based tasks where every step is known and predicatable. Autonomy adapts and decides based on the situation.

Below show how an Agentic AI works.

---
title: Agentic AI (3rd Gen)
config:
  look: handDrawn
  layout: elk
---
%%{init: {
    'theme': 'base',
    'themeVariables': {
      'clusterBkg': '#212529',
      'clusterBorder': '#fff',
      'titleColor': '#fff',
      'textColor': '#fff',
      'primaryColor': '#303030',
      'primaryTextColor': '#222',
      'secondaryColor': '#282828',
      'secondaryTextColor': '#fff',
      'lineColor': '#FFC107'
    }}}%%

flowchart LR

    subgraph Agentic AI

        P(**Prompt ?**)
        A(**Answer =**)
        L[***LLM*** 🦙]
        V[(**VectorDB** 🛢️)]
        T[**Tools** 🛠️]

        P -.-> A
        P --> L
        L --> A
        V --> L
        L <--> T
        A -.-> P

    end

    linkStyle 0,5 stroke:#8080ff,stroke-width:1px
    linkStyle 3,4 stroke:#ffc107,stroke-width:2px

See the additional Tools? This is the secret recipe to move from GenAI to Agentic AI.

Tools may comprise multiple AI agents and MCP servers.

Agentic AI:

• focuses on making decisions rather than on creating content.
• sets to optimize particular goals or objectives and do not rely on human prompts.
• can carry out complex sequences of activities, independently searching databases or triggering workflows to complete tasks.
• has strategic thinking and the ability to respond in real time to the environment.
• focuses on goal-driven behavior and autonomous, real-time decicsion making.

Comparison

#

Below is a simple comparisons on GenAI, AI Agent and Agentic AI.

Here’s an example.

Question: What is the price of Emirates flight from New York to London tomorrow?

GenAI won’t be able to answer this because it has a knowledge cut-off date.

On the other hand, an AI Agent with API access to Emirates will be able to take further actions besides checking the flights/prices. For instance, it can help to make the flight booking if it has been granted to access the credit card info.

Finally, if we need someone who can book a flight for my 7-day trip to London in May and the weather should be sunnny all days. My flight budget is $2000. Here comes the Agentic AI which can complete multi-step tasks and plan the goals for us:

1. Check weather to find 7 days in May
2. Compare airlines for different flights
3. Make booking that fits the budget $2000
4. Suggest for hotel and airport taxi
5. Suggest trip for local visits
6. Suggest trip for shopping plan

Key Takeways

#

First, the most significant difference between Agentic AI and GenAI is autonomy. This means it not only understand our objectives, but it will find ways to achieve them innovatively.

Agentic AI doesn’t just grasp our goals - it creatively paves the way to achieve them.

Second, GenAI focuses on creating contents and Agentic AI focuses on making decisions.

Third, RAG vs Agent/MCP:

• RAG does embedding and Agent does query via MCP
• RAG is one-way to LLM and Agentic AI can be two-way connecting to LLM.