Agentic AI
0
By diviniadmin Artificial Intelligence Development November 18, 2025

Introduction

Large Language Models have changed the way we look at creating content. Whether it is generating code, writing a blog, or brainstorming new ideas, LLMs can really help in overcoming the creator’s block.

However, LLMs operate within fixed boundaries, like they just respond but do not act on the request made. They can process the information, but they cannot decide what to do next.

That’s why, to overcome this, we have Agentic AI. Unlike static LLMs, agentic systems can plan, execute, and adapt dynamically to achieve specific goals. They produce outputs, but they are more action-based on reasoning, context, and environmental feedback.

Limitations of LLMs

Before we move to Agentic AI, let’s understand the shortcomings of LLMs, which paved the way for Agentic AI.

  • Rely on the existing data; no access to the latest data

LLMs are trained on vast but static datasets. Once trained, they can’t access or learn from new information unless updated or fine-tuned. This means their responses may not reflect recent developments, research, or real-time changes in data.

  • Access to non-public data

LLMs cannot directly access proprietary databases, internal systems, or confidential documents. Their output is limited to the information available during training or what’s explicitly provided in the prompt. So, publicly available LLMs, like GPT, LaMDA, PALM-2, and BERT, are unsuitable for tasks that require organization-specific insights or secure data retrieval, and building an in-house LLM can be time-consuming and resource-intensive.

  • Lack of long-term memory

Traditional LLMs operate on a session-by-session basis; they don’t retain information or context across interactions. Once a conversation ends, the model “forgets” past inputs. So, LLMs cannot build persistent understanding or improve based on previous exchanges, unless there is a feature that allows you to save the new instructions in the LLM’s memory.

  • Prone to hallucinations and inaccuracies

Because LLMs predict the most probable next word rather than verifying factual accuracy, they can sometimes generate incorrect or fabricated information. These hallucinations make them unreliable for scenarios that demand factual precision, such as legal, medical, or research-based content.

  • Limited reasoning and autonomy

LLMs lack the ability to plan or make decisions beyond the immediate prompt. They cannot execute multi-step tasks, adapt strategies based on feedback, or act independently. Each task requires explicit human instruction, restricting its usefulness in dynamic or goal-driven applications.

How do LLMs and Agentic AI Approaches to a Task differ from One Another?

One important thing to understand is that LLMs produce the output that we request from them. So, for instance, you want to set up a new development environment for a Python web app using FastAPI, Docker, and PostgreSQL.

So, an LLM, like GPT-4 can generate step-by-step instructions or even write the required Dockerfile, docker-compose.yml, and setup scripts.  Once you get this, you may have to do some manual debugging and setup as per the environment.

On the other hand, Agentic AI can actively plan, test, and iterate using tools and reasoning loops. Here’s how Agentic AI could break down and finish the task when you ask it to set up:

  1. Reads your goal and creates a plan: generate the Docker setup, install dependencies, start containers, and run a health check.
  2. Uses connected tools, like a shell executor, file system access, and an environment monitor, to run commands.
  3. Diagnoses the error (if it occurs) by reading the error log, searches for the fix, updates the script, and retries, all autonomously.
  4. Documents the environment for future use.

So, an Agentic AI system could do things more end to end requiring no (or very less) manual intervention.

What is an AI Agent and Agentic AI?

An Agent is a decision making component which uses LLM to determine which actions to take and in what order – based on user input and the tools available.

Agentic AI is an advanced type of AI system that can reason, plan, able to make independent decisions, and take action towards goals, by adapting dynamically to changes in it’s environment. It is a type of AI system that combines different tools and together makes it possible to plan, act, learn, and improve.

Here are the four characteristics that describe Agentic AI:

  • Proactive: Initiate actions in addition to responses
  • Goal-oriented: Understand the goal and break it down into smaller tasks in order to complete it.
  • Adaptable: Learn from past experiences and adjust as per the feedback
  • Autonomy: Make decisions on its own to reach the goal, but within its parameters

The terms Agentic AI and AI agent are often used interchangeably, but there is a difference between them.

Agentic AI is a broader concept that refers to the next generation of AI systems designed to act with autonomy, reasoning, and goal-oriented behavior. It gives AI the ability to decide what to do next, execute the actions, and adapt as per the outcomes.

And AI agents are the implementation of this concept, where each agent performs specific tasks such as searching data, writing code, sending emails, or analyzing results, often as part of a larger multi-agent system.

Agentic AI Use Cases

Let’s understand few real life Agentic AI use cases, to get a good understanding of what Agentic AI can do for us.

1. Trip Planning

Imagine asking an Agentic AI to plan your next vacation. Instead of simply listing tourist spots like a traditional chatbot, it can autonomously execute each step of the trip planning process by interacting with multiple data sources and APIs.

Here’s how it could work:

  • It can start by checking your calendar for available dates and cross-reference public holiday data through a web search.
  • Next, it can retrieve real-time weather information via a weather API to suggest the ideal travel window.
  • It can then use APIs from travel platforms like MakeMyTrip or Expedia to compare and book flights and hotels that fit your preference and budget.
  • The system could also verify visa requirements through web searches, ensuring you’re prepared for travel documentation.
  • To personalize your experience, the Agentic AI system can search for popular attractions in your destination, check ticket availability through platforms like Headout/Viator etc., and can even buy those tickets.
  • Finally, it can compile the complete trip plan, including travel dates, accommodation, attractions, and visa information. and email it to you as a ready-to-follow travel guide.

In essence, what once required hours of manual effort or coordination across multiple apps could now be completed end-to-end by a single autonomous Agentic AI system, capable of reasoning, planning, and executing seamlessly.

2. Customer Support

In customer service, Agentic AI can function as an intelligent support agent that reads, reasons, and responds, not just replies to user’s queries. Here’s how it could work in a typical support workflow:

  • The system can begin by reading the incoming support email from a customer.
  • It can then searches the company’s internal knowledge base, such as Confluence, product documentation, or JIRA, using a Retrieval-Augmented Generation (RAG) approach to find the most accurate solution.
  • If a relevant answer is found, the AI can log the interaction in the CRM, format the response for clarity and tone, and then send the email reply to the customer, while also copying the support team for visibility.
  • If an answer isn’t found, the AI can log the interaction in the CRM record to track the query, acknowledges the client’s email politely, and escalate the issue to the Level 2 support team via email, copying the Level 1 support team to ensure visibility.

This autonomous workflow allows Agentic AI to manage entire support cycles, from query intake to escalation, with minimal human oversight, ensuring faster resolutions and consistent customer experiences.

3. Executive Assistant

An Agentic AI executive assistant can manage a mix of personal and professional tasks across different tools, without needing step-by-step instructions. For example, imagine you say:
“Send me a summary of unread emails, order Starbucks at 11 AM, book a meeting with John at 4 PM, and arrange an Uber for tonight’s trip.”

Here’s how the Agentic AI system could handle it:

  • It can scan your inbox, filters unread emails, and generates a concise summary of key messages and action points.
  • At 11 AM, it can automatically place your usual coffee order using the Starbucks API.
  • It can check your availability, schedule a 4 PM meeting with John, and then send an invite through the Google Calendar MCP server.
  • The AI could read your travel details from recent emails, determine the trip timing, and book an Uber a few hours before your departure.

In this setup, the AI plans, sequences, and times actions intelligently across multiple services, delivering a seamless, autonomous assistant experience.

How Does Agentic AI Work?

Agentic AI builds on generative AI techniques through large language models to function in dynamic environments. The generative AI models, or LLMs, focus on generating content based on the prompt. The Agentic AI has the capability to apply generative output towards specific goals. A generative AI model, like OpenAI’s ChatGPT, can suggest the best time to visit Switzerland and how to get there, but Agentic AI can execute the suggested actions, like booking a flight and hotels for your stay.

Agentic AI tools can take many forms, but here’s how agentic systems generally work:

Step 1. Perception

Agentic AI systems gather information from different sources, and this could involve analyzing text, images, or other forms of text or data.

Step 2. Reasoning

The LLM analyzes the gathered data to understand the context, identify the relevant information, and formulate potential solutions as per the goal. Like, if the goal is to schedule a meeting, it will parse the text of emails to identify the attendees, their available times, and the agenda of the meeting.

Step 3. Planning

The information is used to create a plan, which involves setting goals as per pre-defined goals or user inputs. It creates a strategy to achieve the goals using decision trees, reinforcement learning, or other planning algorithms.

Step 4. Action

AI evaluates multiple possible actions and chooses the optimal one based on different factors, like efficiency, accuracy and predicted outcomes. Using probabilistic models, utility functions, or machine learning-based reasoning, it determines the best course of action and executes it by interacting with external systems (like APIs, robots, or data) or provides responses to the users.

Step 5. Reflection

Once the action is executed, AI evaluates the outcome and gathers feedback for future development. With reinforcement learning or self-supervised learning, it learns and AI refines the strategies over time.

What’s the Difference Between Agentic AI and LLM?

If you are wondering that if Agentic AI ultimately uses LLM to plan and act autonomously, then how is Agentic AI different from LLM? Well, Agentic AI is not designed to replace LLMs, but to improve on them.

LLM remains the core reasoning engine, but Agentic AI wraps this intelligence in a framework that gives it the ability to plan, act, and adapt.

In a traditional structure, LLM only responds when it is prompted. Like, it generates text or code based on existing knowledge, but it doesn’t know what to do next or whether its output succeeded. However, Agentic AI introduces three additional layers, namely tools, memory, and feedback, to create a framework.

  • LLM: It remains at the center: interpreting prompts, analyzing content, and generating logical next steps.
  • Tools: The LLM is connected to external tools, such as web search, code execution, or data retrieval APIs, so it can perform actions.
  • Memory: Short-term and long-term memory are introduced to retain context across different interactions. Previous steps, results, and user preferences can be saved.
  • Feedback Loops: Each time the agent takes an action, it observes the outcome, compares it with the intended objective, and adjusts its next move.

The iterative cycle: Reason -> Act -> Observe -> Reflect, and Repeat, is what enables autonomy.

Model Context Protocol (MCP)

The MCP, or Model Context Protocol, developed originally by Anthropic, serves as a standardization layer for AI applications to communicate efficiently with external tools, like data, tools, and pre-defined templates. It uses HTTP + SSE (also supports STDIO) for communication and JSON-RPC for messages. Without MCP, interaction with various Tools could become brittle over time if the tool interface changes etc., MCP acts as a USB-C port for AI applications and makes the interaction with various Tools much easier. It also provides dynamic discovery of new Tools, making it very useful.

MCP was introduced to act as a bridge between the model and the external systems. The AI model can easily pull relevant data, validate facts, and perform contextual actions, without requiring manual intervention or specialized integration for each source. Just like how REST APIs provide an industry-standard way to establish communication between two applications, MCP provides a standard for the interaction between AI model and the external systems.

A2A Protocol

A2A is a communication protocol that allows one AI agent to communicate with the other one. It was initially introduced by Google in April 2025, but now it is housed by Linux Foundation as the open-source A2A project. It uses HTTP + SSE for communication and JSON-RPC for messages.

This open protocol allows for interoperability between AI agents built by different providers or using different AI agent frameworks. Usually, multi-agentic AI frameworks, like LangChain and CrewAI, automate the multi-agent workflow within their own systems. However, this protocol acts as a messaging tier that allows the agents to talk despite differences in their architecture.

Imagine having different kinds of agents like CRM agent, Hospital Management Agent, Logistics Agent and your own custom agent and you want all these different agents to communicate with each other to solve a complex business problem. These separate agents can talk to each other using A2A. You can think of A2A as a universal language that the agents use within the ecosystem. It helps in breaking down silos and boost agent interoperability.

Popular Agentic AI frameworks

There are different types of Agentic AI frameworks that you can consider using in your workflows to make the operations intelligent:

  • Crew AI

Crew AI is an open-source orchestration framework to deploy a multi-agent system. It is based on a role-based architecture, which assumes Agentic AI as a crew of workers. So, developers can introduce an agent with a backstory and the agents are assigned complex tasks. A process is identified for how all the agents can work together and how the task will be executed. It can be either sequential (like there is already a preset order) or a custom manager agent can oversee the task delegation, execution, and implementation.

  • Autogen (Microsoft)

AutoGen is an open-source framework released by Microsoft to support the development of multi-agent AI systems to perform complex tasks. The architecture is made of three layers: Core (for developing a scalable and distributed network of AI agents), AgentChat (to build conversational bots on top of Core), and Extensions (to further expand the capabilities to interface with external libraries and APIs). It also provides AutoGen Bench for benchmarking the performance and AutoGen Studio for a no-code development interface for developing AI agents.

N8N, Make, Zapier, Google Agent Development Kit (ADK) and Copilot Studio (Microsoft) are some other Agentic AI platforms that enable workflow automation, API orchestration, and AI-driven task management across enterprise systems.

Conclusion

Agentic AI marks a fundamental shift in how we design and interact with intelligent systems. Instead of acting as passive tools that wait for instructions, agentic systems can plan, reason, take initiative, and coordinate actions to achieve goals with minimal human intervention. AI agents serve as the building blocks of this ecosystem—specialized components that carry out focused tasks—while an agentic AI system orchestrates these capabilities into cohesive, adaptive, and goal-driven workflows.

As this paradigm matures, we will see AI systems that not only complete tasks but manage entire processes end-to-end, collaborate with humans and other agents, and continuously improve through feedback. Organizations that embrace this shift early will unlock dramatic gains in automation and productivity.