In today’s rapidly evolving digital landscape, Artificial Intelligence (AI) is making significant strides, and it is revolutionising operations in various industries such as banking, healthcare, education, manufacturing, travel & hospitality, supply chain and many more. Also, AI is not only driving innovations in the fields of cybersecurity but also actively enhancing fraud detection activities.

There are two major types of technologies leading this AI-driven revolution. One is agentic AI, and the other is generative AI (also known as GenAI). Let’s explore them one by one.

What is agentic AI? How is it different from GenAI?

The term ‘Agentic AI’ refers to a category of artificial intelligence-based systems that can perform a broad range of tasks with little to no human supervision. Some of the most well-known applications of agentic AI are self-driving vehicles like the ones made by Tesla, autonomous rapid transit like the one rolled out by China’s state-owned company CRRC, customer service chatbots used by banks and voice assistants like Alexa.

These systems are very different from generative AI. Since GenAI tools like OpenAI’s ChatGPT, Google’s Gemini and Microsoft’s Copilot usually rely on human inputs to generate output, it is completely opposite in the case of agentic AI, which functions autonomously.

What is LLM?

LLM stands for Large Language Model, and it is a type of supervised machine learning model which is trained on massive amounts of data. LLMs rely on such data to perform tasks, which may range from generating outputs to automating business operations, as well as supporting decision-making. Both GenAI and agentic AI rely on LLMs to perform their tasks.

Let’s discuss more about agentic AI.

Agentic AI Architecture

An agentic AI-powered system consists of multiple LLMs in various interconnected layers that depend on each other to perform tasks or series of tasks, such as retrieving data from databases, making API calls, preparing and running code which may also follow custom logic, error as well as exception handling, gathering real-time inputs to support decision-making, etc.

The perception component gathers inputs from the user as well as the environment. The input can be in any form, such as data, text, images, audio, video, spreadsheets and many more, and even real-time data, such as weather data, stock market conditions, market trends, etc.

These systems follow a chain-of-thought approach to execute each step involved in accomplishing a task by thoroughly analysing the inputs being fed and output requirements. The term ‘chain-of-thought’ refers to the reasoning ability of the AI systems to break down larger and complex steps into simpler ones that are easily manageable and execute these tasks in a sequential manner. All these actions are managed by the reasoning engine.

The feedback mechanisms and learning modules, along with the memory management mechanism, help these autonomous agents learn from previous experiences and contexts and make informed decisions while performing tasks. Feedback allows the system to analyse the results and determine whether the actions it performed helped the user or organisation achieve the desired goals.

Agentic AI Architecture in a Doctorate in Business Administration

Business leaders have always dreamed of an assistant who not only understands the problem but can take the initiative — spotting opportunities, making decisions, and even executing tasks without constant hand-holding. That’s exactly the promise of Agentic AI Architecture. Unlike traditional AI models that just wait for instructions, agentic AI operates like a highly capable intern who has read every business case study ever… and never needs coffee breaks.

In business administration, this means AI agents can autonomously conduct market research, draft strategic proposals, negotiate supply terms, or simulate policy changes — all while explaining the reasoning behind each step. For DBA students, this isn’t science fiction; it’s an emerging field of serious research and real-world experimentation.

In a Doctorate in Business Administration program, Agentic AI Architecture becomes a living laboratory for strategy, leadership, and innovation. Doctoral candidates can design AI agents to run “what-if” simulations of business models, optimize resource allocation, or test crisis-management responses. The focus shifts from AI as a passive analytical tool to AI as an active participant in organizational decision-making.

The academic value is clear: DBA scholars studying agentic systems aren’t just reading about disruption — they’re orchestrating it. They explore governance frameworks, ethical constraints, and performance metrics to ensure AI agents act in alignment with corporate goals and compliance norms.

How autonomous AI agents plan, reason and execute their tasks

Let’s explore in detail how AI agents perform their tasks in various domains and fields, including business administration and information technologies.

Step 1. Analysis of Goal

The agent thoroughly analyses user inputs, constraints, intent, and the objectives that have to be accomplished, and it may also ask clarifying questions and for context.

Step 2. Planning and Information Gathering

After identifying and evaluating the input information and goals, the agent prepares a plan of action to achieve its goal. This may involve breaking down large or complex work items into smaller ones. If necessary, the system will also gather information from external sources on various aspects of the end goal, such as looking for the right tools.

Step 3. Developing Right Strategies

Each work item may have its own unique approach to reach completion depending upon the nature of the work; the agent looks for all possible means to do so, and it can look for alternatives as well if one approach or strategy doesn’t work out.

Step 4. Execution

Once all the appropriate strategies have been identified, these systems will start working accordingly and ensure that every task, including the related sub-tasks, has been completed in accordance with what was planned.

Step 5. Result and Feedback Analysis

After all tasks have been completed, the system asks for feedback from the user to evaluate whether the end goal was met or not. Depending upon the nature of the feedback, it will learn from its experience, and it will reflect on the result. This will help it to perform future tasks with maximised precision and improve over time.

Limitations

While the future for agentic AI looks promising, especially in mission-critical sectors, it may also bring some vulnerabilities with it. Let’s explore them in detail.

The training data can be biased or inadequate, which means that it can make decisions favouring a particular group or context. Organisations must ensure that the training data is free from bias or inadequacies or redundancies.

Lack of appropriate security mechanisms can expose these systems to cyber threats, potentially leading to catastrophic outcomes. This means that organisations, while investing in agentic AI infrastructure, must also invest in robust security mechanisms.

Regulatory compliance is yet another major challenge that companies have to overcome to ensure seamless adoption of agentic AI. Data protection laws in different jurisdictions vary by nature, and the ever-changing regulatory requirements further complicate the compliance process. Since AI systems, be they generative or agentic, rely on massive amounts of sensitive user data, and sectors like healthcare may use these systems to analyse patient history and recommend personalised care plans, protecting such data is of paramount importance. Forging partnerships with tech companies specialising in regulatory technologies can prove to be helpful.