Our Terms & Conditions | Our Privacy Policy
Proactiveness. That’s how Enver Cetin, an AI expert at global Experience Engineering firm Ciklum, described Agentic AI in a Harvard Business Review article.
Agentic AI – the next big thing in the AI ecosystem the country will see in 2025, as Inc42 called it last December – is indeed driving us into yet another phase of transformative growth. And, all this is happening barely two years after Generative AI (GenAI) changed how industries function and individuals work, and the need to improve its resolution process and customisation was felt, accelerating the natural evolution of technology.
AI companies are actively building autonomous AI agents to help their enterprise and startup clients improve customer satisfaction. Many startups are also building in-house AI agents to help enhance the processing of ticket requests, reduce the need for human agents, and so on. From Gupshup to Zomato, Infosys to Zoho Corporation – Agentic AI has become a major focus for companies no matter how big or small they are or in which sector they operate in.
The rapid waves triggered by Agentic AI have fostered a $5.1 Bn market around the world, which is on course to surpass $47 Bn by 2030, averaging more than 44% growth rate. Reports suggest India has a high AI adoption rate, with some stating it accounts for around 40% of global AI agent deployments.
But, what is Agentic AI and how is it making companies smarter? We often see ‘AI agents’ and ‘Agentic AI’ being used interchangeably. Are they really the same?
Let’s understand the concept first.
What Is Agentic AI?
To put it simply, Agentic AI refers to autonomous GenAI agents that are capable of doing complex tasks with little or no human supervision.
These autonomous GenAI agents are more sophisticated in their ability to take actions, and hence, solve more complex tasks compared to what AI assistants could do. The earlier avatar could solve specific tasks such as writing emails or providing customer support, but the advanced version can handle complex tasks such as planning your next vacation, making travel arrangements, taking care of the elderly with human-like bots, and resolving supply chain issues by smarter management of inventories.
The foundation and the backbone remain the same where the open-source or closed-source large-language models (LLMs) are fine-tuned, but they are enhanced with more context and intelligence to analyse and resolve issues independently.
Krishna Tammana, chief technology officer at Gupshup, said that the power of language models and their ability to reason and give answers to questions is already established. “GenAI has advanced quickly over the last two years, bringing in a dramatic improvement in the quality and performance of all these models, including the foundation models,” he told Inc42.
But answering questions by using its reasoning power, logical inputs, and knowledge and information base isn’t enough, he said. There’s a desire to go beyond this and take action on behalf of humans. Such actions could vary from ‘pay my bill’ and ‘buy me a ticket’ to ‘disable my credit card’ and ‘book a doctor’s appointment’.
GenAI-based assistants could address such requests earlier as well, but their actions were limited, which eventually called for human intervention to address customer-specific queries. Agentic AI is making clear transformations there with its autonomous and intuitive nature.
“Once you have a piece of technology that can answer a question, the next thing you can ask the technology to do is, perform an action. There have been a few efforts earlier. For example, there was an effort to create something called ‘large action models’. The idea was the same. Then came agentic models. Why is the name? Because they behave like human agents to perform tasks and actions that you would want them to perform. I see it as a natural extension. Think of it as something on top of LLMs, or as an extension, or the next step where the LLMs function,” Tammana said.
Although Agentic AI and AI agents are often used to mean the same thing, there are subtle differences. These terms are, however, used more loosely and vary in how each company defines them.
span {
margin: 0;
padding: 3px 8px !important;
font-size: 10px !important;
line-height: 20px !important;
border-radius: 4px !important;
font-weight: 400 !important;
font-style: normal;
font-family: noto sans, sans-serif;
color: #fff;
letter-spacing: 0 !important;
}
.code-block.code-block-55 .tagged {
margin: -4px 0 1px;
padding: 0;
line-height: normal;
}
@media only screen and (max-width: 767px){
.code-block.code-block-55 {
padding:20px 10px;
}
.code-block.code-block-55 .recomended-title {
font-size: 16px;
line-height: 20px;
margin-bottom: 10px;
}
.code-block.code-block-55 .card-content {
padding: 10px !important;
}
.code-block.code-block-55 {
border-radius: 12px;
padding-bottom: 0;
}
.large-4.medium-4.small-6.column {
padding: 3px;
}
.code-block.code-block-55 .card-wrapper.common-card figure img {
width: 100%;
min-height: 120px !important;
max-height: 120px !important;
object-fit: cover;
}
.code-block.code-block-55 .card-wrapper .taxonomy-wrap .post-category {
padding: 0px 5px !important;
font-size: 8px !important;
height: auto !important;
line-height:15px;
}
.single .code-block.code-block-55 .entry-title.recommended-block-head a {
font-size: 10px !important;
line-height: 12px !important;
}
.code-block.code-block-55 .card-wrapper.common-card .meta-wrapper .meta .author a, .card-wrapper.common-card .meta-wrapper span {
font-size: 8px;
}
.code-block.code-block-55 .row.recomended-slider {
overflow-x: auto;
flex-wrap: nowrap;
padding-bottom: 20px
}
.code-block.code-block-55 .type-post .card-wrapper .card-content .entry-title.recommended-block-head {
line-height: 14px !important;
margin: 5px 0 10px !important;
}
.code-block.code-block-55 .card-wrapper.common-card .meta-wrapper span {
font-size: 6px;
margin: 0;
}
.code-block.code-block-55 .large-4.medium-4.small-6.column {
max-width: 48%;
}
.code-block.code-block-55 .sponsor-tag-v2>span {
padding: 2px 5px !important;
font-size: 8px !important;
font-weight: 400;
border-radius: 4px;
font-weight: 400;
font-style: normal;
font-family: noto sans, sans-serif;
color: #fff;
letter-spacing: 0;
height: auto !important;
}
.code-block.code-block-55 .tagged {
margin: 0 0 -4px;
line-height: 22px;
padding: 0;
}
.code-block.code-block-55 a.sponsor-tag-v2 {
margin: 0;
}
}
]]>
Agentic AI Versus AI Agents
Agentic AI is the broader concept of solving issues with limited supervision, while an AI agent is a specific component within that system that is designed to handle tasks and processes with a degree of autonomy. That’s how IBM describes the two levels of artificial intelligence.
The Big Blue sees Agentic AI as the technology that can manage and run the overall energy consumption system in a smart home. It uses real-time data and user preferences to coordinate individual AI agents such as the smart thermostat or any other appliances. The agents have individual goals and tasks and they work together within the agentic AI framework to achieve the homeowner’s energy goals.
While the definition might look simple, an Inc42 research gathered that the industry is yet to come to a solid conclusion on their true use.
Anthropic, which focuses on developing GenAI systems and LLMs, wrote in a recent blog post that some customers define ‘agents’ as ‘fully autonomous systems that operate independently over extended periods, using various tools to accomplish complex tasks’. Others use the term to represent more ‘prescriptive implementations that follow predefined workflows’.
“At Anthropic, we categorise all these variations as agentic systems, but draw an important architectural distinction between workflows and agents,” said the post.
“We can think of ‘agents’ as those encompassing a few elements. There are LLMs that give us the natural language processing (NLP) capabilities. There is memory associated with these agents and data stores that provide the context. Besides, there are tools such as APIs and workflow automation and AI capabilities like summarisation, anomaly detection, and forecasting. And, by Agentic AI, we mean these capabilities of the AI agents are put together,” Sujatha Iyer, who heads AI security at ManageEngine by Zoho Corp, said.
India In Agentic AI Age
The biggest use case for Agentic AI in India is the conversation bot. Whether it’s in text-to-speech or speech-to-speech, companies building AI agents are trying to offer Agentic AI solutions tailored to the needs of their customers.
Conversational AI platform Gupshup recently launched its library of 15 pre-built and customisable AI agents. It is providing its Agentic AI capabilities to startups like Lenskart and Cars24, as well as to companies like Kotak and Tata AIA in the financial sector.
Gupshup’s Tammana explained that in the past, the company could offer very rich conversations to its customers that were powered by LLMs, but when it came to taking actions, it had to hand code them.
“For many banks, we had implementations where we hand-coded actions when customers asked ‘show me the balance’ or ‘disable my credit card’. Now, with Agentic AI, we not only do those things faster but also have more sophisticated reasoning and logic to make the decisions for each customer, make recommendations and take the customer through the journey of actually performing that action,” he said.
While earlier it used to take four to six weeks for its clients to resolve some complex issues, now it takes only three to four days, according to him.
With AI agents making deeper inroads into Corporate India, Zoho Corp came up with an AI-powered platform, Zia Agents, to bring autonomous agents into business workflows. It also introduced Zia Agent Studio, a no-code or low-code platform that helps businesses create autonomous agents designed to their needs.
Zomato CEO Deepinder Goyal took to social media this week to announce the launch of Nugget, an AI-native, no-code customer support platform. Built with the capabilities of AI agents, the platform claims to resolve up to 80% of queries autonomously.
Infosys recently told Inc42 that the company is building over 100 GenAI agents for client applications in collaboration with its AI partner ecosystem.
While Hyderabad-based startup Pulse raised $1.4 Mn in seed funding last November to build its Agentic AI platform aimed at SaaS product teams, Atomicwork raised $25 Mn in January as it aims to transform IT service management with its Agentic AI platform.
Speaking to Inc42, Pranav Patil, chief data scientist at AdvaRisk, a startup that helps banks with end-to-end collateral management through its GenAI-powered data intelligence platform, shared that early adoption of Agentic AI will largely be on the customer support front because risks are lesser.
The adoption in algorithmic trading and in improving loan disbursement and loan processing is taking place slowly in the BFSI sector, he said.
Will Agentic AI Have Hallucinations In Check?
Sometimes an LLM, which could be a generative AI chatbot or computer vision tool, perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate. Such fallouts are called AI hallucinations, said an IBM blog post.
The problem of hallucinations surfaces even in Agentic AI, calling for the need to put up guardrails around the models.
Tammana, however, sounded confident about the high degree of sophistication in an autonomous agent. “Earlier, we had to encode every rule, so what we could build before was rule-based systems, as opposed to intelligent thinking systems now.”
ManageEngine’s Iyer said that there has to be a lot of output engineering to reduce hallucinations, and this is where Retrieval Augmented Generation (RAG) system and reinforcement learning can help.
In this rapidly evolving age of artificial intelligence, it now remains to be seen how Agentic AI writes the next major story of customer success in enterprise use cases or if it can go far beyond its current capabilities to improve domains in more business tasks.
[Edited By Kumar Chatterjee]
Images are for reference only.Images and contents gathered automatic from google or 3rd party sources.All rights on the images and contents are with their legal original owners.
Aggregated From –
Comments are closed.