Every company, big tech or startups alike, is betting on agentic AI becoming the biggest trend in the near future. As we move away from small language models, everyone will be likely to increasingly talk about implementing AI agents in their workflow.
Recently, NVIDIA CEO Jensen Huang suggested that all IT departments would evolve into HR for AI agents. Microsoft CEO Satya Nadella similarly highlighted that there would be a swarm of AI agents in the workforce, likening the shift to the rise of Robotic Process Automation (RPA) in recent years. However, it didn’t turn out exactly as predicted.
It was predicted that RPA would automate most of the mundane jobs, which would allow teams to focus on larger tasks. Several industry experts now predict that AI agents are going through the same phase as RPA.
Nikhil Malhotra, chief innovation officer at Tech Mahindra, on an episode of What’s the Point with AIM, pointed out that while a lot of startups would be talking about agentic AI this year, most of the tech would just be RPA. “But the good thing about this would be that these startups will start thinking about agentic loops.”
For instance, when Anthropic released its computer use feature with Claude 3.5 Sonnet, it could move the cursor, click buttons, and type text, as well as fill out forms, navigate websites, and interact with software programmes. This agentic approach has left many wondering about the potential implications for RPA companies and the future of agentic AI and whether it will meet the same fate.
“Wasn’t RPA the exact same thing without an LLM and it failed miserably”
Given the hype around AI agents, the question of their capabilities in the workforce needs to be examined deeply. The current frameworks seem very similar to RPA but with an LLM in the loop. Though that makes a huge difference, adapting them to the workflow still looks like adopting a 10-year-old technology.
It is predicted that the $250 billion SaaS market will be replaced by the $300 billion AI agents market as companies adopt AI agents in their workflows. However, given the huge price difference, people are still not convinced if moving away from the current systems to AI agentic ones is worth it.
Moreover, all RPA companies are also entering the AI agent race. Apart from Salesforce, companies like UiPath and Automation Anywhere have started leveraging AI agents because they believe that both offerings are different. This means that RPA is now actually being upgraded to agentic AI, and not much has changed.
While speaking with AIM, Param Kahlon, EVP and GM of automation and integration at Salesforce, earlier said that autonomous agents also do not mean the end of RPA technology.
“RPA agents were designed to automate repetitive, tedious tasks, such as transferring data between systems when APIs aren’t involved. In contrast, autonomous agents process information more like humans, adapting to situations and making decisions based on changing conditions, enhancing efficiency and effectiveness in workflows.”
Ramprakash Ramamoorthy, director of AI research at ManageEngine and Zoho, told AIM that the dispersion around agentic AI systems in enterprise IT is becoming increasingly polarised. He said that for enterprises, the shift from RPA to agentic AI opens a new era of self-directed operations, which enables faster scaling and better responses to evolving business needs.
“Agentic AI is more than just RPA with LLMs; it’s a transformative evolution that combines automation with intelligent decision-making. While traditional RPA executes predefined tasks, agentic AI learns, reasons, and adapts in real-time, elevating process automation with cognitive flexibility,” Ramamoorthy said.
Agentic AI is RPA 2.0
“Majority of agentic applications are basically workflow automation with some minimal amount of people interactions,” tech YouTuber Shailesh, who runs channel SV Techie, wrote on X. He explained that there might be a reduction in headcount amongst companies, but nothing like the autonomous hype that is being sold.
Anil Kumar, CTO at Exotel, told AIM that calling agentic AI just RPA with LLMs is as unconvincing as saying C++ classes are just C structures with methods. While RPA deals with structured data, agentic AI aims to achieve automation by using LLMs to interpret decision trees.
Taking the example of complex human conversations such as loan negotiations, Kumar said that they cannot be expressed as a decision tree. “Agentic AI like the one used in our bots will work backwards from the goal given to them (which is to negotiate and disburse loans) and navigate the nuances of human conversation,” Kumar said.
“They achieve this by carrying the context of the current conversation, learnings from previous conversations, information from a knowledge base, contractual constraints from a legal document, etc. and make decisions towards achieving the given objective.” He added that if this is implemented as RPA along with LLM, it will be like “giving a script to a child actor who will fluster on stage if others go off script”.
Between 2018 and 2023, AI integration into RPA solutions has steadily evolved. This has enhanced RPA’s functionality with sophisticated AI capabilities. The true breakthrough, however, brought about the emergence of agentic AI in 2024.
Andreessen Horowitz, in its thesis posted in November last year, pointed out that AI will automate operations and eat the world of RPA. The end of traditional RPA is widely discussed in the industry.
Deepak Dastrala, CTO and partner at IntellectAI, told AIM that RPA focuses on automating repetitive, rule-based tasks, which makes it a tactical solution. AI agents, on the other hand, take a more goal-based approach and act more like digital twins of humans, powered by LLMs, equipped with memory, and able to adapt and learn in real time.
“That’s why RPA’s relevance has faded, while AI agents are poised to reshape our work at a level no other automation technology has achieved,” he said.
“The era of cargo cult programming to churn out generic, modular software is dead and buried. In 10 years, RPA and agent studios will be relics of the past. Instead, we’ll see specialised agents, each uniquely designed for specific industries to solve problems end-to-end,” said Arnav Bathla, CEO of Layerup.
Agentic AI can be viewed as RPA 2.0. The rebranding of advanced RPA as agentic AI is often a marketing move to capitalise on AI’s hype. Vendors position their products as “intelligent agents” to differentiate them from traditional RPA, despite the underlying functionality being a continuation of process automation.
The fundamental objective – automating repetitive tasks for efficiency – remains unchanged. However, the fate of agentic AI might end up the same as that of RPA if it fails to evolve and address its limitations.
