Agentic AI: The next wave of intelligent process automation

Automation is helping businesses become more efficient by allowing them to focus their human workforce on the highest value-creating tasks and opportunities.
In the recent past, Robotic Process Automation (RPA) has emerged as one of the key technology drivers of automation. Further, the advent of powerful Large Language Models (LLMs) like ChatGPT, Gemini or Mistral introduced Artificial Intelligence (AI) to the B2B technology stack.
But there is a new revolution brewing. In the next phase, agentic AI is set to drive innovation in an even more intelligent and powerful way. The research firm Gartner predicts that by 2028, 15% or more of day-to-day work decisions will be made autonomously through agentic AI. This is up from 0% in 2024.
In this article, let’s explore what agentic AI is, how it can help enterprises with further automation, and what you can do to adopt this cutting-edge technology.

The evolution of intelligent process automation

Emerging in the early 2000s, RPA was the first step towards intelligent process automation. Leveraging software bots, traditional RPA is built on structured data from centralized systems of record like a CRM or an ERP to automate simple and repetitive workflows or business tasks like payroll processing or regression testing for software engineering — all this, in a rule-based and deterministic way.
While gaining traction in the 2010s, rule-based robots were progressively complemented with AI/ML to add more flexibility in handling processes. In this phase, AI transformed decision-making into a more dynamic activity. This significant progress has been dramatically accelerated in the last two years with advances in Generative AI (GenAI) and, notably, by the development of Large Language Models (LLMs).
This second phase gave way to modern, goal-based AI agents, whose underlying models handle vast amounts of structured and unstructured data. In terms of human-machine collaboration, chatbots now use GenAI to provide responses based on single interactions, or prompts. Next, they use natural language processing (NLP) to provide answers. Some examples of these are SAP’s Conversational AI, Microsoft’s GitHub Copilot or Atos’ very own MO4D assistant — our powerful Machine One for Delivery GenAI-powered developer productivity suite.

The next wave: Agentic AI

Over the last years, these progresses have strongly improved automation capabilities for multiple business processes. But there is a next and third phase coming: agentic AI, in which autonomous systems of agents are self-orchestrated, make decisions and take actions to achieve predefined goals without (or with minimal) human intervention.
The highlight of this third phase is that agentic AI systems don’t need explicit inputs and prompts, and they don’t produce predetermined results! They can learn, adapt, and interact dynamically with their environment, often integrating reasoning, planning, and self-improvement capabilities.
Compared to RPA and chatbots, agentic AI has four unique traits. It is —

• Perceptive, gathering and processing data from multiple resources. These resources can be traditional databases like CRM or ERP but can also include ambient information from Internet of Things (IoT) sensors or digital interfaces.
• Autonomous, making decisions on its own, acting, planning, deciding, based on sophisticated reasoning engines.
• Adaptable, solving problems dynamically within changing environments, learning, collaborating and iterating based on feedback loops.
• Goal-oriented, pursuing and achieving user-defined business outcomes.
  This combination brings positive changes and opportunities to the business.

Profound change, tremendous growth possibilities with agentic AI

These may be in customer service, content creation, hiring, business analytics, medical diagnostics, predictive maintenance, asset management, software engineering, and many more.
For example, an autonomous AI customer agent operates beyond basic question-and-answer mechanisms. As part of the interaction, it checks a user’s outstanding balance and recommends accounts to pay it off. This happens in the background, while the customer asks questions and makes decisions. Transactions can be executed accordingly when prompted. This enhances the customer experience, makes interactions more relevant and swifter, and at the same time decreases financial risk and operational costs for the company.