In the ever-evolving landscape of technology, one thing remains constant: hype. If you’re still catching your breath from the excitement surrounding generative AI (GenAI), brace yourself, because the future waits for no one. While GenAI continues to make waves, a new contender is quietly stepping onto the stage—Agentic AI—or, as many prefer, agent-based AI (because “agentic” sounds a bit peculiar, doesn’t it?). Regardless of the name, this next phase in AI promises to bring both hope and a host of challenging questions. The biggest one? Should you care about Agentic AI?
Let’s take a quick trip down memory lane. Just a decade ago, robotic process automation (RPA) was the talk of the town. The idea was simple: Automate repetitive, high-volume, low-exception tasks. RPA successfully reduced the need for manual data entry and other mundane operations. However, there was a catch. Any time an exception arose—a unique scenario that wasn’t part of the preprogrammed rules— RPA hit a wall. That’s when human intervention vention became necessary.
Now, a new parallel is emerging: agentic process automation (APA). APA isn’t designed to replace RPA but to complement it, taking automation to a more sophisticated level by managing complex, decision-heavy processes. Backed by a new generation of AI models known as large action models (LAMs), Agentic AI has the potential to automate tasks in ways we never thought possible.
LAMs are the muscle behind Agentic AI, but to understand them, we need to look at how AI has historically dealt with tasks that require decision making. Traditional RPA systems handle tasks based on strict, rule-based logic. When something goes wrong, such as an outlier scenario, it requires a person to step in and decide what to do next.
Beyond basic automation
Agentic AI, however, is designed to go beyond basic automation. It uses LAMs to analyze a situation, determine a range of possible actions, and decide on the most appropriate “next best action” to resolve exceptions. Much like the way large language models (LLMs) predict the next word in a sentence, LAMs guess the next action needed to resolve a task.
If that sounds abstract, think of a customer service chat. A typical RPA might automate simple tasks such as checking an account balance. But if the customer says, “I can’t access my account because my password isn’t working,” Agentic AI can take over. It would evaluate whether the password needs resetting, update the customer’s login data, log the issue in the company’s helpdesk software, and even send an email confirmation. Agentic AI is trying to “think” more like a human, working within the context of a broader task while resolving specific issues as they arise.
The key difference between RPA and APA lies in intent. RPA focuses on executing straightforward tasks, while APA is concerned with understanding why a task is being done. This shift toward intent-based automation allows Agentic AI to manage entire processes rather than isolated tasks.
Let’s return to our example of a help-desk. In an RP A scenario, the automation might simply reset the password. In an APA scenario, Agentic AI takes a more holistic approach. First, it assesses the intent behind the request. Is the password reset part of a larger issue, such as a system-wide security breach? Could multiple processes need to be triggered simultaneously? Agentic AI can then select the right processes to run, monitor their progress, and decide whether to escalate the issue or move on to the next step.