From smart processes to truly intelligent solutions

In a world where digitization and automation are the norm, it’s becoming increasingly important to use technologies that can adapt to complex, variable situations. Agentic AI plays a key role in this. Instead of fixed processes that always follow the same steps, we now build intelligent agents that make autonomous decisions based on context and intent.

But what exactly is it, what can you do with it, and how is it different from “RPA”, “LLMs” or other forms of automation?

First real-world example: our HR Agent

We recently built our first UiPath Agent: a smart HR assistant that answers HR-related questions using company information, tools, and logic. This Agent determines on its own which information it needs, where to find it, and in what order to use the available tools to generate a response or execute the right action.

For example, it can handle questions like:

• How many additional vacation days can I purchase?
• Am I entitled to a company car?
• How much care leave can I take?

For employees, this means faster, more consistent, and more reliable answers to their questions without any HR involvement.

What makes an automation truly Agentic?

There’s a lot of buzz around AI Agents. They’re often built using natural language prompts, which makes it easy to get something up and running quickly. But not every use of an AI model automatically qualifies as an Agent.

In practice, we see many so-called Agentic Automation solutions that are essentially regular automations supported by an LLM in one or two steps. That doesn’t make them Agents.

An Agent should not be used for a fixed question with a known answer. If you already know the required steps, where the uncertainty lies, and which AI components you need, then you don’t need an Agent. In that case, you’re building a smart automation combining RPA with LLMs or other NLP models.

Take invoice processing as an example. That process follows a fixed sequence of steps:

1. Retrieve the invoice from an email inbox
2. Extract the information
3. Match it to a purchase order
4. Identify discrepancies
5. Process it in the system

When does Agentic Automation actually work well?

An Agentic Automation is different because it doesn’t follow a predefined route. It’s specifically suited for situations where:

• The input is unstructured
• There’s a lot of variation in the input
• The desired outcome and required actions differ per case
• The Agent needs to decide which tools and sources to use, and in what order

Practical examples include HR or IT questions from mailboxes or ticketing systems like Topdesk or ServiceNow. For instance:

“My laptop is broken, but I have a client presentation tomorrow. What can I do?”
An Agent can look up relevant policies, reserve a replacement laptop, and inform the user without human involvement.

How do you build such an Agent?

With the new UiPath Agent Builder (available in public preview from May 1), building these kinds of agents becomes very accessible. You combine existing automations with intelligence and define how much autonomy the Agent should have. It’s literally the collaboration between the brain (AI) and the hands (RPA).

Interested in a proof of concept?

Curious whether Agentic AI could make a difference in your organization? Get in touch with Jack Klein Schiphorst for an introduction and to learn more about the (free) PoC options we offer at Tacstone.