Agentic AI is reshaping the future of business automation by enabling intelligent agents that reason, plan, orchestrate, collaborate, and autonomously adapt in real-time. Unlike traditional automation, Agentic AI dynamically creates solutions based on live context, driving substantial business impact.
The measurable benefits are compelling. ServiceNow, an early adopter of Agentic AI, reports impressive outcomes: AI agents currently generate $235M in annualized value, manage 400,000 workflows each year, and achieve an 84% rate of customer self-service.
This shift represents more than just an incremental upgrade; it's a revolutionary change in how organizations streamline workflows and engage with users. As Aaron Munoz, Solution Architect and Trainer at CoreX, puts it, "There is a huge revolution on how we're going to have AI helping people, not replacing them, in case anyone wonders what the future may bring."
To fully leverage Agentic AI, businesses must understand how to integrate AI effectively into their existing workflows, recognizing the nuances of successful implementation and the profound advantages it can unlock.
Munoz draws a clear distinction between traditional automation (automation 1.0) and the next evolution (automation 2.0) powered by Agentic AI. In traditional automation, organizations create virtual agents with predetermined workflows; every possible path must be mapped and coded in advance.
For example, a customer service chatbot might have scripted responses for dozens of scenarios: password resets, account inquiries, billing questions, and so on. But what happens when a customer has a unique issue that doesn't fit the script? The system fails, and the interaction gets escalated to a human agent.
Agentic AI changes this dynamic entirely. Instead of following pre-written scripts, AI agents intelligently create workflows during runtime. They reason through problems and generate solutions dynamically based on user inputs and business context, much like a knowledgeable human employee would.
However, this increased autonomy raises important governance questions that organizations must address:
These decisions shape not only operational efficiency but also risk management and compliance strategies.
When thinking about Agentic AI and automation, it’s useful to understand the three “flavors” of human involvement versus autonomy in automated systems or processes.
“The second level, human on the loop, is whenever we provide the AI system with some autonomy, but there is still a stop button,” says Munoz. “We can decide not to go forward with whatever it’s deciding to do, especially for critical systems or something that involves, say, the healthcare or well-being of people, we still want to have a break point.”
Understanding these levels of human involvement helps inform how Agentic AI systems should be designed and deployed. Let's examine how these concepts translate into actual system architecture and operation.
The first step of building an Agentic AI use case is determining the overall business problem or goal to be solved. This outcome is why the agent is being deployed.
Once deployed, the Agentic use case is triggered, either by a user or the system. This trigger leads to the AI Agent Orchestrator in ServiceNow, which handles planning and leading a team of AI agents to address the use case.
This team of agents serves as virtual workers performing tasks to resolve the use case by leveraging platform tools like workflows, skills, knowledge bases, and scripts to complete tasks.
Each AI agent follows a basic process:
An important consideration when building agent teams is maintaining simplicity in individual agent design. As Munoz notes, "It's much easier to separate, divide and conquer the solution to much easier pieces, and then you can update the individual AI agent for one task without affecting what the others can do."
This modular approach enables easier maintenance, clearer troubleshooting, and more flexible system evolution over time.
An obvious question with Agentic AI use cases is, “Where can I find opportunities for AI agents?” Munoz says a good starting point is non-deterministic opportunities, those places where the outcome isn’t known in advance that involve dynamic decision-making and dynamic conversational workflows.
Another great source is simply listening to users because they know the limitations of deterministic workflows. Therefore, it only makes sense to ask them how and why they make the decisions that they do. And finally, take cues from the platform by mining processes for insights on shortcomings that can be addressed by an Agentic AI use case.
Right now, ServiceNow already has out-of-the-box agents ready to go addressing situations across the organization, such as agent-based incident resolution in IT, tuition reimbursement in HR, case resolution and triage in CRM, and service test and repair in TMT.
And the ServiceNow AI Agents Studio allows for creating custom agents using pre-built skills, flows, playbooks, knowledge base articles, SharePoint data, and more.
To illustrate how these principles work in practice, let's examine a detailed example of an agentic AI team handling warranty claims at a manufacturing company.
Diving into an example of an Agentic AI use case, let’s look at warranty claims at a manufacturing plant.
The manufacturer has a warranty claim portal where users can request warranty claims and view claims, and the company has a warranty claim workspace on the administrative side to work on claims.
In this use case, a group of AI agents helps accelerate the process of determining if the request is, in fact, a warranty claim, finding out if the product in question is actually under warranty, and then evaluating the ticket to see if the AI agent can automatically approve the claim.
Another agent tracks all warranty claims to determine if there are multiple claims around the same issue, which would indicate a problem and escalate that discovery to the product team.
The first stage in the agent team is a warranty verifier, which can take text from the claim ticket and compare that information to a knowledge base on what is covered under warranty. These knowledge articles provide the agent with guidelines and logic around the claim. The agent also learns from previous decisions approving similar claims.
If the claim is approved at this stage, the customer is automatically sent an email letting them know the defective product will be replaced by a separate agent tasked with handling automated email communications.
A third agent is looking at the entirety of warranty claims and, in the case of repeating issues, raises a problem ticket based on meeting a certain threshold that is escalated to the product team.
This basic example illustrates how agents take over some of the hands-on tasks involved in resolving warranty claims. The Agentic team can be expanded to handle additional tasks or steps as needed, and the level of automation can be adjusted over time.
When first deployed, the warranty claim use case might have a “human in the loop” approach requiring approvals at each stage, and as users become more comfortable with the agentic flow, outcomes can become more fully automated.
“We always want to emphasize the importance of piloting and minimizing variables as you start an agentic loop step,” says Jay Wigard, Head of Innovation, CoreX. “That ‘human in the loop’ step is an essential starting point, and then as you see the tools working and everything is acting as we would hope then you can start to add in automated triggers.”
The best approach to Agentic AI use cases is to start small, scale quickly, and test and iterate.
“It’s better, just like in other areas of ServiceNow, to have a phased approach,” says Munoz. “Start small with something that is high value and simple to develop instead of waiting for months and not getting a quick win and a quick return on investment.”
He adds, “This can be something very simple, but also beneficial and powerful. Another thing is to test and iterate. It’s not that we have to be perfect and get it right the first time. The benefits of an agile approach applies to AI capabilities like Now Assist and Agentic AI.”