AI is evolving fast. This isn’t exactly a headline. But now it’s not just evolving in how it responds. It's starting to act. That’s the promise of agentic AI: not simply generating answers but identifying problems, taking initiative, and executing solutions, all without waiting for a human prompt. It’s a shift from reactive to proactive, from tool to teammate.
According to Erik Anderson, developer architect on the Innovations Team at CoreX, a good way to view agentic AI is to compare it to traditional AI, with autonomy as a key differentiator.
“If I have a traditional AI, I need to give it every single little instruction and every little piece of context for it to do its job,” said Anderson. “And then it'll always result in a single type of output that’s, ‘Here you go. Now you go do something with it.’”
Agentic AI can work autonomously. How, exactly? How? It's built on large language model (LLM) infrastructure but adds autonomy through prompt chaining, tool use, and scoped access. It’s able to use tools for itself, get information on its own, piece together what it needs to do, and ultimately either provide something for a person to use to complete a task or even perform the action that would normally require human intervention.
Anderson explained that many people have experience interacting with LLMs like ChatGPT and Google Gemini and understand how those work where you pose a question or prompt, and the LLM comes back with an output such as an image or structured text. In this context, AI is a tool that people leverage.
“It was still on you, the end user, to detect the problem and then go out and then basically leverage AI to solve that problem,” said Anderson. “But with agentic, you're basically saying, ‘Okay, I need you to look for these certain types of problems for me so I can focus on other things.’”
Agents are programmed with specific problem-solving capabilities. An agent will see an incident or a vulnerability in ServiceNow, for instance, and be able to look into resources such as knowledge bases, historical data, and configuration management database information and be able to create a larger context than a typical end user would have the time or ability to tap into.
An agent tasked to solve for specific incidents or vulnerabilities will grab the needed resources on its own and generate an output without having to be used as a traditional AI tool and explicitly asked to address individual tasks as they pop up.
CoreX Head of Innovation Jay Wigard said the evolution of AI tools over the last few years has been amazing.
“They’ve really accelerated and leveled up in terms of what they can do,” stated Wigard. “ChatGPT can write the bones of an article or a policy or an outline with minimal input and save you a lot of time. It can help you get past the ‘blank canvas intimidation’ or ‘analysis paralysis’ we sometimes face at the beginning of a task. But it’s nothing compared to the time-saving potential of agentic AI workflows working without direct input, essentially acting as additional team members.”
“Agents can do anything we program them to do,” said Rick Wright, CEO of CoreX. “An agent is a set of prompts that get executed against a large language model and that series of prompts define what the AI agent does.”
Using a call center agent as an example, Wright explained, “We’re telling the agent, ‘You're a call center agent that specializes in resolving email issues. You're highly competent.’”
This basic function defines the series of prompts that describe the activities our call center agent can do. The agent is also outfitted with a series of tools it has at its disposal to complete its prompts. Defining the role of the agent, providing it with prompts to accomplish its task, and outfitting it with the necessary tools to get there are the three basic elements of an agent, Wright reinforced.
“Think of an agent as something very specific, task-oriented and not a broad reasoning of, ‘Hey, go solve any help desk ticket,’” said Wright. “That's not what an agent will do. It's more you invoke an agent to say, ‘Well, I got a ticket that says there's an email outage.’”
Given the email outage issue as a task, our call center agent will know how to resolve the issue using the technologies at its disposal.
Anderson provided a ticket for a password reset as another example of an AI agent acting autonomously. The agent sees the password reset ticket, understands what was requested, and, if it’s configured to complete the task autonomously, triggers that password reset automatically.
In many ways, ServiceNow provides a perfect playground for AI agents, with ServiceNow as the central hub where businesses run and where problems are found and solved. Everything happening inside the platform means agents have access to data and tools to really streamline and automate a lot of business tasks.
“ServiceNow is fertile ground for agentic AI, offering centralized access to the data and tools agents need to identify and resolve issues,” said Anderson. He added that what this means for users is that they can stay put in a central place and have an AI agent assistant go out and do a lot of that grind work.
The default behavior of AI agents in ServiceNow is an agent is instructed to inform the user of whatever task it’s working on, inform the user of what it’s discovering and what it plans to do, and can be configured in the prompt to not proceed beyond a certain point without confirmation from the user. Where it seems appropriate, there can be intersection points built into the automation that act as a stoplight where the agent requires user confirmation before taking an additional step.
With agentic AI, there’s a lot of talk around autonomous behavior, but it’s important to understand that AI agents are built with a purpose, meaning you don’t turn AI agents on in ServiceNow and suddenly those agents can do anything. Each agent is configured to perform certain tasks or solve certain problems.
An example is an industrial OT incident agent. That agent is scoped to look at the industrial workspace inside ServiceNow, and when OT incidents are created, the agent is only able to look at the related industrial plant, items associated with that incident, and incident descriptions. That agent can’t autonomously decide to look at user data when that isn’t part of its assigned task.
“ServiceNow tools are strictly defined in the AI agent configuration where you define what its abilities are. Is it allowed to look at the knowledge base? And if so, what kind of knowledge base? What kind of configuration items can it look at?” said Anderson.
All of these questions are defined when configuring an AI agent. If there are concerns about what it can and can’t see, the scope available to an agent is limited or expanded by its configuration.
“It's not just like you turn it on and it'll learn on its own and just become the super user,” explained Anderson.
Where things will get interesting with agentic AI inside ServiceNow is deploying many agents with configurations that work in concert. Anderson provided a scenario where an AI change agent and a vulnerability agent work together to automate a process that potentially might lead to delays or even problems.
He described a change agent installing a new mobile element in a programmable logic controller (PLC). When the module is selected, the vulnerability agent detects an issue with the firmware needing a patch before it’s installed. The two AI agents have completely different scopes of capabilities and what they can look at.
“The vulnerability agent can look at all the CVEs [Ed. - common vulnerabilities and exposures] and all of the configuration items related to them. And then the change agent can see the same CI, but none of the vulnerability information,” said Anderson. “It's probably able to more see the dependency maps that are related to, ‘If I make this change, what is it going to affect downstream?’”
He added, “With all of your agents working together, they really can be compatible with one another and stretch across the greater ServiceNow platform while still individually being scoped to specific purposes.”
ServiceNow’s recent Yokohama release added thousands of agents across the platform, and it provided tools to make it easy to build custom agents. CoreX will be building custom agents for our customers and other vendors, said Head of Innovation Wigard.
“ServiceNow is the perfect platform for AI business transformation with all of its integration capabilities, enterprise data, workflows, and an ever-expanding set of AI capabilities,” explained Wigard. “At CoreX, we’ve been building integrations and custom apps for the better part of a decade. I love the Now platform, and as developers, we genuinely love building things that expand what you can do with the platform and finding new areas where we can genuinely improve lives.”
He added that CoreX customers operate at massive scales looking for solutions to big problems, and building agentic AI use cases is the next major step at CoreX for helping customers scale and improve workloads.
Like any new technology – especially something as disruptive as agentic AI – there are concerns and maybe confusion.
Anderson addressed concerns about embracing agentic AI in ServiceNow by explaining that agents are designed to be as visible as possible from both an admin and end user perspective. The people are part of the actions taken by agents, and maybe more importantly, users can see the benefit of agentic AI pretty quickly.
Looking at a key takeaway on agentic AI in ServiceNow, Wigard pointed out governance and change management has always been a major aspect of digital transformation.
“It’s one thing to say, ‘We want to do big things.’ But you need governance and change management to make sure everything works well and that you don’t sink the ship in the process of building toward the next big thing,” he said. “In that sense, AI is similar to all major technology changes. It is going to take planning, a careful thoughtful implementation, and governance to get the results you want. But in the end, it’s going to be worth it. And it’s going to open up huge benefits for those willing to take that journey.”