Using AI Agent to Extract Key Topics from Incidents

[Alisa Tipisova]

In the previous post we built a custom Now Assist skill to identify key topics in incident records and store them for reporting. It worked well—but as we explore GenAI capabilities further, another question comes up:

Can we do the same thing using AI Agent instead of Now Assist?

Let’s find out.

What We Want to Achieve

Just like before, our goal is to:

• Analyze incident Short Description and Description fields

• Identify the key topics users are reporting

• Store those topics in a custom table for reporting and trend analysis

This time, instead of relying on a custom skill triggered from Flow, we’ll use an autonomous AI Agent that runs behind the scenes.

Creating a Custom AI Agent

Head over to AI Agent Studio and click Create and Manage to build a new agent.

Give it a name, write a short description, and configure the two main building blocks:

• Role – What is this agent? For example: “An expert at extracting key topics from incidents using natural language analysis.”

• Instructions – What should the agent do, step by step?

Defining the Agent’s Tasks

In this case, the agent needs to:

1. Retrieve the incident details (Short Description and Description)

2. Get the list of existing keywords from a custom table

3. Analyze the incident text and match it with existing keywords

4. Insert new records into a keyword mapping table

We’ll handle all this by adding tools the agent can use.

Adding the Tools

Go to the Add Tools and Information section. Here’s what we need:

1. Script Tool – Fetch Incident
   Retrieves the short description and description based on the incident sys_id

   

2. Script Tool – Get Existing Keywords
   Queries the custom table for all known keywords (this is the same script I used in my previous post)

   
3. Record Operations Tool – Insert Keywords

   Inserts a new record into the keyword mapping table with two inputs:

   • Keyword

   • Incident sys_id

Once that’s done, the agent has everything it needs to complete the task from end to end.

Please note: I don't tell HOW to identify the keywords in the incident details, this is where AI magic happens.

Triggering the Agent

AI Agents support multiple trigger types:

• Record created or updated

• Scheduled execution

• Email received

For this use case, we want the agent to run automatically when a new incident is created. So we configure a record-based trigger accordingly.

Activating and Testing

The agent doesn’t need to be visible in the UI—it runs completely in the background. Just click Activate.

Then head over to the Test screen to try it out. Provide the incident number and click Start Test.

Here’s what happens:

• The agent pulls the incident description

• It fetches existing keywords

• It compares them and selects relevant matches

• It inserts the final keywords into your custom table

No workflows, no scripting logic outside the agent setup—it just works.

Now Assist vs. AI Agent — Which One Should You Use?

Both approaches solve the same problem, but they offer different strengths:

Feature	Now Assist Skill	AI Agent
Setup Location	Skill Kit Console	AI Agent Studio
Trigger	Flow Action or Manual	Record change, email, or schedule
LLM Prompt Customization	Full control with examples and formatting	High-level instructions; more abstract
Data Fetching	Requires scripting via Tools	Requires scripting via Tools
Automation Integration	Needs Flow to execute and store results	Fully autonomous once activated (can be supervised as well)
Best For	Use cases that need prompt precision + control	End-to-end tasks that should run independently

Reusing the Now Assist Skill Inside an AI Agent?

Since I had already built a custom Now Assist skill to retrieve existing keywords and identify key topics, I wondered—can I reuse that work inside the AI Agent?

Turns out, yes.

I added a new Now Assist Skill Tool to the agent and simply selected my existing custom skill.

Because the skill already handles keyword retrieval, I no longer needed a separate script tool for that. So I removed the Get Existing Keywords tool.

Next, I updated the AI Agent’s instructions to reflect the simplified logic—removing anything related to fetching keywords.

 And that’s it. I let the agent run—and as you can see, it works just as well with the Now Assist skill embedded directly into the AI Agent!

Final Thoughts

Both Now Assist and AI Agent can help you extract valuable insights from incident records—but they shine in different ways.

With Now Assist, you get full control over prompt design and output formatting. It’s a great choice when you want to guide the GenAI logic closely and integrate it into Flows or UIs.

AI Agent, on the other hand, excels at orchestrating tasks independently. It can read data, call logic, and write results—without a single Flow.

And the best part? You don’t have to choose one or the other.
In my setup, I reused the Now Assist Skill inside the AI Agent to simplify logic and avoid duplication. The AI Agent focuses on orchestration, while the Now Assist Skill handles the natural language analysis.

It’s a great example of how the GenAI building blocks in ServiceNow can work together—efficiently and flexibly.

If your business has a use case in mind, chances are:
you can build it.

[anubhavkapoor76]

@Alisa Tipisova - Nice write-up 👏. Providing complete screenshots for scripts would have been beneficial for the community readers.