First impressions of ServiceNow Now Assist Agentic AI

Here are some things that go into making a proper Agentic AI solution on the ServiceNow platform.

1. Define the use case. This is where it all starts. You need to come up with well thought out use cases.  A use case should be repetitive, cover routine tasks, and save time to bring value once it is up and running.  Once you have a use case defined, you can start concentrating on the technical pieces of the puzzle like how many AI Agents will be needed. Another important part of the use case definition is to ask the question, “Is Agentic AI the best solution for the problem I’m trying to solve?”  This is important because there is a cost to using Agentic AI on the ServiceNow platform.  There may be alternate solutions, such as simple scripts or workflows that work just as well. To be successful, Agentic AI should be used prescriptively and wisely.
2. Define the Agent logic. At the core of the AI Agent is the definition of the logic for it to reason and carry out tasks. The AI agent needs to be instructed on what to do and how to do it so that it can act autonomously. As I found out, this takes a lot of care and thought, otherwise you will get undesired results. This makes it both a science and an art, as it takes being creative and a level of precision to get the job done. Fortunately, ServiceNow makes giving the AI agent instructions as easy as typing up what you want it to do in normal conversational text.
3. Tools, tools, tools. – This is one of the most important aspects of building your Agentic AI solution. Once you define the instructions for the AI agent, you need to give it the tools to work with. Every AI agent will need at least one tool defined. For example, if you want to look up and update a record, you will need potentially two tools, one to do the lookup and one to perform the update. Even though the AI agent determines autonomously what tool to use, careful thought and preparation are needed to make sure that the right tools are in place.  I initially thought I’d need just a couple of tools defined but wound up with five for my simple use case. On the ServiceNow platform, there are a variety of tools that can be used:
   • Catalog items
   • Conversational topic
   • Flow actions
   • Subflows
   • Record operations
   • Now Assist Skills
   • Search retrieval
   • Scripts
   • Web search
4. Understand triggers. – AI agents need to know when they are to act. This is done by adding triggers to the use case. In ServiceNow, a trigger is defined by selecting a table and setting the conditions at which it should be activated or by selecting a schedule. For instance, you can set up a trigger that launches when a record in the incident table has a status of Resolved. Once again, careful thought must be put into this, or you will have AI agents called to act when you do not intend for them to do so.

So, you can see, there is a lot that goes into a properly built Agentic AI solution. While the technology is incredibly powerful, it isn’t a magic pill that is prescribed, it is more like a workout routine. The beauty of this is that once it is in place the rewards can be immediately realized.

Final thoughts