The Now Assist Skill Kit: Prompt Engineering is Here!

The upcoming Xanadu release of ServiceNow introduces a powerful new feature—the Now Assist Skill Kit (NASK)—which allows clients of ServiceNow to create their own skills for use natively in the platform. At the heart of this capability is prompt engineering, a technique used to guide AI models in generating the desired output.

In this blog we will explore the essentials of prompt engineering, discuss the benefits of the Now Assist Skill Kit, and provide a practical guide for ServiceNow users who are new to prompt engineering.

What is the Now Assist Skill Kit?

The Now Assist Skill Kit is designed to make it easier for developers and administrators to extend and enhance the value gained from Now Assist in your ServiceNow environment.

Some of the key features of NASK include:

🔑 Custom Skill Development: You can design new skills that address specific needs within your organization. These can be embedded within the Now Assist Panel or through custom UI Actions for any record. Note: depending upon the UI in question, the Now Assist Panel may not be available.

🔑 Integration with ServiceNow Applications: These skills can interact directly with any application in the platform- it isn't limited to one specific product area.

This is the feature we've all been waiting for. We've seen clients have success with the out of box skills (which are still a phenomenal starting point) but this just brings it to the next level.

Basics of Prompt Engineering

Prompt engineering refers to the technique of crafting specific inputs, or prompts, to instruct AI models on how to generate relevant and accurate responses. Here are a few key principles for effective prompt engineering:

1. Provide a Persona

A powerful technique in prompt engineering is to assign the AI a specific persona. By doing so, you can guide the tone, style, and type of responses the AI provides, making the interaction more consistent with the intended context. For instance, if you're creating a customer service assistant, giving the AI a friendly and professional persona ensures it delivers responses that align with customer expectations. In the Now Assist Skill Kit, this is particularly useful when the assistant must take on roles such as an HR representative or IT support agent. Throughout this blog, we will refer to a use case where a Content manager needs to review knowledge submissions for reading level to ensure the full employee base can understand the article. A recent study from the Literacy Project showed 50% of Americans have a reading comprehension of 8th grade or less. For our knowledge scenario, we will target a 6th grade reading comprehension level.

Starting prompt: Review the knowledge submission and provide guidance if the reading level is not 6th grade.

✅  Improved: You are a Content Manager. Review the knowledge submission and provide guidance if the reading level is not 6th grade.

2. Clarity and Precision

The clearer and more specific your prompt, the better the AI will perform. For example, the improved prompt above may not be clear enough. Let's refine it further:

✅  Improved: You are a Content Manager who specializes in assessing the reading level of knowledge submissions before they are turned into employee facing knowledge articles. Review the text of the provided knowledge submission so that you can provide guidance to the user about which elements of the text are above 6th grade reading level and difficult to understand.

3. Use of Examples & Format

AI models can benefit from being shown examples of the desired output. This can be rather helpful if you are looking to get a specific format or type of response from the AI.

✅ Improved: You are a Content Manager who specializes in assessing the reading level of knowledge submissions before they are turned into employee facing knowledge articles. Review the text of the provided knowledge submission so that you can provide guidance to the user about which elements of the text are hard to understand.

You should provide the output in the following structure (using the provided text below as an example):

[General feedback] Some of your phrases can be verbose and challenging to understand.

[Quote from provided text] “Regaining access to your local account can be accomplished with a few simple steps, provided you have the necessary tools or security questions set up.”

Issue: The phrase "can be accomplished" and "provided" could be made more straightforward.

Suggestion: "You can regain access to your local account by following a few simple steps, as long as you have the right tools or security questions set up."

4. Chain of Thought

For more complex tasks, break them down into smaller steps. This approach, known as chain-of-thought prompting, leads to more accurate and logical results. Instead of a single, complicated request, break it into steps: first retrieve data, then analyze it, and finally summarize the results.

✅ Improved:

You are a Content Manager who specializes in assessing the reading level of knowledge submissions before they are turned into employee facing knowledge articles. Perform the following steps:

1. 1. Perform any required research to understand what is meant by 6th grade reading level to ensure you can provide adequate feedback.
   2. Review the text of the provided knowledge submission to assess its current reading level
   3. Determine the primary attributes of the text that cause it to be at a higher reading level.
   4. Determine examples for each attribute determined in the prior step.
   5. Give suggestions on how to improve each example determined in the prior step.
   6. Provide the output in the following structure (using the provided text below as an example):

[General feedback] Some of your phrases can be verbose and challenging to understand.

[Quote from provided text] “Regaining access to your local account can be accomplished with a few simple steps, provided you have the necessary tools or security questions set up.”

Issue: The phrase "can be accomplished" and "provided" could be made more straightforward.

Suggestion: "You can regain access to your local account by following a few simple steps, as long as you have the right tools or security questions set up."

5. Continuous Refinement

Prompt engineering is as much art as it is science. Through the examples above, we hope you can see that there are many ways to continually refine what you give the NASK to perform its function. Continuous refinement and governance is a key aspect of this- work towards a prompt that provides good and consistent results, then monitor and incorporate feedback from users on how to improve longer term.

Don’t let perfection be the enemy of the good- once you have reached a stable prompt and have tested it sufficiently, let your end users start getting value out of it as soon as possible!

Ready to Put it in Action?

The Now Assist Skill Kit, when paired with effective prompt engineering, offers a powerful way to enhance the ServiceNow platform. By learning good prompt design, administrators and architects can create AI-driven assistants that not only resolve incidents faster but also improve user satisfaction.

Need more help getting started with NASK or exploring Now Assist at large? Feel free to reach out for guidance!