Earl Duque
Administrator
Administrator

Building the Promptception App – Four Episodes of Live Coding Chaos, Curiosity, and Coffee

 

Over four one-hour live streams, we took a half-baked app idea and pushed it from concept to working ServiceNow application. The “Promptception” app’s mission was simple in theory: take a short, one-sentence prompt from a user and have Now Assist’s generative AI expand it into something richer and more structured using the R.I.S.E.N. framework.

 

Key Tools & ServiceNow Features Used in the Series

 

Before diving into the episode-by-episode recap, here’s a quick rundown of the main tools and platform capabilities that shaped the build:

Development & Source Control

  • ServiceNow Studio – For creating the core app structure, tables, and portal components.

  • GitHub Integration – Linking the ServiceNow app to a GitHub repo for version control and collaboration.

  • Cursor/Claude code – An AI-powered code editor used to generate and modify XML, scripts, and widget code directly from prompts.

ServiceNow Components

  • Tables & Fields – Our usual core app structure, literally including this just to be exhaustive.

  • Service Portal Pages & Widgets – Built both manually and with AI assistance to display prompts and allow user submissions.

  • Flow Designer – Automated prompt expansion workflows and integrated Now Assist Skill Kit actions.

  • Now Assist Skill Kit – Created a custom skill to format and expand prompts using the R.I.S.E.N. framework.

AI & Prompt Engineering

  • RISEN Framework – Structured prompt format for more predictable AI outputs.

  • Prompting Techniques – No-shot, one-shot, and few-shot approaches tested for best results.

  • Debugging AI Output – Identifying and correcting AI-generated code issues while preserving best practices.

 

Of course, nothing about building in real time is ever actually simple. Here’s how each episode unfolded.

 


 

Episode 1 – The Evolution of the Show and First Lines of Code

 

 

The debut of Live Coding Elevenses wasn’t just about kicking off a new show format, it was about setting the stage for our build. We talked about the “Live Coding Multiverse” (Afternoon Tea, Second Breakfast, and Elevenses), laid out our streaming schedule, and teased a backlog of app ideas. Then we rolled up our sleeves and:

 

  • Created the ServiceNow app in a fresh repo, linked to GitHub

  • Designed the Prompts table with fields like submitter, user input, expanded prompt, and display toggle

  • Built baseline Service Portal pages for input and list display

  • Let Cursor (an AI-powered code editor) generate new fields and widget code from plain-English prompts

  • Pushed our first changes to GitHub and tested AI-generated components

It was part platform work, part AI experiment, and part coffee-fueled ramble about where the series could go.


Episode 2 – AI Can Code, But You Still Have to Debug It

 

 

Lauren McManamon joined me to push the app further and test how far AI-driven development could go before hitting the reality of bugs and best practices. Highlights included:

  • Using Cursor to generate and modify Service Portal widgets directly from prompts

  • Watching AI generate code that… didn’t quite work, forcing us into manual debugging

  • Discussing widget architecture: multiple small widgets vs. one large one

  • Walking through why some fields are still better added manually (and when AI shortcuts can cause headaches)

  • Wrestling with GitHub sync issues between Cursor and ServiceNow Studio

This was the episode where “AI builds your app for you” met “you still need to know what the code is doing.”


Episode 3 – Teaching the AI to Think Before It Talks

 

 

With Now Assist Skill Kit in play, we shifted from raw app structure to how the AI should expand prompts. We:

  • Reviewed the app’s progress and remaining gaps

  • Introduced prompt hygiene concepts like no-shot, one-shot, and few-shot prompting

  • Broke down the RISEN framework for structured AI prompts

  • Built and activated a custom Now Assist skill for prompt expansion

  • Integrated the skill into Flow Designer so expanded prompts could be stored automatically

By the end, the app wasn’t just capturing prompts, it was smart enough to format them in a consistent, AI-friendly way.


Episode 4 – The Breakfast Smoothie Bug and the Sweet Taste of Victory

 

 

What was meant to be a quick wrap-up turned into a full-blown debugging saga. Somewhere in the process, Now Assist decided that every prompt needed to be about healthy breakfast smoothies. We spent the hour:

 

  • Digging through flow logs and JSON output to trace the problem

  • Tweaking prompt formatting and even stripping Markdown to test theories

  • Building a custom action to re-format the AI’s JSON output

  • Discovering the real culprit, a wrong data pill in the Flow Designer action

  • Fixing the input, testing again, and finally seeing the app behave correctly

It was a perfect example of live development reality: hours chasing a weird bug for a one-line fix.


Four hours, one app, countless side tangents, and a lot of lessons on building with AI in ServiceNow. You can watch the full arc on YouTube to see every win, every detour, and every “why is it doing that?” moment along the way.

 

Follow us on YouTube (https://www.youtube.com/@ServiceNowDevProgram), our next app-build starts next week where I plan on building a Discord bot managed by ServiceNow!