Regal Copilot

Regal's Big Bet

Our new mantra: Regal builds; customers guide.

Managing and optimizing Regal, a full suite CCaaS platform with omnichannel AI agents, requires ongoing management, monitoring, and iteration cycles from customers and Regal's internal teams. Regal's new vision of AI building AI aims to fundamentally change that so primary workflows would instead be handled by an AI-driven Copilot feature that works alongside users.

The goal of Copilot is to shorten time-to-value and scale quality and performance across every use case with minimal manual intervention.

We have the MVP — Now What?

The Problem

A couple weeks after the MVP launched, the data looked promising:

The MVP validated demand; however; adoption and intent resolution remained below target:

With adoption and intent resolution rates far from target goals, what would be the next highest impact investment to close that gap?

After presenting ideas informed by competitive analysis at the kickoff of this project, the CTO, PM, and I aligned on a broader vision for Copilot: an agentic layer that could both collaborate with users in a dedicated workspace and surface throughout the platform contextually. Two directions emerged:

01

Workbench

A dedicated environment where users and Copilot could build, test, and iterate together.

02

Widget

An embedded experience that could surface throughout Regal to provide contextual assistance inside existing workflows.

Due to resource constraints, both ideas could not be pursued simultaneously. The core strategic question became:

Should we improve Copilot's visibility throughout the app or create a dedicated collaborative experience where users can work with Copilot?

Investing in the Widget would be a discoverability bet and investing in Workbench would be a workflow bet. I had a hunch that the Workbench represented a higher-leverage investment.

Evaluating Two Futures

Validating Ideas for Design Direction with Prototyping & Testing

To validate my hypothesis, I built clickable Prototypes on Claude to run a usability study. I interviewed 5 Copilot MVP users. All self-proclaimed "tab warriors" said the current workflow - Copilot in one tab and the builder in another - works just fine.

Workbench Prototype

Widget Prototype

When introduced with the Workbench concept, the reaction was an immediate strong preference (lots of oohs and aahs). The split-pane environment didn't just reduce friction, it changed how users perceived Copilot - from an assistant to collaborator.

"I'm building with the AI and it's not AI building for me."

Copilot MVP User, Usability Study

"It feels like I have a buddy."

Copilot MVP User, Usability Study

The widget was received with less fanfare but users saw value in it for in-context debugging but did not think it was a natural fit for building.

A few key expectations emerged:

• Users expected the Workbench to be editable: making minor changes themselves in the workbench and prompting Copilot in the chat area for bigger changes
• Seeing a diff increased trust by making Copilot's changes feel reviewable and controlled

Understanding How Experts Build Agents

Working with Constraints

With Workbench validated as the direction, I initially advocated for iframing the existing AI Agent Builder into the Workbench Panel. It would provide users an editable environment immediately - allowing us to learn quickly.

As we explored the concept further, I began questioning one of my own assumptions: if Copilot was responsible for much of the construction process, should users still interact with an interface designed for manual configuration?

To answer that, I shadowed five members of Regal's FDE team, mapped their workflows from creation through launch, and conducted a card sorting exercise to understand how they mentally organized the agent building process.

FDE Workflow

FDE Mental Model

The observations from these sessions revealed that users worked in phases, spending most of their time iterating on agent behavior while testing continuously, and refining specific portions and preserving syntax they've had previous success with.

The insights led to three design principles that shaped the Workbench experience:

A Collaborative Workbench

Hi Fidelity Designs

The resulting Workbench is organized around the natural progression of agent creation observed during research. Rather than surfacing each configuration upfront like we do today, the experience is restructured around how users naturally build agents. By guiding users via stages, the Workbench reduces cognitive overhead while preserving the flexibility and control.