AI Agent Builder

Designing a No-Code AI Agent Builder

Regal's AI Agent Builder is a no-code platform for deploying customer-facing AI agents.

Principal Product Designer

As the sole designer owning the end-to-end design process, I built Regal's AI Agent Platform from the ground up — from strategy and research to shipping the MVP and beyond.

Product Strategy User Research UI/UX Design Prototyping & Testing

Q1 2025 (MVP launch) → Q3 2025 (V2)

AI Agent Builder interface
The Problem

Stuck Between Power and Usability

How can Sales Leaders innovate and enable AI in their processes?

When AI Agents became a product offering on Regal, there was no dedicated in-app tool to build them, so customers had no way to get started on their own and were completely dependent on Regal's FDE team to deploy an AI Agent, which wasn't scalable.

Our users are AI-curious and want to automate workflows, but they're focused on business outcomes, not figuring out how to prompt or tinker with AI.

Takeaway

We needed to build a tool that both customers could use to build robust AI agents without too much technical expertise and leaning too much on Regal's FDE team.

Research

Getting to Know the AI Landscape

How can we empower our users to innovate with and benefit from AI?

Our existing "Sales Leader" persona — a tech-lite, non-engineer who invests deeply in hitting their conversion and revenue goals and is curious about AI — from a previous UX study I conducted, helped shape how I framed my research:

Competitive Research

Most tools were either too complex or too simple — and all assumed users could prompt. From the research I was also able to identify requirements to design for, like the need for a testing experience.

Competitive Analysis

Subject Matter Interviews

Talking with internal teams surfaced two things:

  • Building an effective AI agent follows a kind of formula. There's a consistent structure.
  • Testing wasn't a final step — it happened with every incremental change.
Subject Matter Interview Framework

Customer Insights

Some of our customers were already experimenting with AI to start automating repetitive tasks but didn't know where to start or had issues getting something production ready.

"I'm optimistic about the power of AI on our business - but we haven't found a tool that works for the way we actually run our business"

Sales Leader, Current Regal Customer
Hypothesis

From Insight to Direction

Designing for a New Experience

Based on feedback from customers and learnings from SME interviews and competitive analysis, Regal's Agent Builder experience needs to inspire instead of overwhelm. By designing a guided experience that also limits upstart effort, we believed opinionated structure could outperform open-ended flexibility for non-technical users.

Hypothesis

If Regal's AI Agent Builder is designed to be structured and guided, then non-technical users would be able to create successful AI Agents - without needing extra support from internal resources or Regal's internal teams.

Ideation Phase

Wireframing & Sketching

Aligning with Stakeholders Before Designing

This was a 0-to-1 greenfield project, so I started the ideation phase with some sketches.

Notebook sketches

The sketches helped me formulate design questions, so I translated those into illustrative mocks and brought them to a kickoff meeting with my PM and CPO to get input before committing to real design work:

Form vs Wizard wireframe

During this meeting, I facilitated alignment around the Form UX by grounding my recommendation with insights from my research:

The Takeaway

A structured form was the only pattern that didn't fall into either trap.

From Decision to Flow

Defining the MVP

The meeting helped me define an MVP flow made up of 3 pages:

MVP flow diagram

Each page then went through multiple rounds of design options — except the list page, which reused an existing pattern. Knowing that AI Agents was a pivotal moment for Regal, I made a deliberate choice to push the template page visually, despite directions to stay within existing styling to signal a meaningful shift in Regal's platform.

Template page explorations

The builder form also went through rounds of layout exploration. The core challenge was figuring out how to organize all the required pieces into a single page that felt structured — not overwhelming.

Builder page explorations
Product Strategy & Design Decisions

Design Framework

Translating Insights to Product Direction

The biggest design challenge was balancing clarity and complexity - the solution needed to support a wide range of requirements such as guided prompting, customizable agent settings, and function-calling flows (Actions) into a single experience.

I drove three key decisions that shaped the product's direction:

01

Templates as Launchpad

Rather than opening on a blank builder, I pushed for templates that reflected the most common use cases from interviews. This gave users a concrete starting point, reduced time-to-first-value, and let them see what "good" looked like before they customized anything. I presented this framing to leadership to get buy-in over a more open-ended approach.

Templates as Launchpad
02

Structured Inputs Over Freeform Prompting

I made the call to break prompting into discrete, labeled sections rather than a single open text field — the industry default. This reduced cognitive load, passively taught users what makes an effective agent, and produced higher-quality outputs from day one. The tradeoff was flexibility for power users, which I scoped as a V2 concern.

Structured Inputs
03

Validate Before Launch

Since testing is woven throughout the build process, I pushed for an in-product testing experience so users can fluidly edit and test an agent without leaving the builder. Collapsing the feedback loop into one place meant users can deploy confidently rather than discovering problems when the agent is already live.

04

Scoped Out Power User Entry Points

I pushed for two additional ways to start building an agent: a blank canvas for more experienced users who didn't need the guardrails, and the ability to upload a call recording or script to auto-generate an agent — a more sophisticated solve for the blank canvas problem that met users where their existing knowledge actually lived. Both got deprioritized due to resourcing and timeline constraints.

Design

Hi-Fidelity Designs

Bringing AI Agents to Life

The high-fidelity designs translated complex product requirements into a cohesive, intuitive interface. Regal users can now build, test, and deploy AI agents with confidence.

Single State Agent Workflow — Updated 2026

Use the prototype below or view it in Figma to walk through the end-to-end flow of building an AI Agent — from selecting a template to configuring the agent to testing.

01 — Template Page

Discovery Aid

Templates reduced the cognitive load of starting from scratch and helped users understand what's needed to deploy an AI agent. It also helped spark ideas on how a user can leverage AI agents for other use cases.

02 — Builder Page

Translating a Workflow into UI

The builder interface made the underlying structure of effective AI agents visible. By listing out each part of the "formula", a user can passively learn the requirements of a successful AI Agent while building.

03 — Model Settings

Configuration without Complexity

Technical capabilities like voice and model settings were designed in approachable dropdowns so AI configurations can feel manageable.

04 — Actions

Conversation AND Execution

Translated backend capabilities into composable building blocks so non-technical teams can create operational AI Agents without having to write a lick of JSON.

05 — Settings

Designing for Real World Behavior

Beyond the core configurations, the builder also accounted for edge cases and conversational nuance so agents can feel more human and handle the unexpected.

06 — Testing

QAing While Building

Testing in-context meant users could catch gaps in their prompts, tune the voice, and build confidence without spinning up a live campaign first.

07 — Agent List Page

Lifecycle Management

AI Agents are now a core part of user's Regal experience. A management layer reinforced that an AI agent is a living system that can be continuously refined and requires monitoring.

Template page Builder page Model settings Actions Settings Test mode Agent list page

Results — Q2 2025

The Results

From Launch to Impact

Within the first quarter after launching the MVP Agent Builder, customers could get to a live, deployed agent much quicker with less reliance on Regal's internal team. The structured onboarding model reduced implementation friction enough that customers could independently deploy production-ready agents across multiple industries. We saw successful agent deployments across multiple use cases — validating that the structured, guided approach worked in practice:

Medicare Advantage

Qualification

96%

CSAT for AI Agents

Roadside Assistance

Notice of Disablement

60%

Containment Rate

eCommerce

Customer Support

73%

Solution Identified

Insurance

Collections

60%

Collection Rate (Even with Humans)

These outcomes validated the core design bets — templates reduced onboarding friction, structured inputs produced more reliable agents, and the scoped MVP let us iterate quickly once real usage patterns emerged.

8

Paying AI Agent brands
(goal: 10)

31

Live AI Agents deployed
(goal: 35)

422k

Tasks completed by agents
every 30 days

Where the Product Stands Today

From MVP to Full Platform

My involvement didn't end at MVP. I helped shape the post-launch roadmap based on emerging user needs — and the Agent Builder has continued to mature.

MVP · Q1 2025

Foundation

  • Templates
  • Structured Inputs
  • Test Audio + LLM
  • Core Actions

V1 · Late Q2 2025

Knowledge & Testing

  • RAG / Knowledge Base
  • Test Cases + Simulations
  • Expanded Actions

V1.5 · Late Q2 2025

Analysis

  • Post-Call Analysis
  • Collected Variables

V2 · Q3–4 2025

Advanced Control

  • Conversation Flows
  • SMS AI Agents
  • Start from Scratch

Q1 2026

Platform Maturity

  • Agent Drafts + Variants
  • AI Agent Versions
  • Global Nodes + Actions

Together these features mark a shift from "build your first agent" to "manage a fleet of agents" — which is exactly where our most mature customers needed the product to go.

Beyond the Product

I documented the Agent Builder's patterns into a reusable component library — giving the team a foundation for how AI surfaces are designed across the product.

Multi State Agent Workflow — Q2 2026

Use the prototype below, or view it in Figma, to walk through a more complex end-to-end flow — exploring the broader platform experience including Conversation Flows, multi-state agents, and Agent Variants for iterative experimentation.

01 — Simulations

Ready for the Real World?

Stress test how AI Agents handle nuanced and real-world scenarios by simulating full customer conversations against a set of predefined test cases. By representing expected and edge case interactions, the testing suite builds confidence in agent behavior before deployment.

02 — Knowledge Base

Grounding Agents with Context

Connect documentation like FAQs, pricing, policies, and product information directly into the agent experience by creating a knowledge base in a dedicated workflow separate from the prompting logic. Separating knowledge management from prompting makes agents easier to maintain as documentation evolves.

03 — Conversation Flows

Scaling Beyond Linear Conversations

Conversation Flows introduced multi-state logic that allowed users to design branching conversational paths based on customer intent, enabling AI agents to support more operationally sophisticated workflows.

04 — Agent Variants

Designing for Continuous Optimization

Run multiple agent variants simultaneously and compare performance across prompts, tone, and strategies. Variants enables iterative experimentation driven by real production outcomes.

Simulations Knowledge Base Conversation Flows Agent Variants

Update — Q1 2026

Built to Last

As the product matured, the MVP design choices became the foundation that made more advanced features approachable. The early decisions created room for the complexity of the platform it is today.

The platform has grown from a handful of early adopters to one handling millions of customer interactions every month.

43

Paying AI Agent brands
↑ from 8 at launch

610

Live AI Agents deployed
↑ from 31 at launch

3.1 Million

Avg monthly tasks completed
↑ from 422k at launch