Scaling UX with an AI Design Agent

〰️

Scaling UX with an AI Design Agent 〰️

TL;DR

To help our UX team keep pace with a demanding roadmap, I designed and built the UX Agent—an AI-powered assistant that automates early-stage UX analysis. Triggered from Aha!, it synthesizes product data and user research to generate design guidance, flow diagrams, and task tickets—cutting design ramp-up time nearly in half. The result: faster starts, deeper insights, and a scalable foundation for design excellence across the organization.

Problem Statement

The design team was navigating the challenge of supporting an ambitious, multi-suite product roadmap with limited time and resources. The volume of work was increasing rapidly, and expectations for UX to contribute early and meaningfully to product planning were higher than ever.

I saw an opportunity to rethink how design work was initiated and supported. I wanted to scale our impact without compromising craft. I saw AI not as a replacement for designers, but as a strategic tool to amplify their effectiveness—by moving faster, reducing repetitive work, and surfacing deeper insights from day one.

AI is uniquely powerful at synthesizing large volumes of product requirements, user feedback, and research transcripts—surfacing patterns and opportunities far more efficiently than any human could. By embedding this intelligence directly into our workflow, I set out to transform how our team worked, not just what we delivered.

Solution

To bring AI into our workflow in a meaningful way, I personally designed, built, and deployed the UX Agent—an internal AI assistant that automates early-stage UX analysis and integrates directly with our existing tooling in Aha!

I began by setting up a dedicated UX workspace inside Aha!, giving our team a way to independently manage tasks while staying tightly linked to product features. This setup brought visibility to UX work and allowed us to connect our efforts to roadmap planning and OKRs—something that hadn’t existed before.

With this foundation in place, I developed the entire UX Agent system myself—including the backend logic, AI integration, and automation infrastructure. I built a Flask/Python application, hosted on Render.com, and connected it to Aha! via webhooks and API calls. The workflow was fully automated and tightly tailored to our internal design and planning processes:

  • When a product feature was created in Aha! and marked with a “Needs UX” flag, a webhook sent the feature data to my Flask app.

  • The app validated the flag, then routed the content to an OpenAI assistant I trained using OpenGov-specific materials: product documentation, HeyMarvin user interviews, our internal design system, and accessibility standards.

  • The AI returned structured UX analysis, including detailed design considerations and mermaid.js flow diagrams for multi-step or multi-screen experiences.

  • The Flask app then used Aha!’s API to automatically generate a design task—linked to the originating feature and populated with the UX guidance—within the correct product suite’s workspace.

The result was a seamless, intelligent PM → UX handoff that gave designers an immediate, AI-powered starting point—saving time, improving planning accuracy, and allowing more energy to be spent on actual design work.

This was more than just automation—it was augmentation. Designers remained fully in control of their process, but now had a smart assistant trained on the company’s ecosystem working alongside them from day one.

Example Output: AI-Generated UX Guidance

Below is a real excerpt from the UX Agent’s output, automatically generated from a new product feature in Aha!.

click on image for full view
Example: AI Generated Flow diagram using Mermaid.js.  Designers can use mermaid.live to view diagram.

Selected Output from the UX Agent

  • “Use a tabbed interface to separate Associated Rate Codes, Accounts, and Version history for clarity and navigability.“

  • “Add a persistent side panel for quick access to rate code version history.“

  • “Ensure all data elements are hyperlinkable to facilitate fast cross-referencing.“

  • “Apply conditional logic to UI based on user roles (e.g., Admin vs. Clerk).“

  • “Follow WCAG 2.1 AA standards for all interactive elements.“

  • “Include a future-proof UX flow for adding/editing upcoming rate code versions.“

AI doesn’t just generate ideas—it generates context-aware, standards-aligned design direction. This gave our designers a structured head start without compromising craft.

Impact

The launch of the UX Agent marked a major shift in how design work was initiated and executed. By automating the early stages of UX analysis and planning, I cut the time to start and complete design work by nearly half—freeing up designers to focus more deeply on problem-solving, refinement, and innovation.

The response across teams was overwhelmingly positive. Product Managers appreciated the speed and structure it brought to the planning process. Designers gained confidence and clarity, receiving rich, context-aware guidance without the usual scramble to interpret loosely defined feature briefs.

Beyond productivity gains, the UX Agent also helped normalize AI as a supportive force within the design org. It reduced uncertainty around AI’s role in our discipline by demonstrating its value as a tool—not a threat. I consistently emphasized the message: AI isn’t here to take your job. It’s here to help you do it better—and someone who uses AI will.

The system proved so effective that it inspired interest in expanding the agent model—connecting the PM Agent to the UX Agent, and exploring how a future Engineering Agent might close the loop from planning to implementation. We were no longer just improving workflows—we were pioneering a fully automated PM → UX → ENG handoff pipeline.

The UX Agent became more than a tool—it became infrastructure. And it gave our design team something we hadn’t had before: the ability to scale insightfully, not just reactively.

Future Vision

Building the UX Agent was only the beginning. With the foundation in place, I began exploring how we could extend AI’s value even further—moving from insights to actual design outputs.

My next steps focused on integrating design automation into the workflow. I explored ways to have the UX Agent generate:

  • Clickable HTML wireframes for early-stage concept validation

  • Auto-generated Figma files preloaded with layout structure and design system components

  • Production-ready HTML front ends, built directly on our internal component libraries

I experimented with tools like Cursor (using MCP servers to translate Figma into HTML), Builder.io’s Figma plugin, and Vercel for deployment-ready front-end builds. The goal wasn’t to eliminate design work—it was to compress the time between idea and implementation, and to give designers a stronger, more structured starting point.

In parallel, I started exploring agent-to-agent integration:

  • The PM Agent creates a feature

  • The UX Agent analyzes it and generates design tasks

  • A future Engineering Agent receives that design and initiates development planning

This vision of a fully connected, intelligent product development pipeline—PM to UX to ENG—has the potential to transform how teams work together. Not through top-down mandates or rigid automation, but by supporting humans with intelligent, contextual tools that amplify creativity, speed, and alignment.

🛹