TL;DR

CM.com set the goal of becoming an AI-first company, but in a developer-first culture, AI was mostly seen as a way to automate processes, not to solve user problems. I led the design of AVA, our first AI design and implementation framework, and shifted the conversation from “what AI can do” to “what users truly need.”

[X] teams trained · [X] AI use cases defined · 5 user jobs for AI · AVA Beta live, first AI agent in development

Context

Like many tech companies, CM.com set the goal of becoming AI-first. As part of Aurora, our Design Language System, we needed to be ready to integrate AI in a way that made sense for users, keeping usability, clarity, and trust at the center.

The challenge wasn’t only technical. In a developer-first culture, AI was often seen purely as an automation layer. Without a user-centered approach, every team would build its own AI experience in its own way, repeating the same inconsistency problems we had just solved with Aurora.

During our DLS sprints, a group of designers and developers came together with three goals:

1. Create the first AI chat for our platform.

2. Define UX patterns for AI that any team could reuse.

3. Enable AI support for complex processes, not just simple tasks.

Starting with people, not technology

Task: Align teams on how AI could actually help our users.

Since our end users are internal teams across the company, my role was to keep the process open and collaborative. Before the sprint, I ran a workshop with POs, developers, and support teams to present existing AI UX patterns in the industry, brainstorm how AI could help their users, and collect real use cases from different product areas.

Findings:

• Our platform could use AI for simple tasks like content creation, auto-filling data, filtering, and generating insights.

• Many processes still required manual review and external tools, so AI needed to support complex work, not only simple prompts.

• Transparency is critical. Users need to understand what the AI is doing and why.

I then organized two company-wide workshops to change how teams approached AI: start with user pains and goals, not the technology, and only then explore how AI could assist or enhance those experiences.

Impact: Teams generated [X] practical AI use cases, each aimed at solving at least one clear user pain. AI stopped being seen as a “nice to have automation layer” and became a strategic design tool, focused on supporting people, not replacing them.

AVA Chat: Building trust first

We started with AVA Chat, a natural-language interaction that could simplify many processes. We covered the essential AI interaction patterns, including conversational flow, input methods, and response types (text, media, code, tables).

But based on our findings, transparency was the priority. If users didn’t trust what AVA was doing, nothing else would matter. So we designed trust into the core of the chat:

• Showing sources and citations, so users could verify answers.

• “Show your work” steps, so users could follow the AI’s reasoning.

• Clear data ownership and prompt transparency, so users always knew what was being used and why.

AVA Panel: When chat isn’t enough

After testing AVA Chat, we saw that natural language alone wasn’t enough. [Add what testing showed, for example: “users lost context when switching between the chat and the platform” or “X of Y users struggled to complete multi-step tasks.”] Complex processes needed more structured AI assistance that worked alongside the main platform view.

So we created the AVA Panel, which lets users keep their platform visible while interacting with AI, and provides proactive suggestions and step-by-step guidance. In a second workshop, we defined three core roles for its first version:

• Assist: Complete tasks, fill in details, and move processes forward without interrupting the workflow.

• Simplify: Automate and streamline processes where possible.

• Guide: Provide timely suggestions and tools to keep users on track.

Defining AI’s jobs

With the general patterns defined, there was still one risk. As we learned with Aurora, AI wouldn’t bring real value if it wasn’t designed around real user needs.

Using our user feedback database, personas, journeys, and research from projects I’ve run across the company, I defined five main jobs for AVA, each with corresponding UX patterns that the DLS provides:

1. Onboarding new clients

2. Setting users up for success

3. Real-time troubleshooting

4. Automating routine tasks

5. Customer feedback analysis

This gives every team a clear starting point. Instead of asking “where can we add AI?”, they can ask “which of these jobs does my user need help with?”, making AI easier to integrate and more valuable for users.

The Results: Launching AVA to the company

At Dev Days 2025, one of CM.com’s most important internal events, I officially launched AVA as our first AI design and implementation framework. I presented how AVA was created, from research and UX foundations to collaboration with AI engineers, key use cases, and how AVA simplifies AI implementation across multiple products.

The launch positioned AI as an accessible, user-centered capability rather than a technical experiment. [Add a result, for example: “X teams asked to apply AVA patterns in their products after the talk.”]

[Image: Dev Days talk]

Impact

• AVA Beta is live in our testing environment, and I’m leading a new round of workshops to help teams integrate AI patterns into their products.

• AVA has moved beyond UI. I’m now collaborating with our internal AI team, positioning AVA as the foundation for designing agents that bring real user value.

• The first UX-shaped AI agent is in development. Together with one of our SaaS product teams, we used user journeys, research, and testing to identify the key pains the agent should solve, while keeping quality and ethics at the center.

This marks a milestone for CM.com: UX now actively shapes how AI behaves, not just how it looks.

“AI is just the tool, the impact comes from what you can do with it.”
— Conclusion of the AI discussion session

What I learned

The hardest part of AI design is the mindset. The patterns were well documented in the industry. Getting a developer-first company to start from user pains instead of technology was the real work.

A design system makes AI scalable. Because Aurora already had structure, tokens, and user job mappings, we could define AI patterns once and let every team reuse them.

What I’d do differently: [Write your own. For example: set success metrics for AVA from the start, so we could measure its impact from the first beta.]