How an AI-woven process made Aeris 10x faster by removing the design-to-development handoff entirely.
By James Lee · VP of User Experience Engineering, Aeris Communications
design
removed
“I need you to reimagine”
Shortly after I joined Aeris in December 2025, my CTO Drew Johnson gave me a brief that was five words long: “I need you to reimagine.” Not improve. Not optimize. Reimagine. What I found was an engineering organization that already builds sophisticated technology. What I could contribute was making sure the experience customers interact with every day reflects that same standard. The first thing I questioned was the handoff, and with it, our dependence on the traditional SaaS approach to design.
The breaking point
My first project was the Aeris Agentic AI User Experience across our entire portfolio, one of the biggest product-related initiatives. It gave me a close look at a problem every design leader knows but few name honestly. We design in one tool. Engineering builds in another. Between them sits a translation layer of redlines, specs, and review meetings, and every translation introduces guesswork. The handoff was not making us efficient. It was manufacturing friction and calling it process.
So I made a decision that surprised my own team. We would not improve the handoff. We would remove it.
What we built instead
We cancelled our SaaS design tools. No Figma. No Adobe. Not as a cost measure, but as a position: design capability should never depend on a tool. A team that is tool agnostic can adopt whatever comes next, and in the AI era, what comes next arrives monthly. Design at Aeris now happens directly in code, with AI as the production layer.
One thing we are clear about: we use AI to amplify, not to shortcut. The work is still hard. The thinking is still ours. What changes is the output. Where we used to produce one, we now produce many, at the same level of quality and intention. Leveraging AI is not about doing less. It is about establishing a new efficiency where effort stays constant and results multiply.
- AI-first production. Designers work with Claude Code and GitHub Copilot inside VS Code, choosing models by the complexity of the task. The design conversation and the build are the same activity.
- HTML from the first sketch. I still begin with hand sketches for inspiration, then capture them straight into the AI tool, never through a design tool to be digitized. AI bypasses that middle process, so the first digital artifact is already fully interactive HTML.
- Version control as the design library. Work lives in Bitbucket branches. The latest design is the latest commit, visible to PMs and developers without change rights; EXG, our experience team, controls versioning, so the source of truth stays protected while staying transparent.
- Aeris Design System 2.0 as the foundation. Fully token-based and AI-driven, so speed never costs us consistency. Every output is already on brand and on pattern.
“Handoff” no longer describes anything we do. What design produces is what engineering receives. Code-ready was a requirement, not an aspiration, validated with our developers before we committed. Design carries an enterprise product a certain distance; the back-end, the data, and the systems work is where our developers step in. What changed is where their time goes: into hard engineering problems, not into reinterpreting design intent.
One thing worth naming directly: mobile is not a shrunk version of the desktop experience. It is the same product, built from the same components, expressing the same design language. What adapts is the layout. What never changes is the intent. Because we work in code from the start, this is not something we solve at the end of a project. It is already true the moment the first artifact exists. Screen size stopped being a handoff problem the day we stopped having a handoff.
Designing the agentic experience, agentically
There is a fitting symmetry here. The Aeris Agentic AI Platform has a prompt-native user experience: users do not navigate it like a dashboard, they express intent and the system orchestrates the work. You cannot evaluate that from a flat mockup. Our HTML-first process meant we designed the agentic experience in the same medium our users live in, from the first iteration. We were not just using AI to design faster. We were using AI to design AI, digitally authentic from the very first artifact. A platform should feel as well engineered as it actually is.
The hardest part was never the technology
Setting up the process took a couple of weeks. Getting people to trust it took two months. The resistance was not irrational: designers build identity around their tools, and setting that aside feels like starting over. I had to be assertive on one principle.
Reimagining the process meant changing my own daily behavior first. A leader cannot ask a team to let go of something he is still holding. My team in Bucharest carried the heaviest load of this transition, and they deserve enormous credit for where we are today.
Build the capability, then open it up
At our yearly tech summit I went further than my team expected: the AI-woven process should be accessible to everyone in product, including PMs and sales engineering. The pushback came from inside my own organization. I fundamentally disagreed. A design system is not the last asset a team owns. It is evidence of a capability, and the capability is what we actually build.
Think of how automakers build a supercar that reaches sixty in under two seconds. It was never meant for everyone; it exists to push core capability forward, and that capability elevates everything else they ship. Opening up capability does not diminish the team that built it. It attracts partners who want to work with them. Today, the same team that pushed back is becoming the strongest proof of the argument.
What the numbers buy us
Design changes that once took a multi-week pipeline now complete in a day or two. We removed about half of our meetings and roughly 90% of the traditional design process. We move ten times faster, with the constraint no longer our capacity but simply deciding what to do next.
The fair question is where that time goes. It does not go to doing more of the same work faster. It goes to the two things AI cannot do for us: judgment and authenticity. When twenty interactive options cost what one mockup used to, the scarce skill is deciding, not making. And when anyone can generate, the differentiator is whether the work is genuinely yours. AI supplies the speed and the volume. We supply what makes the result unmistakably Aeris. Speed and volume are becoming table stakes. Judgment and authenticity are becoming the craft.
What comes next
We treated our own organization as the first customer of this transformation. The next step is making it a repeatable practice across Aeris, and eventually sharing it externally, because the gap we closed between design and development is not unique to us; it exists in nearly every product organization. For our customers, it shows in products where the experience is native to the code, not translated into it. We did not bet on a tool. We bet on a way of working where design thinks in outcomes, produces in code, and ships without translation.
I was asked to reimagine. We transformed.
James Lee is VP of User Experience Engineering at Aeris Communications, where he leads design, front-end engineering, and technical publications. He joined Aeris in December 2025.
