RPS Overview
Dispatchers were planning flights on systems built in the 1960s. The work meant moving between tools, rekeying numbers by hand, and holding the full picture in their heads because nothing held it for them. Skyplan was meant to fix all of it, an ambitious system spanning the entire flight lifecycle, and it proved too large to land. What survived was RPS, scoped to the one calculation that has to be right before a plane leaves the ground.
The timing added a second problem. While we were rethinking how dispatchers work, we were also being asked to rethink how we work, bringing AI into the design process to move faster without giving up rigor. Two unfamiliar problems at once: one system to modernize, one practice to reinvent.
My Role
Senior UX Designer IV
Stakeholders
- Pilots
- Dispatchers
- Performance engineers
- Flight operations
Tools
- Miro
- Figma
- Teams
- Figma Make
- Zoom
- Claude
- Jira
Research
RPS served dispatchers, pilots, performance engineers, and flight operations. What follows is one piece of it: the Summary modal, built for dispatchers. We started with the chief dispatchers. We mapped their workflow, every tool it took to get through a shift, and where it broke down. Then we went to the dispatchers themselves, interviewing them and watching them work, looking for the manual steps worth automating and the ones worth rethinking entirely.
We also had a head start. The Summary modal was a pared down version of the Sandbox, an experience we had already designed and delivered for admins, chief dispatchers, and flight operations. So rather than gather requirements in a document, we built the prototype with AI and brought it to the stakeholder group. When they asked for a change, we made it in the room and showed them.
I worked alongside our UX researcher on every prototype. I brought the structure, the layout and hierarchy the team had already agreed on, and together we fed it into Figma Make to build into the prototype. The agreement always came first. AI built what we had decided, not the other way around.
Sketch + Wireframe + Prototype
AI broke our process. Research, sketching, wire-framing and prototyping stopped being four steps and collapsed into one, and each loop took more iterations than the last as the tool gained shape.
Fidelity became the problem. What came out of Figma Make looked finished, and finished work invites people to build. Teams pulled screenshots of work in progress and started development from them, and we had to flag that work as design debt to be addressed later. So we turned the fidelity down. We had Make render the prototype as low fidelity wireframes, which kept the conversation
Testing
Once the chief dispatchers were satisfied, we put the clickable prototype in front of the dispatchers themselves. They gave us the feedback only daily users can, we incorporated it, and we brought the changes back to the stakeholders.
High fidelity Prototype
With the requirements settled, we applied the design system built for Skyplan and RPS and moved the work to high fidelity. I manually adjusted the screens while my researcher updated the Make prototype in parallel, so stakeholders and developers always had something fully clickable to interact with, not a static file to interpret. We handed off both: the screens for specification, and the prototype for behavior.
Other RPS Screens
RPS and Skyplan were designed in dark mode. Dispatchers work in dim rooms for entire shifts, and an interface that glows white in that environment is a physical problem, not a stylistic one.
The Summary modal is the exception. It launches from an application that lives in light mode, and matching it kept users feeling like they never left. Consistency with the surrounding experience mattered more than consistency with our own system.
My Learnings
- The worry that AI is coming for design work assumes the tool decides what gets built. Ours never did. It built what we had already agreed on, and it built it fast enough to change in front of a room. The value only became clear once we stopped asking it to think and started asking it to keep up.
What's next?
- RPS is in development, and we're consulting with the dev team to resolve design questions as they surface. We'll re-engage during QA to capture the design debt from the screenshots that got built too early.
- We're teaching the rest of the design team how to fold AI into their workflow, including how to keep token costs down while doing it.
Teaching the practice
I built the team a prompting guide for Figma Make. I built it with AI. It assumes no prior experience, addresses the fear that AI is coming for design jobs, and ends with a prompt template that bakes the cost saving habits into the structure itself: state the goal, the users, and the constraints up front, then ask for one screen at a time.
