The Problem

Data stewards managed data agents entirely from the command line — upgrading and rolling back versions by hand. Seeing what a change would affect meant a long chain of commands; knowing whether an agent was even online meant hunting through the CLI or digging into its history on screen. The commands were long and easy to fat-finger.

The Real Problem

Two teams, one root cause: the CLI made routine, high-stakes work — version changes on live agents — both blind and error-prone. Back-end command logs showed people repeatedly mistyping the lengthy commands and burning time just to answer "what depends on this?" The fix wasn't a better command. It was making the system's state visible.

What I did

Combined two separate efforts version management (Data Integration) and an agent health dashboard (Data Configuration) — into one coherent control surface.

Designed the version-management experience in Figma upgrades, rollbacks, and a clear view of dependencies and downstream impact that the CLI kept hidden.

Designed and built the health dashboard as an experience engineer agentically, in code — so stewards get status, history, and instant outage alerts at a glance instead of querying for them.

Climbed into the back end with the engineers to learn how versioning actually works — affected files, dependencies, failure modes — so the UI reflected real system behavior, not a designer's guess.

Pushed for new interaction patterns over the familiar table layout, making the case to PM and engineering for why they served users better.

Impact

~80% fewer task errors in internal testing validated with support engineers and fresh-eyes back-end engineers. (Production rollout was paused by a company-wide stack migration.)

Dependency impact that took a chain of CLI commands to uncover became visible at a glance.

Customers, shown the work via support engineers, expected a large cut in the time spent managing and monitoring agents.

The Hardest Call

Two fronts. Convincing the PM and engineers to leave the familiar table-style interface for new interaction patterns took repeated discussion and a clear UX rationale. Harder still was the technical climb: to design version management honestly, I had to understand how it actually works under the hood — which files change, what dependencies break, how the process runs — sitting with engineers until the system model was mine, not theirs. That depth is exactly what let me build the dashboard rather than just mock it.