Lead Designer — Summer Internship

Agentic Web: Designing a Campus-Scale Internet of Agents

As the sole designer on a three-person team (1 designer, 2 developers), I led the UX and interface design for the Agentic Web, a campus-scale "Internet of Agents" that lets researchers describe tasks in plain language and have the system automatically route them to the right AI agents. My work over the five-week sprint focused on turning a developer-built, technically-exposed prototype into an interface usable by three distinct audiences: non-technical researchers, agent developers, and PIs/admins.

I began with a heuristic audit of the live interface, producing eight prioritized findings across usability, admin/user separation, system status, and feedback — for example, flagging that raw API capability tags (email/calendar/delete_event) and debug info (gpt-4o-mini, ctx:) were being surfaced to end users, and that internal jargon like "ADS Agents" needed plain-language replacement. From there I defined the user flow around fully automatic, natural-language routing (so users never have to pick an agent), designed a full system-status model (idle / thinking / responding / error), added an empty state with suggested prompts, and introduced a per-response feedback loop.

I built an HTML prototype of the redesigned researcher view, began a separate admin view for agent management and monitoring, and prepared a PI presentation framing each design decision against an audit finding — closing with the open strategic question of whether agents should be visible to users at all.

AI Agents Heuristic Audit Systems Design

Heuristic audit of the live Planner Agent prototype — Heuristic audit of the live prototype — eight prioritized findings spanning usability, admin/user separation, system status, and feedback.

HTML prototype of the redesigned researcher-facing view.

Early admin view for agent management and monitoring, built for PIs and admins as a distinct audience from researchers.