Designing a university run entirely by AI agents

Exploratory research

User interviews

Solutions

User flow

The agent map

A Lead Orchestrator holds each student's record end to end. Below it, agents hand off in sequence: Admissions interviews and decides, Enrollment collects the $250 deposit, the General Course Agent builds timetables and rolls grades into a GPA, Placements matches certified students to real placements, and the License Agent preps them for the exam.

InsightDesigning this meant deciding who talks to whom and when, not just what each screen looks like. One early Course Agent taught pre-surgical hygiene, scored down to micro-skills with required retests.

Iteration: video interviews become audio

I wanted admissions interviews on video, the way a real admissions officer would run them. We could not solve it in time.

InsightWe shipped audio instead, so students could talk rather than only type. An open trade-off, with video still on the list.

Low-fidelity exploration

Early exploration was the system before the screens: mapping agent handoffs and student states, then wireframing where students meet each agent.

High-fidelity design

The final designs carry students from interview to coursework to grades while always making clear which agent you are talking to and what happens next.

Prototype (recording)

Usability testing

I tested the product myself throughout, alongside a second dedicated tester as each agent stage came online.

Results

• →Typing a whole admissions interview felt cold, which pushed us to the audio option
• →It was easy to lose track of which agent you were talking to, so each got a clear identity in every conversation
• →Retests did not say what to redo, so score feedback now points at the specific skill

Measuring success

Was the project successful and why?

Paused, on purpose

Xel.care has zero customers; we paused before reaching Placements. The deciding insight came from research, not the build: most people wanted a paycheck more than care work, and a passion gap is harder to close than a cost gap. The stakeholder paused Xel.care to focus on AfroHealth. For me, this is the project where I stopped designing products like they belonged to someone else.

Metrics and KPIs

• →Flow built end to end through the Course Agent stage, including hands-on skills training
• →Zero customers at pause, stopped at the Placements stage
• →Research redirected the model: paycheck motivation outweighed passion for care work

Future vision / next steps

• →Apply the model to fields where passion already exists but affordable education does not
• →Xel.tech is one direction under consideration
• →Agent video calls remain on the list