M
← All work
AI AgentsSalesAi · 2025 · Intern

From an expert-only setup to a flow anyone can finish.

Creating an agent used to mean building separate inbound and outbound agents, then hunting through scattered settings. I restructured it into one clear flow where a single agent does both.

1
agent, both directions
reliance on CX to build
setup completion · TBC

Role · Product & UX design — research, IA, flows, UI, design system

01

The problem

Creating an agent was effectively an expert task. You couldn't make one agent that did both inbound and outbound — you built, and often duplicated, a separate agent for each.

Settings were scattered with little explanation, and our users aren't especially tech- or AI-savvy. It was confusing enough that CX usually built agents for people.

Before — the scattered settings (recreated mock)
Fig. 01 — Before: scattered settings, recreated to protect privacy
02

Understanding why

The flow assumed a technical confidence users didn't have, and its structure mirrored the system's internals instead of the user's goal.

People didn't need more options — they needed a confident path, and to understand what each choice actually meant.

What mattered vs. what was exposed
Fig. 02 — The gap between what mattered and what the UI exposed
03

The solution

One agent that handles both inbound and outbound. Settings grouped logically, each with a plain-language description, surfaced where it makes sense.

Far simpler for non-technical users — CX still helps, but it's no longer essential. It also finally looks modern and polished, like a sophisticated AI company.

After — the grouped, guided creation flow (recreated, dummy data)
Fig. 03 — After: the grouped, guided creation flow (recreated, dummy data)
04

Outcome

Placeholder for shareable results — e.g. more users creating agents on their own, and less CX involvement.

The designer's bonus: the work produced a clean, well-organized design system that's now easy to reuse for faster iteration.

Where I'd take it next

Keep simplifying — describe the job in plain words and have the agent pre-configure itself, with the structured flow as the fallback for fine-tuning.

Reflection

Designing for non-technical users taught me that clarity beats power — the best setting is often the one you never have to think about.

Privacy note — screens are recreated with dummy data and details simplified; the real product evolves with business decisions and user needs.

Next project

Using an agent — the workflow

Read →