Margo Sultenfuss

← / CASE STUDIES

2025–2026 · Agent.ai

Agent Builder

Interaction Design1-to-Scaled

A low-to-no-code agent builder that took 18,000+ builders from idea to a live, running agent.

Role
Lead Product Designer
Scope
Led redesign of the core agent-building tool on a platform that grew to 2M+ sign-ups across 180+ countries.
Outcome
Opened a powerful but technical tool to non-technical builders without losing the power users who depended on its depth.
FIG. 01The redesigned agent builder, in the live product.

87,000+

Granted builder access ([TODO: define '87,000+ access' meaning + window])

49,000+

Published agents (cumulative, all-time; not post-redesign)

18,000+

Publicly published agents

[PENDING MIXPANEL]

Start-to-publish conversion, before vs. after redesign

Most builders kept their agents private.

/ THE PROBLEM

Agent builder was powerful but originally built for technical users. We needed to open it to non-technical builders without breaking the depth power users relied on. Every decision had to serve both audiences at once.

/ WHAT I BUILT

Restructured agent creation from four pages into a two-page flow ("Create" and "Settings"), reordered around how people actually think through building an agent. Redesigned the action library (seven tabs, no search, no descriptions) into a slide-out panel with fuzzy search and new categories. Added a live preview panel with a toggle for underlying code and context, so anyone could test and debug what they'd built. Redesigned the UI, implementing updated styles.

FIG. 02The two-page agent creation flow ("Create" and "Settings"), restructured from four pages.
FIG. 03The live preview panel, with a toggle for underlying code and context.
FIG. 04The action library, redesigned as a slide-out panel with fuzzy search.

/ HOW I PROVED IT

Validated through user testing with both technical and non-technical builders, confirming the redesign held up for power users while opening the tool to newcomers. The tool was later acquired by HubSpot for its own Agent Builder, external validation of the core bet that the agent-building layer was the thing worth owning.

/ The Long Version

Read the full case study

Coming soon
← / CASE STUDIES