Challenge
Create an intuitive marketplace that could help employees work more efficiently today while establishing a foundation for a growing ecosystem of agents, builders, and future capabilities.
Visa was expanding how employees could use AI in their everyday work. As the sole product designer, I translated an early platform vision into an MVP marketplace for discovering, using, and creating purpose-built agents across consulting workflows.
Create an intuitive marketplace that could help employees work more efficiently today while establishing a foundation for a growing ecosystem of agents, builders, and future capabilities.
I turned a broad platform vision into a usable marketplace model, introduced research before launch, and helped the team prioritize what needed to change for MVP.
The MVP launched globally to 2,000+ employees as the organization’s first enterprise agent marketplace and part of its broader digital and AI transformation.
I organized the experience around three clear actions: find an agent, use it in a conversation, or create one. The interaction model helped people understand what an agent could do, when one was active, and what happened when they moved between agents.
The MVP began with only with six agents, I designed the experience for a much larger catalog it would need to support later, using a repeatable information model and discovery patterns that could expand without redesigning the experience each time.
User research wasn’t part of the original delivery plan. I made the case for adding a week of validation before release, then used seven sessions to identify and prioritize changes for MVP.
Testing exposed where people needed clearer language, stronger guidance, more context about agents, and changes to the creation flow. Those findings helped the team focus engineering effort on the changes most important to address before launch.
A new marketplace needed a clear entry point. When a full onboarding experience wasn’t feasible, I preserved the user need with a lightweight walkthrough that introduced the core features without slowing launch.
Multiple participants lost track of which agent they were speaking with. You strengthened the active-agent indicator so the system state remained clear throughout the conversation.
Experienced builders wanted to test behavior before naming and branding an agent. You changed the sequence and clarified the flow so creation better matched their working model.