• Shipped · 2026
  • UX/UI Designer
  • B2B SaaS
  • AI agents
  • Figma · Claude Code

Creating a no-code builder for custom agents

A dedicated space in a security platform where customers compose AI agents of their own — any model, any tools. Designed from scratch and delivered through a design-to-code pipeline in about half the usual build time.

An MCP server's settings open beside the Custom agents navigation

TL;DR

Background
A security automation platform whose users wanted to improve their operational efficiency. Beyond the agents the product already provided, customers needed a way to connect any LLM provider or MCP server and build custom agents to fit their own work.
Problem
The requirements for the new feature showed that it was too bulky to sit in the usual configuration panel on the automation canvas. This required a rethink of the information architecture. At the same time, role-based permissions meant the design had to separate who may create and edit an agent from who may run it.
Approach
I worked closely with engineering, product, and customer success. Rather than designing every screen, I started with a prototype that tested the structure and behaviour. Then I created key screens and documentation, additions to the design system, and an agentic design-to-code and audit workflow.
Outcome
A new section for configuring and managing providers, MCP servers and agents. Workflow AI nodes gained a live view of the agent’s reasoning and an optional human decision checkpoint. Repeatable work now moves to agents each security team shapes to its own needs, so attention lands where judgement is needed.
Role & responsibilities
UX/UI Designer, hands-on from research through implementation
Team
1 UX/UI Designer, 1 Front-End and 1 Back-End engineers, 1 PM, 1 CS Manager
Scope
Information architecture, key screens and prototype, design system updates, UX writing, design QA, code contributions.

This case study is under NDA. Contact me to learn more about the project and my approach.