Agentic Design Kit

Overview

Designing the core agentic experiences across IBM.

Role

UX Design Lead

Date range

January 2025 - April 2025

Designing the core agentic experience for all IBM products.

The challenge

The Carbon for AI Essentials and Chat Kits gave IBM a foundation for consistent AI interaction patterns, but as AI agents arrived, that paradigm quickly needed to be rewritten. As an emerging technology, we needed a framework that could define how agentic systems should look, feel, and behave inside enterprise products.

The solution

The Carbon for AI Agentic Kit: a set of patterns, principles, and guidelines for designing agent-driven experiences such as reasoning and memory. This work set the foundation for watsonx Orchestrate, IBM’s one-stop-shop for AI agents.

Approach

Because agentic AI was just barely beginning to hit the market, it was my job to be ahead of the curve so that our product teams could implement agentic experiences as quickly as possible.

Because of this, I stepped in to build a pipeline from IBM Research directly to Carbon. This mutually beneficial relationship reduced duplicative work and, more importantly, gave us an advantage in the market.

My role

As UX Lead on this project, I:

Designed agent interaction flows and patterns such as multi-step reasoning, memory, and building agents.
Created a pipeline from IBM Research to get ahead of emerging patterns and into product as quickly as possible.
Co-created with 5 key product teams tasked with implementing early agentic features to ground the work in validated use cases.
Coached designers through their implementation of agentic experiences in their products.

Key features:

Reasoning trace
Showing the step-by-step of how an agent reaches conclusions and makes progress
Agentic plans
Enabling human-in-the-loop oversight, especially in high stakes scenarios
Memory
Saving and managing key pieces of information to improve the agent’s context and personalization
Agent Builder Flow
Detailing how an agent gets created and deployed across products

Reasoning trace
Showing the step-by-step of how an agent reaches conclusions and makes progress
Agentic plans
Enabling human-in-the-loop oversight, especially in high stakes scenarios
Memory
Saving and managing key pieces of information to improve the agent’s context and personalization
Agent Builder Flow
Detailing how an agent gets created and deployed across products

Outcomes

This work, combined with our chat interface, became the core experience for the newly debuted watsonx orchestrate, IBM's one-stop-shop for enterprise-ready agents. It has became the foundation for watsonx.ai's AgentLab.

Released at IBM's THINK 2025 conference, these agents are creating business value for clients such as the US Open, who used these two agentic products to build and power its MatchChat feature.