Hi, I'm Purnima.
I design AI for enterprise tools — the unglamorous, mission-critical software that keeps businesses moving. Most recently, I led the design of the Integration Agent inside IBM webMethods.


Designing the Integration Agent for IBM webMethods

How I led the design of an AI agent that builds enterprise integrations from natural language — and the interaction principles it left behind across IBM.


Role
Lead Designer
Integration Agent, webMethods
Year
2024 — 2025
Platform
IBM webMethods Integration
(part of IBM Hybrid iPaaS)
Teams Involved
Lead Designer (Me)
+ 3 designers
Pembroke Research
IBM Carbon Design
Cross-IBM AI Working Group
What I contributed
Integration Agent · Jul 2024 — Dec 2025 · 18 months
My role
Design lead and sole design owner. I am named in IBM's delivery plan as the owner of user journeys, flows and designs for both the Public Preview and the GA release. I ran the design of the agent end to end, mentored three junior designers on adjacent AI surfaces, and represented the portfolio in the Cross-IBM AI Working Group.
The motivation
Integration specialists were spending roughly 60% of their week rebuilding the same handful of workflow patterns, while business users waited weeks in a queue for integrations nobody had time to build. The design question was not "where do we put AI" — it was "which work should stop being human work, and who decides?" Everything I designed traces back to that.
What I designed
  • The end-to-end conversational flow — nine stages from prompt to deployed, running workflow
  • Every fallback and error branch: unresolved triggers, ambiguous actions, cancelled dialogs, abort paths
  • The AI entry-point pattern — the onboarding banner, sample prompts and prompt bar now reused across the Hybrid iPaaS point products
  • The hand-off model between chat and canvas, including when the agent must stop talking and open a real configuration dialog
  • The mapping review and test-execution moments, where the agent proposes and the person approves
  • Five interaction principles for agentic AI, since adopted by four other IBM product teams
  • Storyboards, service models and journey flows used to align Product, Engineering and Offering Management
What I didn't do
I did not write production code, run the model evaluation, or own the go-to-market. Research fieldwork was run with the Pembroke UX Research team, not by me alone. Visual components came from Delite and IBM Carbon rather than being drawn from scratch. Three designers worked alongside me on the sibling agents; their work is credited at the foot of this page.
Confidentiality
This case study covers capability that shipped publicly and has been discussed in IBM's own marketing, the Forrester Wave, and press coverage. Unreleased roadmap, internal metrics and patent detail are deliberately left out. Diagrams here are redrawn from my working artifacts rather than lifted from internal decks.
The rest of this case study
🔒

The rest is on request.

Everything past this point — the research, the process artifacts, the nine-stage flow, the screens and the measured impact — is covered by an NDA. I share it on request rather than publishing it.

Request access

I usually reply within a day.