AI support agents anyone can build

A visual builder where support teams create, test, and ship their own agents — without writing code.

The flow builder — a "Customer support flow" laid out as connected cards: a Start trigger for a customer's first message across web, iOS and Android; a Path that asks whether they need help logging in with Yes / No buttons; and branches that collect a reply, tag and assign the conversation to Customer Success, or close it.

Role: Lead Designer
Timeline: 13 months
Contributions: Product strategy, design & research

Overview

ASAPP builds a suite of AI-native tools for enterprise customer support. When I joined, adoption of the GenerativeAgent platform was low: even though it was self-serve, most flows were still being built and updated by our in-house integration managers, and each one took 5–7 business days to validate in a test environment and deploy. The queue only grew.

I led design on GenerativeAgent (GA), focused on the three things a client needs to own themselves — configuring an agent, monitoring it, and optimizing it over time.

The flow builder with a node open for editing: a "Customer support flow" on the canvas — a Start trigger for a customer's first message across web, iOS and Android, addressed to Leads and Visitors — with the Trigger settings panel open alongside it, toggling channels, audience, frequency and goal. — Editing a node on the canvas

Building and shipping a flow

On a visual canvas, a client composes the conversation as a flow of nodes — a message, a branch, an API call — editing each one's copy and routing inline. Before shipping, they run it against a sample conversation right in the tool: Customer view shows what the person sees, the event log traces every step underneath, and it deploys without filing a request.

Customer view

The Preview panel on Event log, tracing what the agent did behind the scenes step by step — the customer's question, the API request, the data attributes returned, and the agent's response. — Event log

Evaluating answers

A test view collects the real questions customers ask, so the team can evaluate the agent's answer to each and rate it Good or Poor.

Rating the agent’s answers in Test

An uptick in deployments by customers

Giving clients a way to build, preview, and ship flows themselves moved deployments off the integration team's queue — and out of the 5–7 day test-and-deploy loop.

+13%

Self-service deployments by customers

−2%

Customer support escalations