Skip to content

What Rocket Money Had to Rethink to Build an AI Agent

Aaron Dignan shares how building Rowan made Rocket Money rethink data definitions, interfaces, engagement, unit economics, brand voice and risk.
James Gross2 min read

Building an agent made familiar questions difficult again at Rocket Money. At BRXND NYC 2026, Aaron Dignan shares lessons from Rowan, a personal finance agent, starting with the deceptively simple question: what is a bill? The talk follows the changes that reached across data, design, product, monetization, brand and legal.

Key takeaways

  • Make definitions explicit for agents; labels built around an app can depend on context only employees understand.
  • Measure customer outcomes and assess the cost of individual agent capabilities.
  • Choose the simplest interaction for the task before designing another screen.
  • Bring brand, legal and risk teams into ongoing product design and review.

In this session

00:00 Introducing Rowan and six familiar questions

Dignan introduces Rowan and explains how an early data misunderstanding revealed questions the organization had to answer again.

02:25 What is a bill? Rebuilding data for agents

He describes rewriting metadata and definitions so an agent could distinguish bills, subscriptions and other transactions.

05:31 What should this look like? Rethinking the interface

The team experiments with a messaging interface and asks which interaction best serves each customer task.

08:54 What is engagement? Measuring customer outcomes

Dignan discusses moving beyond app opens and time spent toward outcomes such as money saved, time saved and peace of mind.

11:49 What are we selling? The economics of agent actions

He explains why the frequency and compute cost of each agent action matter when designing a consumer business model.

15:33 Who are we? Designing an adaptable brand voice

Rowan's voice becomes a starting point that can adapt to users, rather than one fixed personality for every conversation.

Dignan describes legal and risk teams working with traces, evaluations and guardrails as continuing design partners.

20:42 What each function had to change

The closing recap shows how building an agent changed the work of each function and invites teams to identify their own underlying questions.

Recorded at BRXND NYC on November 5, 2026, at The TimesCenter in New York.

Written by

James Gross

James Gross

Co-Founder

More from James Gross

Alephic Writing

Ideas for building the AI-native enterprise.

New essays on AI, marketing, software, and the organizations being rebuilt around them.

Continue reading

More from Alephic

All Writing

Previous · November 12, 2026

What It Takes to Put AI to Work

Noah Brier and James Gross discuss AI adoption, marketing's role in change, model costs, and the company knowledge agents need to do useful work.

What Rocket Money Had to Rethink to Build an AI Agent | Alephic