Tony, co-creator of the widely cited 2015 interactive explainer "A Visual Introduction to Machine Learning," now works as a product designer at Augment Code, a coding-agent startup, where his focus is UX for coding agents. That career line — from scroll-driven ML pedagogy to the front edge of agent tooling — adds some context to why the decade-old R2D3 piece keeps resurfacing on Hacker News.

The explainer, built with D3.js by Tony and Stephanie under the project name R2D3, uses a dataset of San Francisco and New York home listings to walk readers through decision trees, classification, overfitting, and the bias/variance tradeoff. The approach was deliberate: readers draw intuitive boundaries before encountering formal vocabulary. It worked. The piece introduced a generation of practitioners to concepts — split points, leaf nodes, training versus test data — that now appear routinely in discussions of the systems those same practitioners are building.

R2D3 describes itself as "an experiment in expressing statistical thinking with interactive design," not a company. That framing explains the sparse output and the outsized influence. The scrollytelling format became a template that ML educators and technical writers have copied repeatedly; few have matched the original.

The biographical details circulating in the latest HN thread — that Tony holds an MFA from the School of Visual Arts and that Stephanie has held data roles at Stitch Fix and Netflix before moving to Meta — have not been independently verified by Agent Wars, and readers should treat them as unconfirmed. What is visible from public records is Tony's current role at Augment Code.

His background makes him an unusual hire for an agent-tooling startup. Most UX work in the coding-agent space has been shaped by engineers with limited training in data visualization or statistical communication. The question of how to surface agent reasoning, uncertainty, and failure modes to users remains genuinely unsolved. A designer who spent years thinking about how to make decision boundaries legible to non-specialists is working on a relevant problem.

Augment Code declined to comment for this article.