In this episode of DataLatam, we welcome back Josiah Parry, who joins us from Seattle, Washington. Josiah divides his time between building open-source tooling and working on the ArcGIS Bridge at Esri, connecting the ArcGIS ecosystem to R. His trajectory began in sociology, where an early introduction to R inspired his transition into urban informatics at Northeastern University, followed by roles at RStudio (now Posit), Circana (NPD Group), and the R Consortium.
We discuss the origin and technical foundation of Ricochet, Josiah's platform designed to help data scientists deploy applications across R, Python, and Julia without infrastructure friction. Josiah explains that while data scientists are inherently developers, deployment has historically been a significant barrier. To guarantee security in multi-tenant enterprise environments, Ricochet utilizes an in-house rootless container runtime built in Rust, completely isolating application processes, eliminating root privileges, and integrating redundant audit logging with OpenTelemetry.
The conversation turns to the evolving role of AI assistants in software engineering and the growing trend of "vibe coding". Josiah distinguishes between using AI solely for prompting—which produces useful prototypes that often mask fragile implementations—and disciplined AI-assisted development, where experienced developers maintain active oversight of source code. Without strict supervision, frontier models often optimize for machine-centric solutions that compromise long-term human readability and software extensibility.
We also explore the fundamental divergence between software engineers and data scientists. While software engineers prioritize building systems that pass test suites and scale reliably, data scientists are defined by continuous inquiry. Josiah and the team examine how data professionals constantly question initial outputs through methods like causal inference and variable importance metrics (such as SHAP), refusing to accept a metric until its underlying mechanisms are understood.
Looking ahead, Josiah reflects on the societal consequences of cognitive offloading to AI models. Drawing parallels to Frank Herbert's Dune, he warns against a growing divide between those who retain the discipline to learn core fundamentals and those who rely passively on automated answers. We contrast this view with cultural technology shifts, underscoring the enduring value of human expression and deep domain mastery.
Relevant Links
Resumen en español
En este episodio conversamos con Josiah Parry, miembro del equipo de analítica espacial y ciencia de datos en Esri y creador de Ricochet, una plataforma orientada a llevar soluciones de ciencia de datos a producción de forma segura. Josiah analiza por qué R pertenece a entornos productivos, el diseño de contenedores seguros con Rust, los límites del vibe coding frente al desarrollo asistido por IA, y el debate sobre el impacto del aprendizaje automatizado en el pensamiento crítico humano.