A ruby gemstone, playing on Ruby on Rails as the foundation for a career that now spans data and writing

That work taught me something that carries straight into AI: a clean foundation is invisible until it breaks.

Feed a model incomplete or untraceable data and you get content that sounds confident and carries no signal. The model won’t warn you. People call the result a hallucination, when half the time it’s bad inputs doing exactly what bad inputs do.

That failure mode is easy to miss because the output still looks polished.

For mission-driven teams there’s a second risk past accuracy. If your data overrepresents the audiences that are easy to measure and thins out the ones you actually serve, the AI spreads that skew. It gets efficient at telling a story that leaves people out.

Before any of this reaches a marketing funnel, a few things are worth tracking. Where each dataset came from and whether you can trace it. Whether your analytics events fire the way you assume they do. Where your CMS taxonomy has coverage gaps. Whether your inputs reflect the people you serve or just the ones cheapest to collect.

This is where the two halves of my background meet. The data work and the writing work sit with the same person for a reason. A clean data layer gives the writing accurate material to stand on. And someone has to read both the spreadsheet and the story and catch the moment one starts lying to the other.

AI inherits whatever the systems feeding it already earned. Thin foundation, and the smartest model on top of it only sounds smart.