Late 2025, I spent a weekend going deep on GenAI internals — tokenizers, transformers, attention mechanisms, the KV cache. Not because I needed to implement any of it, but because I wanted to understand what I was working with. Product people who treat AI as a black box make worse decisions than those who understand at least the shape of the system.

That weekend led me to sign up for the AI Product Academy cohort with Marily Nika, Ph.D — six weeks of use cases, discussion, and building. Structured learning after self-directed exploration turned out to be the right sequence. The weekend gave me enough vocabulary to ask better questions; the cohort gave me frameworks to organize the answers.

What I came away with: leveraging AI well requires structured thinking and a working understanding of how LLMs transform data. Knowing where to use guardrails, when to let AI do the work, and when the human needs to stay in the loop — these aren’t instincts you develop by reading about AI. They come from building with it, breaking it, and iterating.

I completed the Maven AI Product Management certification in January 2026. The credential matters less than the calibration it gave me.