I had already spent decades building websites, databases, airport systems and digital platforms when generative AI entered my work.
I started using GPT-3.5 shortly after ChatGPT became publicly available in December 2022. At first it was exploration — understanding this new tool, its strengths, its limitations, and what it could add to a way of working I had already developed.
Very quickly, it began helping me with something that actually mattered: Acoustonauts, my charitable foundation.
GPT-3.5 began helping me think, structure, research and build while I was developing Acoustonauts.
I studied how models responded to language, structure, context and iteration. One project pushed that learning particularly far: creating a tarot deck using GPT and MidJourney.
I explored different tools, models and approaches. I programmed with them, built MVPs, experimented, broke things and rebuilt them.
AI moved from something I experimented with into something I could architect into websites, applications, platforms, automations and products.
My work expanded into orchestration, routing between models, agentic workflows, token efficiency, evaluation and human review.
Today I can go beyond simply using the tools and help others understand where AI is useful, how to integrate it, and where human judgment remains essential.
Today, generating an image can feel almost effortless. In 2023, getting a single card exactly where I wanted it could take hours of prompting, iteration and refinement.
That process taught me that AI was not magic. It was a medium — and quality still depended on patience, language, structure, experimentation and taste.
Tap a card to bring it forward.
“What can AI generate?”
“What can AI actually do?”
Making AI understandable and useful.
Understanding and improving digital visibility.
AI-assisted marketing analysis and recommendations.
The question became more interesting: what combination of models, tools, automation, context and human review produces the best result?
Sometimes the answer is AI.
Sometimes it is a database query, a better workflow, a simpler interface — or removing a process entirely.
The objective is not to use more AI. The objective is to build something that works.