The Messy Middle: Notes on Building AI Products

Most of what we see in tech is the polished final product: the big launch on LinkedIn, the clean UI in the demo, the press release announcing "overnight" success. This blog is about everything that comes before that.

It's about the messy, unglamorous, and often frustrating process of actually making things work. It's a space dedicated to the long, winding road of building, not just the destination. This focus on the "how" and "why" is a direct reflection of my own career path a non-linear journey from sociology to founding my own company, and eventually to building AI platforms.

What We Don't See on Twitter

The headlines celebrate the launch of a new AI feature. What they don't show are the three weeks the team spent debating the right tech stack, the messy, unstructured data we had to clean just to get started, or the first three prototypes that completely failed.

That part, the part that happens out of sight is the most interesting and most important. I’ve learned the most valuable lessons in those moments:

• ...debugging a Shopify theme at 2 AM as a founder, realizing that I was the entire system and there was no one else to call.
• ...discovering that a complex recommender system wasn't struggling because of the algorithm, but because the entire data foundation beneath it was flawed.
• ...finding small but meaningful wins by building a simple LLM-powered tool that saved a team hours of manual work each week.

This is the "messy middle," and it’s where real product development happens.

A Workshop, Not a Lecture Hall

So, what will you find here? This isn’t a "how-to" blog with rigid rules or frameworks that promise to solve all your problems.

Think of this space less like a lecture hall and more like a workshop or a garage. It's my digital notebook, a collection of reflections, practical lessons learned from specific projects, and questions that often don't have easy answers. Some posts will be about successes, many will be about the lessons learned from things that didn't work.

Expect Sparks, Not Fireworks

The point isn't to give you polished answers. It's to share an honest look at the process of building AI-powered systems, the good, the bad, and the complicated.

If you're curious about that real, in-the-trenches work and you're okay with things being a little messy, you’ll probably feel at home here.

Welcome.