Why I built it
AI made writing fast. It also made everyone sound the same.
I write a lot. This blog is fifteen years of thinking about architecture, teams, and the decisions that don't have a clean answer. Writing is how I figure out what I actually believe — so when AI writing tools arrived, I leaned in hard.
They were fast. Uncomfortably fast. But something was off. Everything I generated started to sound like everything everyone else generated — the same cadence, the same “it's not just X, it's Y”tic, the same glossy, oddly American polish. The words were fine. They just weren't mine.
The problem isn't that the model can't write. It's that it doesn't know you. It forgets you the moment the tab closes, improvises your facts, and starts every draft from zero. For a founder, an operator, a creator — anyone whose voice is the entire point — a tool that quietly flattens that voice is worse than no tool at all.
So I built Zsper. Not a chat window with a longer memory — a pipeline. It keeps a persistent, inspectable memory of how you think, learned as a by-product of the writing you already do. Before it drafts a single line, it selects from that memory: your stances, your stories, the way you actually make an argument. Then it writes, and shows you the provenance of every choice.
It's the same philosophy as everything else on this site. AI writes the code; you still make the call. Zsper drafts. You stay the author. Nothing it learns and nothing it writes goes out in your name until you say so.