Journal

Shipped, useful, prototype, experiment. A quick defense of the verdicts I put on every build — and why the throwaways earn their keep.

Most builds in this challenge live for a day. This one turned into a real product with payments, multi-tenancy, and 263 commits. Here's what changed.

Two years, 73 builds, honest verdicts. The things I expected to learn, and the things I didn't.

A dare to myself: ship one useful app a day for a month, and treat coding agents as collaborators instead of autocomplete.

Sunday Hype has maybe ten users. My kids are some of them. Building something useful for a small, specific audience turns out to be one of the more satisfying things I do.

I built two RAG systems this year. One does document search. One extracts structured knowledge from documents. They're completely different products, and I conflated them for too long.

My nephew Brian died in an avalanche on February 22nd. I built the memorial site in a day, for the funeral on March 6th. Some things I'll remember about building it.

I'd been building tools that do one thing. MCP made me realize I should be building tools that know things — and can tell an agent about it.

NarrativeFlow has three agents coordinating brand narrative, email sequences, and campaign analysis. The system works. The failure modes are just harder to debug.

After a dozen builds, the clearest thing I've learned is where the line is between what you can delegate and what you can't.

Two Laravel apps, a mock API, scoring logic, and a message template for the COO to paste into an email. Built in one day. That's the part that still feels weird.

I built a chatbot that talks like my dad. It taught me more about persona prompting than anything I've read about it.