About this site

Plain Text is where I write about what happened after words stopped being the thing I used around software and became the way I started building it.

I came into software sideways. I had an English degree, worked as a bank teller and call center trainer, then got a chance to become a Salesforce admin. I did not know how to write much code, but I could take a messy process, pull it apart, figure out where each piece belonged, and put it back together in a way someone else could follow. Salesforce Flow felt a lot like editing a chapter to me. The logic was visible, I could move it around, test it, find the part that did not belong, and keep going. I did not know it at the time, but that way of thinking would become the bridge between the writer I had been and the software builder I wanted to become. The first post on this site tells more of that story.

LLMs made the bridge real. The same skills I used to shape training, break down workflows, and work through a story could be used to guide a model through real software work. That started with Storyweaver, a little command-line writing app that could carry the memory of a story from one chapter to the next. It was the first app I built and also the first one I let an AI agent tear apart because I was too excited by a confident recommendation to ask what it planned to protect. Learning how to recover from that sent me into Git, branches, pull requests, and the parts of development I used to think belonged to other people.

The work has grown since then. I am building AgentShelf with Gabe and Edgar, trying to make useful AI agents available to people who should not have to become AI engineers before they can put one to work. I have written about the brain, the job, and the app because the word agent should not require a wall of jargon. I have compared two review systems running against the same code and found that the harness around a model can change the cost, the trust, and what happens when it misses something. More recently, I stopped juggling every AI coding session myself and watched Claude dispatch a crew of Codex agents into a workspace I had been quietly preparing for months. Somewhere along the way this stopped being a story about writing better prompts. It became a story about building better places for agents to work.

So this is not a tutorial site in the usual sense, although I do explain the things I am learning. It is closer to a field journal from inside the work. You will find origin stories, rough first versions, expensive mistakes, experiments from my own systems, and numbers I pulled because I wanted to know whether something really worked or only felt like it did. I am not interested in hiding the rough edges or pretending one model wins every time. The model matters, the prompt matters, the context and the app around it matter too, and the human still has to decide what is worth building and whether the result is actually any good.

If you came into technology through writing, operations, training, Salesforce, no-code tools, or any path that made traditional development feel like somebody else's territory, you are probably in the right place. I still feel mid-crossing. I just have more of the map now, and I am writing it down as I go.

New posts when something I have been building, breaking, or measuring turns into something worth saying.