How I build AI automations, and what I learn doing it. Real projects, with the details, costs, and lessons included.
My board game app's search blends exact-word matching with search by meaning, and every so often it returns a result no person would pick. A decision model now double-checks each result before anyone sees it. I built a test harness first, to prove it was worth adding.
Read the post →
A chatbot gives you one answer from one point of view. I wanted to see each decision the way a strategist, an operator, a marketer, and a lawyer would, and then make the call myself.
Read the post →
I don't open a dashboard to check my App Store sales anymore. They sit on a small e-ink screen on my desk and update twice a day. Here's how it works, and why it isn't really about the App Store.
Read the post →
Picking the AI model behind an automation feels like a one-time decision. It isn't. In one week, re-testing two of my own apps against new options made one better and faster and another cheaper, before anything broke.
Read the post →
The first question most owners ask about AI is whether it'll send something in their name that they wouldn't. The answer isn't yes or no. It depends on how sure the system is, and you decide where that line sits.
Read the post →
I handed one of my least favorite chores to a decision model. It labels every new GitHub issue for less than a penny per hundred, and it flags the ones it isn't sure about instead of guessing.
Read the post →
My task manager has no priority field, on purpose. So I tried calculating priority fresh instead: an AI model answers a few questions about every open task, and code turns the answers into a ranking. All 238 tasks take about a second and cost under a penny.
Read the post →
I had 16 paper-only state tax forms to fill out for my LLC, with the same details on every one. AI found them, downloaded them, and filled them in. My part was about 10 minutes of supervision.
Read the post →
Twice a day, an AI agent goes through everything I've captured: it fixes vague titles, applies tags, files tasks where they belong, and answers questions I've left in the notes. By the time I open a task, it's ready to act on.
Read the post →
Most people use AI in a chat window, copying things in and pasting answers out. MCP connects it to the tools where the work actually lives. Here's the first thing I did with it, and where the same idea fits in a business.
Read the post →