Blog
Notes on building software where AI does a specific job.
Before an AI builds your app, ask what it plans to build
An AI coding tool can build the same app several ways. See the options for the screens, the business rules, and the records, and choose them yourself.
Ask what it intends to build before you let it code
A coding agent can build the same request several ways. Ask for the options in the front end, the business logic, and the schema, and choose before it writes code.
When Someone Says 27B, They Mean Parameters
The number on a model name is a count of parameters, not a score, a price, or how many words it knows.
AI as the Coordinator of Human Work
The useful change is not cheaper judgment. It is AI packaging the work so a person can make one call without holding the whole system.
The Engagement Loop: Recurring, Personal, and Not Creepy
Run the loop on events, not a calendar, and never let an outreach be specific enough to trace back to a single post.
Finding Wealth Management Prospects Without Making Anything Up
Qualify prospects from public filings and permitted data, with every number traceable to a source.
Stop Asking the Model for the Answer
Ask the model for the method and the code. The answer should come from running that code, not from the model's opinion.
Your AI Isn't Rambling Because It's Broken. It's Rambling Because You Let It.
Models ramble when the prompt leaves room for it. These phrases set the shape, the tone, and the length.
Token Maxing: Is Burning Your Budget Actually Making You Productive?
Spending the whole token budget is not a measure of productivity. It pays when the work is verifiable, parallel, and worth doing sooner.