Dispatch

“"It seems good" is a feeling, not a measurement.”

What an Eval Actually Is

Everyone asks which AI model is best. The teams that ship reliable AI ask a quieter question: how do we know if it's working?......

“A model's confidence and its knowledge are two separate things. Nothing in the output tells you where one ends and the other begins.”

Why AI Makes Things Up

People call it hallucination, like the model glitched. It didn't. It did exactly what it was built to do.......

“The dangerous errors hide in the places you stop checking.”

Don't Trust the Answer

I evaluate AI output for a living. Trusting it has almost nothing to do with how good the answer looks — here's what I actually watch for.......

“I used to write clearer instructions. Now I write clearer permissions.”

Rooms I Didn't Ask For

A month ago a model built most of my app without asking me anything, and it taught me something about judgment instead.......

“Fluent is not the same as right.”

The Quiet Variable

Last week I almost shipped something I knew was wrong. On fluency versus rightness, why the stupid questions hide the good work, and......

“Most interesting work doesn't arrive as a clean spec.”

Turning ambiguity into programs

How I take a fast-moving, loosely-defined goal and turn it into a structured program that actually ships — the mental model I use every day......