Posts taggedai

Cheaper implementation makes everything else more expensive. Eight assumptions about working with coding agents that hold locally and break at scope.

A benchmark from Surge AI drops an agent into a company with a 20 to 124 page policy document and grades whether the document actually constrains it. The best model passes 36.2 percent of tasks.

I was using AI before I had tried to understand the task. Taking regular days away from it exposed gaps in my knowledge and improved the way I use it.

A simple story about mice and men takes on a heavier weight when the "cheese" is the coding work you spent years mastering. No advice here—just reflections from the edge of the AI disruption, left open for you to explore.

How I use focused specifications, reviewable changes, and independent validation when working with coding agents.