AI Trends

I build agentic systems inside organizations that already have systems, which is a different job from building one on a clean page. The interesting decisions are almost never about the model. They are about what an agent is allowed to touch, what happens on the day it gets something wrong, and whether the team can still run it after I leave. So I start with the process that actually hurts and the number that would prove it got better, then build the smallest thing that moves it. I integrate what is already there rather than replacing it, keep each step reversible, and hand over a system the team can extend without me. What I write here comes out of that work, the parts that held and the parts I got wrong first.