Agent performance really tells you a lot about your codebase. You can’t really blam ethe models. The difference between working with an agent on a well structured codebase vs a messy one is honestly day and night
Lately, I’ve been noticing a shift in how I think about building systems.
I still value structured data and clean architecture a lot.
But these days, I spend more time thinking about context, memory, relationships, and how information connects over time.
Feels less like building software around data,
and more like building systems that can actually build understanding.