Performance is sometimes a design/architecture decision, in which case it needs to be thought about up front. That kind of performance isn't an "optimization" that can be deferred until later.
I wonder if there's a correlation between people who get really into agentmaxxing and people with addictive personalities.
Anecdotally, I don't have an addictive personality and I've never felt a need to maximize agent output.
It's really important to be clear on how things are *defined*.
When I read a variable or function name, I want to know what that name is defined as.
I should be able to create a dictionary with every name in the codebase and a clear definition.