I find myself in a pattern during a software project where I go up to do huge swathes of work with agent(s) that touch large areas of the code. Then I identify a bottleneck or performance I don’t like, I go down and dig deep in to systems engineering to benchmark performance and look at syscalls to understand what is going on. This is often very granular and low level.
I don’t know if this is quite the right analogy, but it feels like a sine wave when I fly high orchestrating many concurrent agents across a large portion of the codebase, then dive down in to the depths of smaller portions of the code to do systems level work. Maybe software engineering has always been like this, but it has become more noticable as I iterate on things in faster cycles with agents.