Comment by awfm9

I used to do this, and do it well. Nowadays, I avoid it like the plague. Not just because of the advent of AI agents, but also. I usually try to condense the core business logic of the application into a tight sequencer, and then every type of slower workload has a manager for it, with queue, dispatching. All logic remains linear, easy to review and follow. Concurrency is basically just handled at the level of kicking off some work, and then funneling the result back into the sequencer. Easier to test, highly scalable concurrency.