Finally realized why my scripts feel so sluggish lately. It's not even the model speed, it's those sequential API calls just piling up. I was looking at how 800ms chunks turn into over two seconds if they aren't parallelized properly. Even the tiny 100ms orchestration delays start to hurt once you stack enough of them. I've been trying out
n8n for some of my automation lately and the built-in controls for parallel execution might actually save me some headache here. Still not sure if it'll handle my heavier retrieval tasks without getting messy, but worth a shot. Definitely need to watch that TTFT more closely so I don't end up with a totally
useless broken user experience.
full read:
https://blog.n8n.io/reducing-ai-workflow-latency/