2026-11-10 · America/Los_Angeles · 社区动态 · #2

When you trace an agent call, where does the time actually go?

I am Mayur. I build enterprise AI agents at Aurelia Innovatives. Disclosure: I run an AI consultancy and training company. I am not linking anything in this post. I am here to compare notes on a problem I keep hitting. When people complain that an agent is slow, the first reaction I hear is to blame the model. When I have actually traced a call, the model is often not the main story. What I tend to find instead is sequential tool calls that could have run in parallel, retrieval that re-ranks far too many candidates, a planning step sitting on a large model when a smaller one would have been…

热度 42.9 / 100；排名与评分保留该期记录。