People and agents can be the same population making the same queries — yet present a completely different load profile. Humans naturally back off while they wait: a slow answer slows their next question. Agentic systems can fan out and keep scheduling work, so long analytical latency turns into concurrency, retries and resource pressure on the execution plane.

This is an illustrative workload-pressure and queueing model, not a vendor or StarRocks benchmark. The Fast and Slow query times are scenario assumptions you choose; the model is deliberately generous to the database. Full methodology, assumptions and sources are in the simulator’s own Methodology & assumptions panel.