the other thing is averaging over days says nothing about spikes in the rate - I imagine very few systems see more-or-less constant traffic over the course of an entire day

This. Customers are not organized in neat lines waiting their turns around the clock IRL

Pretty common in industrial control. Your sensors take measurements at a constant rate all day everyday. Sometime a pretty high rate e.g. IEC 61850 is at 4 kHz.

I worked on a SaaS CMS. A major consumer PC manufacturer was a customer and their "Welcome to your PC" app would pull their news feed from our system. People would turn on their PC and they'd pull a little JSON. 100s of Thousands of request per day, no big deal. The manufacturer decided that they wanted the updates to show up for everybody at the same time on the hour. Serving 100k req/s for a minute is a much different problem to solve.

you could use mechanical turk for 10k tasks per day

Anything other than operations per second raises my suspicions--a minute is a huge duration for most computing, the speaker is just angling to multiply the measure by 60 for whatever reason.

I snorted at the Jenkins suggestion… and you’re right… spooling an agent all that nonsens… lol I imagine telling “the business” that your customer requests are handled timely by a jenkins job queue

"It's not a lot" has basically been my career up until now, at least the bits where I interacted with a back-end; in the few cases where I witnessed a new back-end being set up, they over-engineered it without fully understanding the actual problem or needs; the engineering was often wishful thinking as well, as in, they wished they had the problems that their proposed solution would fix. Cargo culting, in other words.