Splitting cost into build, run and keep alive is the most useful framing for anyone comparing quotes. Most proposals name the first, guess at the second and leave out the third, which is why the real bill arrives a few months later.
The example of a ten turn task costing around 24 times a single reply is worth putting in front of buyers early. It is the kind of detail that changes a design decision, such as summarising context or ending sessions sooner, before anything is built.
On client work we price the build as one fixed scope agreed up front, with the expected run cost per task written next to it and measured on real examples during acceptance. Keep alive work, such as model upgrades and prompt drift, gets its own line instead of hiding inside support.
When you compare quotes with clients, how often do vendors name all three parts?