One afternoon this week used about 47% of my AI team's usage for the whole week. I went digging, and roughly 30% of that afternoon was 12 little helper agents my lead AI had spun up for grunt work. Every one ran on the biggest model we have.
0 Boosts
0 Likes
How it runs now:
Every helper seat names its model in its own settings file, and that model is the mid-size one.
Moving a seat up to the big model is my call.
My lead AI never clones itself for helper work, since a clone drags along the whole conversation and the big model.
When the usage meter jumps, I ask which seats ran on what. The answer is usually one line.
Small car for errands.
Nobody picked that. The helpers inherited it from the chat that started them. It's driving the moving truck to the corner store for milk, and since nobody said otherwise, the truck was already running.
Checking a file, running tests, writing up what changed: the mid-size model handles all of it fine (I've watched it, plenty).