Agreed on the cognitive budget point. For the "simple obvious things" case though, the bigger cost lever is just routing to a smaller model rather than compressing output from a big one.
Most API traffic in prod is routine stuff where a smaller model gives identical results at a fraction of the cost. Caveman-style compression saves maybe 30-50% on output tokens per request. Model selection saves 90%+ on the requests that don't need frontier reasoning.
The tricky part is reliably knowing which queries actually need the big model. You can get surprisingly far with embedding-based classifiers on the input, but there's always an error rate, and the cost of a false negative (sending a hard query to a weak model) is worse than the cost of a false positive.
As those tokens flow through the QKV transforms, on 96 consecutive layers, they become the canvas where all the activations happen. Even in cases where it's possible to communicate some detail in the absolute minimum number of tokens, I think excess brevity can still limit the intelligence of the agent, because it starves their cognitive budget for solving the problem.
I always talk to my agents in highly precise language, but I let A LOT of my personality come through at the same time. I talk them like a really good teammate, who has a deep intuition for the problem and knows me personally well enough to talk with me in rich abstractions and metaphors, while still having an absolutely rock-solid command of the technical details.
But I do think this kind of caveman talk might be very handy in a lot of situations where the agent is doing simple obvious things and you just want to save tokens. Very cool!