Token anxiety: rationing tokens and overspending both burn you out
EVs have range anxiety. The AI community has its own version: token anxiety, the fear that an LLM exhausts its context or its credits before arriving at a solution.
There are two failure modes, and they don't look alike.
Empty tank. Daily Pro cap is closing in. You ration prompts, attach less context, step down the model, compress conversations early, split sessions, hop providers, watch the meter between every prompt, settle for the first draft. The same anxiety that pushed the downgrade pushes the corner-cutting that follows.
Full tank. You'd think more tokens fix it. They don't. With unlimited capacity, the marginal cost of any prompt is zero, so you offload the trivial (renaming a variable, looking up a flag, reformatting a paragraph), let chats grow long with stale code, never close anything out. You babysit agents from the checkout line, from bed, from the grocery store. The model gets to forget. You don't.
The cure isn't a bigger battery. The post argues it's knowing the route: decide what the work is worth before you ask, spend where the answer earns it, hand the small tasks back to yourself.
My personal practice is to downgrade my plan every few months for a month at a time. The cap forces intentional use, and the spare hours go elsewhere.
https://starikov.co/token-anxiety/
How do you regulate? Anyone here deliberately keeping themselves in the middle of the tank?