"How can I save on tokens?"
Fair question. Tokens are getting more expensive, you're burning through your weekly quota by Tuesday, and more people around you are switching to Chinese models running at up to 10% of the cost for comparable results.
Token economy is a real topic if you're getting value out of AI.
Here's my answer: don't change models. Get smarter about the context you hand your AI.
Months ago I was trying to work out why my 5-hour quota burned up in 30 minutes, every single time. That sent me deep diving, and I came across a term: progressive disclosure.
The idea is that if you structure your context files, AI will just add to its context what's most relevant. Not "everything."
Same setup, same files, a fraction of what the model has to read.
Token economy improved, but then so did performance. The model stopped tripping over things that had nothing to do with the job at hand.
Context engineering doesn't have to require a PhD. It's about you deciding what the AI will be allowed to know before it reads your instruction at all.
And it's unglamorous enough that most teams skip it, which is exactly why it's still an edge.
Next time you're wondering how to save tokens, ask your AI to help you build your context around progressive disclosure.
--
Further reading: for me the master of this is Jake Van Clief (Lost & Lucky). Link in the first comment.
