Token density
Build-along · ~10 min at your own keyboard
A rambling voice transcript and a tight bullet list can carry identical information at wildly different token cost — and bloated context doesn't just cost money, it makes answers worse. Needle-in-haystack: burying facts in noise degrades recall. The same thing happens to a whole session. Early on, the model is in its smart zone: sharp, following every instruction. As the session fills, quality slides into a dumb zone, slowly and with no warning: it forgets an instruction from earlier, repeats a fixed mistake. That slide starts well before the window is full. Plan around the smart zone: one task per session, and dense context so each task gets the sharp part.
The long version: Token density ↗
Do this
0/4Checkpoints
- ▸You should now see the original input and a compressed version side by side.
- ▸You should now see a smaller word or token count for the dense version.
- ▸You should now see a follow-up task run correctly from the dense version alone.
The three ways this goes wrong
- The compressed version dropped a fact. — Ask: 'list every name, number, and date in the original and confirm each is in the summary.'
- The dense version is not much shorter. — Say 'bullets only, no filler, max 100 words' and name the limit.
- You cannot measure the size. — Paste each version into a free token counter, or compare word counts.
Where these ideas come from: Smart zone / dumb zone, Token (AI Hero dictionary). Paraphrased here.
Share what happened
Paste the before/after sizes and the dense version.
Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].
🔒 Don't paste confidential, personal-health, or payment info.