Burn less, ship more: the case for token optimization · HackerLangs