Anyone has any good recommendation or sources on how to be more token efficient when using Codex / Claude Code? Presumably from prompting or systems design perspective?
As a *very* inexperienced software engineer, i find the popular opinion that SWE is doomed to be extremely lazy. If you’re building LEGO buildings then sure Claude code or Codex is gonna build a fancy one for you. Yay.
But if you’re gonna build an actual building with agents w/o any swe knowledge then I’m not staying in your building brother.
Every day I get partially baited to prototype something along the mental wellness angle. But I’m also adamant on launching something like that because:
1) cost of getting something wrong (even if it’s not a prediction/classification tool) is catastrophic
2) i am in 2 minds whether people would even use a scripted tool rather than talk to a real person/hotline
This is so good. The most interesting part of this is you can distinctively hear signature sounds/singing techniques from several of the biggest Mandopop stars. Getting one shotted by AI music was not on my 2026 bingo card.
https://youtu.be/zOMblRrWmBM?si=XbtWJLTy5OM9I5og
a good friend asked if he could come over because he quarrelled with his wife. I said sure of course. Turned out to be smallest of small things.
Many such cases?
