super helpful tip for using claude code:

don't "go down the rabbithole" of feedback when it isn't working well

learn what you can (e.g. what didn't work, what you think the right path might be), and then clear your session. optionally give it the context from the last try. the model is surprisingly stochastic.

from this interview. Recently I've done a 180 on Anthropic—their product quality is very strong, and it's becoming clear that being a foundation lab is a major competitive advantage on product.

Meta's moves make a lot of sense in that lens.

design @splits | prev @event and @latenightcrew

from this interview. Recently I've done a 180 on Anthropic—their product quality is very strong, and it's becoming clear that being a foundation lab is a major competitive advantage on product.

Meta's moves make a lot of sense in that lens.

https://www.lennysnewsletter.com/p/anthropic-co-founder-benjamin-mann

Can vouch for this. Let its mistakes guide you down a better path.

Totally agree. Claude is incredible if you use it carefully. If you start getting into too many aimless suggestions, start over. If your project is getting unwieldy for it to help code - parcel out functions. You can even just give it a little context and it doesn't need the whole picture to help out.

It's my favorite AI.